{"id":"04b222e7-54e7-4980-8bf9-e3b24729aab3","entityType":"agent","slug":"clawhub-deciqai-hindsight-bias","name":"Hindsight Bias","canonicalUrl":"https://www.xpersona.co/agent/clawhub-deciqai-hindsight-bias","canonicalPath":"/agent/clawhub-deciqai-hindsight-bias","generatedAt":"2026-10-12T11:15:16.504Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-12T01:29:00.615Z","emptyReason":null},"description":"Activate when: someone says 'I knew it all along' or 'we should have seen it coming'; a post-mortem is blaming someone for not predicting an outcome; a team... Skill: Hindsight Bias Owner: deciqai Summary: Activate when: someone says 'I knew it all along' or 'we should have seen it coming'; a post-mortem is blaming someone for not predicting an outcome; a team... Tags: latest:1.0.4 Version history: v1.0.4 | 2026-07-16T18:02:19.835Z | user Description tail link + agents machine-readable metadata line (deciqai.com/s/hindsight-bias.json) v1.0.3 | 2026-07-08T11:05:08.651Z | use","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 961 downloads reported by the source. Last updated 10/12/2026.","installCommand":"clawhub skill install s17a4mqcnk515kvaca5ze55d0x88pfpx:hindsight-bias","sourceUrl":"https://clawhub.ai/deciqai/hindsight-bias","homepage":"https://clawhub.ai/deciqai/skills/hindsight-bias","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/deciqai/hindsight-bias","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/deciqai/skills/hindsight-bias","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":60,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"Activate when: someone says 'I knew it all along' or 'we should have seen it coming'; a post-mortem is blaming someone for not predicting an outcome; a team... "},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-12T01:29:00.615Z","emptyReason":null},"protocols":[{"protocol":"OPENCLEW","label":"OpenClaw","status":"self-declared","notes":"Declared in the public agent profile."}],"capabilities":[],"verifiedCount":0,"selfDeclaredCount":1,"capabilityMatrix":{"rows":[{"key":"OPENCLEW","type":"protocol","support":"unknown","confidenceSource":"profile","notes":"Listed on profile"}],"flattenedTokens":"protocol:OPENCLEW|unknown|profile"}},"adoption":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-12T01:29:00.615Z","emptyReason":null},"stars":null,"forks":null,"downloads":961,"likes":null,"task":null,"library":null,"packageName":null,"latestVersion":"1.0.4","tractionLabel":"961 downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-12T01:29:00.601Z","emptyReason":null},"lastUpdatedAt":"2026-10-12T01:29:00.615Z","lastCrawledAt":"2026-10-12T01:29:00.601Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-13T01:29:00.601Z","lastVerifiedAt":null,"highlights":[{"version":"1.0.4","createdAt":"2026-07-16T18:02:19.835Z","changelog":"Description tail link + agents machine-readable metadata line (deciqai.com/s/hindsight-bias.json)","fileCount":7,"zipByteSize":18250},{"version":"1.0.3","createdAt":"2026-07-08T11:05:08.651Z","changelog":"Footer now uses /c/<slug> short link (fixes UTM truncation when SKILL.md is read in a terminal)","fileCount":7,"zipByteSize":18362},{"version":"1.0.2","createdAt":"2026-07-08T00:50:11.720Z","changelog":"Refreshed content + GitHub star link in footer","fileCount":6,"zipByteSize":13679},{"version":"1.0.1","createdAt":"2026-07-07T20:33:44.507Z","changelog":"Add catalog categories and topics","fileCount":5,"zipByteSize":10789},{"version":"1.0.0","createdAt":"2026-06-29T06:18:02.054Z","changelog":"Initial publish","fileCount":5,"zipByteSize":11020}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s17a4mqcnk515kvaca5ze55d0x88pfpx:hindsight-bias","setupComplexity":"low","setupSteps":["Setup complexity is LOW. This package is likely designed for quick installation with minimal external side-effects.","Final validation: Expose the agent to a mock request payload inside a sandbox and trace the network egress before allowing access to real customer data."],"contract":{"contractStatus":"missing","authModes":[],"requires":[],"forbidden":[],"supportsMcp":false,"supportsA2a":false,"supportsStreaming":false,"inputSchemaRef":null,"outputSchemaRef":null,"dataRegion":null,"contractUpdatedAt":null,"sourceUpdatedAt":null,"freshnessSeconds":null},"invocationGuide":{"preferredApi":{"snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-hindsight-bias/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-hindsight-bias/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-hindsight-bias/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-hindsight-bias/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-hindsight-bias/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-hindsight-bias/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-12T11:15:16.498Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-hindsight-bias/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-hindsight-bias/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-hindsight-bias/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-hindsight-bias/trust"}},"reliability":{"evidence":{"source":"runtime-metrics","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No trust, reliability, or runtime telemetry is available."},"trust":{"status":"unavailable","handshakeStatus":"UNKNOWN","verificationFreshnessHours":null,"reputationScore":null,"p95LatencyMs":null,"successRate30d":null,"fallbackRate":null,"attempts30d":null,"trustUpdatedAt":null,"trustConfidence":"unknown","sourceUpdatedAt":null,"freshnessSeconds":null},"decisionGuardrails":{"doNotUseIf":["Contract metadata is missing or unavailable for deterministic execution."],"safeUseWhen":[],"riskFlags":["missing_or_unavailable_contract","trust_data_unavailable","schema_references_missing"],"operationalConfidence":"low"},"executionMetrics":{"observedLatencyMsP50":null,"observedLatencyMsP95":null,"estimatedCostUsd":null,"uptime30d":null,"rateLimitRpm":null,"rateLimitBurst":null,"lastVerifiedAt":null,"verificationSource":null},"runtimeMetrics":{"successRate":null,"avgLatencyMs":null,"avgCostUsd":null,"hallucinationRate":null,"retryRate":null,"disputeRate":null,"p50Latency":null,"p95Latency":null,"lastUpdated":null}},"benchmarks":{"evidence":{"source":"no-benchmark-data","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No benchmark suites or observed failure patterns are available."},"suites":[],"failurePatterns":[]},"artifacts":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"high","updatedAt":"2026-10-12T01:29:00.615Z","emptyReason":null},"readme":"Skill: Hindsight Bias\n\nOwner: deciqai\n\nSummary: Activate when: someone says 'I knew it all along' or 'we should have seen it coming'; a post-mortem is blaming someone for not predicting an outcome; a team...\n\nTags: latest:1.0.4\n\nVersion history:\n\nv1.0.4 | 2026-07-16T18:02:19.835Z | user\n\nDescription tail link + agents machine-readable metadata line (deciqai.com/s/hindsight-bias.json)\n\nv1.0.3 | 2026-07-08T11:05:08.651Z | user\n\nFooter now uses /c/<slug> short link (fixes UTM truncation when SKILL.md is read in a terminal)\n\nv1.0.2 | 2026-07-08T00:50:11.720Z | user\n\nRefreshed content + GitHub star link in footer\n\nv1.0.1 | 2026-07-07T20:33:44.507Z | user\n\nAdd catalog categories and topics\n\nv1.0.0 | 2026-06-29T06:18:02.054Z | user\n\nInitial publish\n\nArchive index:\n\nArchive v1.0.4: 7 files, 18250 bytes\n\nFiles: examples/ai-boom-obviously-coming-hindsight-2022-2026.md (8514b), examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md (9849b), examples/caplan-posner-cheney-anesthesia-outcome-study-1991.md (4772b), references/sources.md (2372b), skill-card.md (2337b), SKILL.md (9389b), _meta.json (133b)\n\nFile v1.0.4:SKILL.md\n\n---\nname: hindsight-bias\ndescription: \"Activate when: someone says 'I knew it all along' or 'we should have seen it coming'; a post-mortem is blaming someone for not predicting an outcome; a team is reviewing a past decision and the outcome is coloring the judgment; a decision-maker is being evaluated on what happened rather than what was knowable at the time.\n  Do NOT activate when: contemporaneous pre-decision records exist and match current memory (bias is bounded); the goal is explicitly pattern-recognition from outcomes rather than evaluating the original decision-maker. More: deciqai.com/c/hindsight-bias\"\n---\n\n# Hindsight Bias\n\n## Overview\n\n**Hindsight bias** — the \"I knew it all along\" effect — is the tendency, *after* learning an outcome, to misremember your prior judgment as having been closer to that outcome than it actually was. Fischhoff (1975) demonstrated three distinct components: **memory distortion** (\"I said it would happen\"), **inevitability** (\"it had to happen\"), and **foreseeability** (\"anyone should have seen it\"). Each requires a different countermeasure; the structural fix for all three is pre-commitment documentation.\n\nComposes with `probabilistic-thinking`, `premortem`, `confirmation-bias`, and `survivorship-bias`.\n\n## When to Use\n\n- A post-mortem is becoming an exercise in blame for an \"obviously foreseeable\" outcome\n- Someone says \"I knew this would happen\" without contemporaneous records\n- A decision is being evaluated against its outcome rather than the information available when made\n- An investor, judge, or jury evaluates a past decision with knowledge of how it turned out\n- Someone calls a market move \"obvious in retrospect\" — e.g. the AI boom, AI capex buildout, or frontier-AI valuations were \"clearly inevitable\" / \"anyone should have seen the AI adoption wave coming\"\n\n**Not when:** pre-commitment documentation exists and matches current memory; the goal is pattern-recognition from outcomes; the failure to predict was a genuine process failure with clear pre-outcome signals.\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user has a specific post-outcome narrative → run The Process directly.\n- **Coach mode:** user is unfamiliar or has no concrete case → guide step by step.\n\nIn Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.\n\n1. One-line: before judging a past decision by its outcome, ask what was actually knowable in advance — check the contemporaneous record, not your current memory.\n2. Check fit against When to Use / When NOT to use. If records exist and match memory, bias is bounded.\n3. Elicit the specific outcome and the claim of foreseeability. What outcome happened? Who is being credited/blamed?\n> **[WAIT — do not advance until user responds]**\n4. Run The Process one step at a time: reconstruct the pre-outcome information set, decompose the three components, evaluate decision process vs. outcome.\n> **[WAIT — do not advance until user responds]**\n5. Close by naming the insight uncovered and the structural fix (decision journal, pre-mortem, blind post-mortem).\n> **[WAIT — do not advance until user responds]**\n\n## The Process\n\n**Step 1 — State the outcome and the claim:** outcome that occurred / who is being credited-blamed / claim of foreseeability (verbatim) / time elapsed.\n\n**Step 2 — Reconstruct the pre-outcome information set:** what was knowable / what was genuinely uncertain / what contemporaneous records exist / consensus view at the time. *Nixon test: if informed contemporaries gave the outcome ≤30%, \"anyone should have seen it\" is hindsight bias.*\n\n**Step 3 — Decompose the three components:** memory distortion (current memory vs. records) / inevitability (is outcome framed as the only possible result?) / foreseeability (\"anyone should have seen it\" applied to genuinely uncertain ex-ante info?). Countermeasures: memory → pull records; inevitability → list 3-5 alternative outcomes; foreseeability → identify what pre-outcome info would have made it predictable.\n\n**Step 4 — Evaluate the decision process, not the outcome:** was the decision reasonable given available info / expected value across plausible outcomes / would you make the same decision again with the same info. *Bridgewater matrix: good decision + bad outcome = bad luck (don't change). Bad decision + good outcome = good luck (don't celebrate).*\n\n**Step 5 — Install pre-commitment documentation:** decision journal (who/what/when/why/probability) / pre-mortem record / public prediction logged to internal forum / pre-registered evaluation criteria.\n\n**Step 6 — Run a blind post-mortem:** read decision journal before outcome data / evaluate against recorded reasoning / introduce outcome data only then / distinguish process error from outcome variance.\n\n## Output Template\n\n```markdown\n# Hindsight Bias Analysis: <outcome>\nOutcome / Foreseeability claim (verbatim) / Who is credited-blamed:\nPre-outcome info set (knowable / uncertain / records / consensus / ex-ante probability):\nThree-component diagnosis (memory distortion Y/N / inevitability Y/N / foreseeability overclaim Y/N):\nDecision-process evaluation (process quality 1-5 / reasonable Y/N / EV / same decision again Y/N):\nStructural fix (decision journal owner / pre-mortem owner / blind post-mortem protocol Y/N):\n```\n\n*→ Method in Action: [Baruch Fischhoff's Nixon-China Trip Study, 1972-1975](examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md) · [Anesthesia Malpractice Outcome-Bias Study, 1991](examples/caplan-posner-cheney-anesthesia-outcome-study-1991.md)*\n*→ 2026 lens: [\"The AI Boom Was Obviously Coming\" — retrospective inevitability in the ChatGPT/Nvidia era (2022–2026)](examples/ai-boom-obviously-coming-hindsight-2022-2026.md)*\n\n## Pack: Hindsight Bias Patterns\n\n| Domain | Common hindsight manifestation | Structural countermeasure |\n|---|---|---|\n| Investment | \"Obvious in retrospect\" trade | Decision journal with logged thesis and probability |\n| Engineering | \"How did anyone miss that bug?\" | Blameless post-mortem; reconstruct info state at time |\n| Hiring | \"I knew this person wouldn't work out\" | Logged interview rubric and pre-hire confidence rating |\n| Legal | \"The defendant should have foreseen the harm\" | Ex ante risk assessment; \"reasonable person at the time\" |\n\n## Applying It Well\n\n- Pull contemporaneous records before forming any judgment about what was \"knowable\"\n- Evaluate decision quality and outcome quality on separate axes — never conflate them\n- The structural fix is records, not vigilance — warnings reduce but do not eliminate the bias\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\n\n## Common Rationalizations\n\n**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**\n\n| Fake move | Reality |\n|---|---|\n| [D] \"I remember exactly what I said at the time\" | Confident-memory subjects were *just as biased*. Memory without records is unreliable. |\n| [D] \"But it was so obvious in retrospect\" | Hindsight makes everything look obvious. Test: did informed contemporaries give it >50% probability before the outcome? |\n| [D] \"Anyone with sense would have predicted this\" | Check the contemporaneous record — pre-event surveys, market prices, expert forecasts. |\n| [D] \"The outcome was inevitable given the structure\" | List 3-5 plausible alternative outcomes the same structure could have produced. |\n| [D] \"Process doesn't matter; outcome is what matters\" | Outcome alone confounds skill and luck. Evaluate them separately. |\n| [D] \"We learned the lesson — that's all that matters\" | If the \"lesson\" is hindsight-biased, the next decision will be miscalibrated. |\n| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |\n\n## Red Flags\n\n- Post-mortem conducted with no pre-mortem record to compare against\n- \"Obvious in retrospect,\" \"anyone should have seen it,\" \"we knew this\" without contemporaneous documentation\n- \"Lesson learned\" focuses on the specific failure mode, not on the broader uncertainty space\n- Decision-maker blamed/credited based primarily on outcome rather than process\n- Emotional or financial stakes are high — the bias is *larger* in high-stakes cases\n\n## Verification\n\n- [ ] Contemporaneous records pulled (emails, memos, forecasts, market prices, decision journals)\n- [ ] Three components diagnosed separately (memory / inevitability / foreseeability)\n- [ ] At least 3 plausible alternative outcomes enumerated for the same starting structure\n- [ ] Decision quality evaluated separately from outcome quality\n- [ ] Ex-ante probability estimated using pre-outcome information only\n- [ ] Pre-commitment documentation installed for future similar decisions\n- [ ] Post-mortem protocol blind to outcome data until after process review\n\n---\n\n*Part of **deciqAI Knowledge Skills** — 227 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. **See it run → https://www.deciqai.com/c/hindsight-bias** · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.*\n\n*Agents: latest version & machine-readable metadata → https://www.deciqai.com/s/hindsight-bias.json*\n\nFile v1.0.4:_meta.json\n\n{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"hindsight-bias\",\n  \"version\": \"1.0.4\",\n  \"publishedAt\": 1784224939835\n}\n\nFile v1.0.4:references/sources.md\n\n# Sources — hindsight-bias\n\n> *Primary sources for the [hindsight-bias](../SKILL.md) skill.*\n\n## Sources\n\n- Fischhoff, B. (1975). \"Hindsight ≠ foresight: The effect of outcome knowledge on judgment under uncertainty.\" *Journal of Experimental Psychology: Human Perception and Performance*, 1(3), 288-299. The foundational paper.\n- Fischhoff, B., & Beyth, R. (1975). \"'I knew it would happen': Remembered probabilities of once-future things.\" *Organizational Behavior and Human Performance*, 13(1), 1-16. The Nixon-China study.\n- Slovic, P., & Fischhoff, B. (1977). \"On the psychology of experimental surprises.\" *Journal of Experimental Psychology: Human Perception and Performance*, 3(4), 544-551.\n- Christensen-Szalanski, J. J. J., & Willham, C. F. (1991). \"The hindsight bias: A meta-analysis.\" *Organizational Behavior and Human Decision Processes*, 48(1), 147-168.\n- Hawkins, S. A., & Hastie, R. (1990). \"Hindsight: Biased judgments of past events after the outcomes are known.\" *Psychological Bulletin*, 107(3), 311-327.\n- Kamin, K. A., & Rachlinski, J. J. (1995). \"Ex post ≠ ex ante: Determining liability in hindsight.\" *Law and Human Behavior*, 19(1), 89-104.\n- Caplan, R. A., Posner, K. L., & Cheney, F. W. (1991). \"Effect of outcome on physician judgments of appropriateness of care.\" *JAMA*, 265(15), 1957-1960. Outcome bias in expert medical peer review.\n- Baron, J., & Hershey, J. C. (1988). \"Outcome bias in decision evaluation.\" *Journal of Personality and Social Psychology*, 54(4), 569-579.\n- Roese, N. J., & Vohs, K. D. (2012). \"Hindsight bias.\" *Perspectives on Psychological Science*, 7(5), 411-426. The canonical review.\n- Tetlock, P. E., & Gardner, D. (2015). *Superforecasting: The Art and Science of Prediction.* Crown. ISBN 978-0804136693.\n- Allspaw, J. (2012). \"Blameless PostMortems and a Just Culture.\" Code as Craft (Etsy engineering blog).\n- OpenAI (2022). \"Introducing ChatGPT.\" OpenAI blog, November 30, 2022 (contemporaneous \"research preview\" framing of the launch; used in the 2022–2026 AI-boom hindsight example).\n- Nvidia Corporation (2023–2025). Quarterly financial reports and investor relations disclosures on data-center revenue, plus contemporaneous reporting of market-capitalization milestones (Reuters, Bloomberg, Financial Times). Basis for the ex-ante-vs-ex-post valuation record in the AI-boom hindsight example.\n\nFile v1.0.4:examples/ai-boom-obviously-coming-hindsight-2022-2026.md\n\n# Method in Action: \"The AI Boom Was Obviously Coming\" — Retrospective Inevitability in the ChatGPT/Nvidia Era (2022–2026)\n\n> *Example for the [hindsight-bias](../SKILL.md) skill.*\n\nFischhoff's lab work and the Caplan-Posner-Cheney medical study both isolated hindsight bias in controlled settings. The 2022–2026 AI boom is the same distortion playing out at civilizational scale and in real time: after ChatGPT's launch and Nvidia's run, a dominant narrative emerged that the generative-AI wave and the compute buildout were \"obviously coming\" — a claim that, if believed, corrupts how investors, operators, and policymakers calibrate the *next* set of bets.\n\n**Step 1 — State the outcome and the claim.** The outcome: OpenAI released ChatGPT on November 30, 2022; it reached a consumer adoption curve widely reported at the time as among the fastest on record (roughly 1M users in its first days and, by contemporaneous estimates, ~100M monthly users within about two months). Over 2023–2025, Nvidia's data-center GPU business drove its market capitalization from well under \\$1 trillion into the multi-trillion-dollar range, briefly the most valuable company in the world, and hyperscalers announced enormous AI-infrastructure capital expenditure. The foreseeability claim, heard constantly by 2024–2025: *\"This was obvious — transformers had been around since 2017, scaling laws were published, of course LLMs and the GPU buildout were inevitable; anyone paying attention should have seen it coming.\"* Who is credited/blamed: those who bought Nvidia early or funded AI look like visionaries; those who didn't are cast as having missed the obvious. Time elapsed: roughly 1–3 years — short enough that contemporaneous records are abundant, which is exactly what makes this case testable rather than merely arguable.\n\n**Step 2 — Reconstruct the pre-outcome information set.** What was genuinely knowable in mid-2022: the transformer architecture existed (Vaswani et al., 2017); OpenAI's scaling-law papers and GPT-3 (2020) were public; Nvidia GPUs dominated ML training. What was genuinely uncertain: whether a chat interface would trigger mass *consumer* adoption; whether enterprises would pay at scale; whether capability would keep improving with scale or plateau; whether Nvidia's margins and lead would hold against custom silicon and competitors. The contemporaneous consensus is the tell. Before the launch, generative AI was a niche within ML; the prior consumer-AI cycles (chatbots, IBM Watson, voice assistants) had underdelivered relative to hype, priming informed observers toward skepticism. OpenAI itself reportedly launched ChatGPT as a low-key \"research preview,\" not expecting the response — a strong signal that the people closest to the technology did not treat the outcome as inevitable. *Nixon test:* if you had polled informed contemporaries in October 2022 on \"a chatbot will reach ~100M users within two months and a chip vendor will become the world's most valuable company on AI demand,\" the assigned probability would have been far below 30%. That gap is the measure of the hindsight distortion.\n\n**Step 3 — Decompose the three components.**\n- **Memory distortion** (\"I said it would happen\"): people now recall having been more bullish than their contemporaneous record shows. The countermeasure is records — 2021–2022 fund letters, analyst notes, forum posts, and price action. The market's own pricing is the cleanest record: Nvidia traded at a small fraction of its later valuation, and most AI-exposed equities were not priced for this outcome, which is direct evidence that the aggregate informed view did *not* foresee it.\n- **Inevitability** (\"it had to happen\"): the boom is retold as the only possible result of transformers + scaling + GPUs. Countermeasure — list 3–5 plausible alternatives that the *same* starting structure could have produced: (a) capability plateaus after GPT-3.5-class models, deflating the wave; (b) a competitor or open-source stack commoditizes inference and compresses Nvidia's margins; (c) the consumer interface flops and generative AI stays a B2B API niche; (d) an early high-profile safety, legal, or hallucination failure triggers regulatory freezes; (e) custom accelerators erode the GPU moat faster. Each was a live branch in 2022.\n- **Foreseeability overclaim** (\"anyone should have seen it\"): applying \"obvious\" to ex-ante-uncertain facts. The existence of the ingredients (transformers, scaling laws) did not make the *outcome* — timing, magnitude, and specifically the consumer-adoption and valuation shocks — predictable. Ingredients being visible is not the outcome being foreseeable.\n\n**Step 4 — Evaluate the decision process, not the outcome.** *Bridgewater matrix:* an investor who bought Nvidia in 2022 for a *wrong* reason (e.g., a gaming-GPU thesis) and profited had a good outcome from a mediocre process — good luck, don't over-learn from it. An analyst who assigned the AI boom a reasoned 15% probability and sized a bet accordingly ran a *good process* even though the base case didn't happen — don't punish that as a miss. Conversely, a skeptic who reasoned carefully from the weak track record of prior consumer-AI cycles made a defensible decision that a low-probability branch overturned: bad outcome, not necessarily bad process. The danger of the \"obviously coming\" narrative is that it rewards process by outcome, teaching everyone to chase the last trade and to treat the winners as oracles — precisely the miscalibration that inflates the *next* bubble.\n\n**Step 5 — Install pre-commitment documentation.** The structural fix is a decision journal, not vigilance. For each AI-era bet, log contemporaneously: the thesis, the probability assigned to the key uncertain event (adoption, capability continuation, margin durability), the disconfirming signals that would change your mind, and the position size. This converts \"I always knew AI would be huge\" into a checkable record. Applied forward — to 2025–2026 questions like AI capex sustainability, data-center ROI, and whether frontier valuations are justified — a pre-registered thesis is the only thing that will let you distinguish, later, between having been right and merely remembering yourself as right.\n\n**Step 6 — Run a blind post-mortem.** When reviewing AI bets in 2026 and beyond, read the decision journal *before* looking at returns. Evaluate the reasoning against the information available at decision time; only then admit the outcome data. This is what separates \"our AI thesis was well-reasoned and got lucky/unlucky\" from \"our thesis was sound.\" Without it, every post-mortem collapses into the inevitability narrative, and the lesson learned will itself be hindsight-biased — miscalibrating the next capital-allocation and risk decision, whether that is doubling down on AI infrastructure or fleeing it.\n\nThe mapped steps:\n1. Outcome and claim: ChatGPT's launch and Nvidia's rise; \"obvious — the ingredients existed, anyone should have seen it coming\"\n2. Pre-outcome information set: ingredients knowable (transformers, scaling, GPUs); outcome uncertain (consumer adoption, valuation magnitude, margin durability); pre-launch consensus skeptical; ex-ante probability well below 30%\n3. Three-component decomposition: memory distortion (records/prices contradict recalled bullishness); inevitability (5 live alternative branches in 2022); foreseeability overclaim (visible ingredients ≠ foreseeable outcome)\n4. Process vs. outcome: Bridgewater matrix separates lucky winners from good process; the narrative wrongly rewards process by outcome\n5. Pre-commitment documentation: decision journal with logged thesis, probability, disconfirmers, and size — applied forward to AI capex/valuation bets\n6. Blind post-mortem: read the journal before returns; keep the lesson from becoming hindsight-biased and miscalibrating the next bet\n\n*Sources: OpenAI, \"Introducing ChatGPT\" (blog, November 30, 2022); Vaswani et al., \"Attention Is All You Need,\" NeurIPS 2017; Brown et al., \"Language Models are Few-Shot Learners\" (GPT-3), 2020; Kaplan et al., \"Scaling Laws for Neural Language Models,\" 2020; Nvidia investor relations and quarterly reports, 2023–2025 (data-center revenue and market-capitalization milestones as widely reported by Reuters, Bloomberg, and the Financial Times); Fischhoff, B. (1975), \"Hindsight ≠ foresight,\" Journal of Experimental Psychology: HPP, 1(3), 288–299.*\n\nFile v1.0.4:examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md\n\n# Method in Action: Baruch Fischhoff's Nixon-China Trip Study, 1972-1975\n\n> *Example for the [hindsight-bias](../SKILL.md) skill.*\n\nThe empirical foundation for hindsight bias as a formal cognitive phenomenon is **Baruch Fischhoff**'s doctoral work at Hebrew University, published in two landmark 1975 papers:\n\n> Fischhoff, B. (1975). \"Hindsight ≠ foresight: The effect of outcome knowledge on judgment under uncertainty.\" *Journal of Experimental Psychology: Human Perception and Performance*, 1(3), 288-299.\n>\n> Fischhoff, B., & Beyth, R. (1975). \"'I knew it would happen': Remembered probabilities of once-future things.\" *Organizational Behavior and Human Performance*, 13(1), 1-16.\n\nThe Nixon-Beyth study (often called the \"Nixon-China study\") had a clean design. In February 1972, before President Richard Nixon's two-week diplomatic tour of China (Feb 21-28) and the Soviet Union (May 22-30), Fischhoff and Beyth presented Israeli undergraduate subjects with a list of 15 possible outcomes:\n\n- Nixon would meet Chairman Mao Zedong\n- The U.S. would establish a permanent diplomatic mission in Peking\n- The U.S. and China would issue a joint communiqué\n- Nixon would announce a U.S. policy of two-Chinas (recognizing both Taipei and Peking)\n- A group of American POWs from Vietnam would be released\n- ... (10 more)\n\nSubjects estimated the probability of each event, on a 0-100% scale. The trips then occurred. Some events happened (Nixon-Mao meeting, joint communiqué), others did not (no permanent mission established, no two-Chinas announcement, no POW release).\n\nThen, two weeks, two months, and four months after the trips, Fischhoff asked the same subjects to *recall* their original probability estimates. He compared the recalled estimates with the actual original responses.\n\nThe findings, in Fischhoff and Beyth's own words:\n\n> \"Subjects' memory for the probabilities they had originally assigned was systematically biased in the direction of the actual outcome. For events that had occurred, subjects recalled having predicted higher probabilities than they actually had recorded. For events that had not occurred, they recalled lower probabilities than they had recorded. The mean absolute error in recall was 9.4 percentage points at two weeks, rising to 11.5 percentage points at four months. The direction of the shift was consistent with hindsight bias in 70% to 85% of cases, depending on the event and the time delay.\"\n>\n> — Fischhoff & Beyth (1975), p. 8.\n\nThe effect was not eliminated by:\n- High confidence in memory (subjects who were *most* sure they remembered correctly were just as biased).\n- Interest in the political subject matter (engaged subjects were biased just as much as uninterested ones).\n- A short time delay (the bias was already substantial at two weeks).\n\nIn a companion study, Fischhoff's 1975 *JEP:HPP* paper used the British-Gurkha War of 1814 — a historical event most subjects had no prior knowledge of. He gave subjects a 150-word description of the strategic situation and asked them to estimate the probability of four possible outcomes: British victory, Gurkha victory, stalemate, no military resolution. Critically, different groups of subjects were told different outcomes had actually occurred before being asked for their probability estimates. The instruction was explicit: \"Estimate the probabilities *as if you did not know which outcome occurred*.\"\n\nFischhoff's finding:\n\n> \"Subjects told that the British had won assigned the British victory a mean probability of 57.2%. Subjects told that the Gurkhas had won assigned a Gurkha victory a mean probability of 38.4% — substantially higher than the 21.3% assigned by subjects told there had been no military resolution. The mere knowledge of the outcome shifted estimated probabilities upward by 15 to 25 percentage points, *despite explicit instructions to ignore that knowledge*. Subjects were unable to disregard outcome information when reconstructing how probable the outcome 'should have looked' beforehand.\"\n>\n> — Fischhoff (1975), p. 293.\n\nThis was the breakthrough finding: the bias is not in deliberate self-deception, nor in motivated reasoning, nor in failure of effort. It is in the cognitive impossibility of reconstructing a prior epistemic state that has been overwritten by subsequent information. The mind cannot un-know what it knows.\n\nThe implications were operationally massive:\n\n**Post-mortem unfairness.** Every post-disaster analysis (investment loss, business failure, military defeat, medical misdiagnosis, engineering accident) is conducted by people who know the outcome. Their reconstruction of \"what should have been obvious\" is systematically biased toward making the actual outcome seem more predictable than it was. Decision-makers are routinely judged as negligent for failing to foresee outcomes that, ex ante, were genuinely uncertain.\n\n**Legal liability.** Tort law often hinges on whether a defendant \"should have known\" of a risk. Hindsight bias systematically inflates jurors' assessments of foreseeability. Empirical studies (Kamin & Rachlinski 1995, *Law and Human Behavior* 19(1)) show that mock jurors evaluate the same precautionary decision as negligent at much higher rates when they know an accident occurred than when they evaluate the decision ex ante.\n\n**Investment evaluation.** Returns are visible; the decision process is not. A fund manager who took a 30%-probability bet that paid off looks like a genius; one who took the same bet and lost looks like a fool. Both decisions had identical expected value. Hindsight bias rewards lucky decisions and punishes unlucky ones, generating systematic miscalibration in portfolio assessment.\n\n**Strategic planning.** \"We should have predicted the financial crisis / the pandemic / the rise of X / the fall of Y\" — these claims are almost always hindsight-biased. The information available before the event would, in honest reconstruction, support a much lower probability estimate than the post-event narrative suggests.\n\nThe follow-up literature established the bias as universal and structural:\n\n**Slovic & Fischhoff (1977).** *JEP:HPP* 3(4). Showed the bias persists when subjects are explicitly trained in it and warned about it. Self-vigilance is weakly effective.\n\n**Christensen-Szalanski & Willham (1991).** *OBHDP* 48(1). A meta-analysis of 122 studies confirmed the bias is robust across populations, domains, and methodologies. Effect size: Cohen's d ≈ 0.39, considered moderate-to-large in social psychology.\n\n**Roese & Vohs (2012).** *Perspectives on Psychological Science* 7(5), 411-426. The canonical review. Decomposed the bias into the three components (memory, inevitability, foreseeability) and reviewed evidence that each has distinct underlying mechanisms and distinct countermeasures.\n\n**Tetlock (2015).** *Superforecasting.* Crown. ISBN 978-0804136693. The Good Judgment Project's calibration-training methodology is, in operational terms, an industrial-scale antidote to hindsight bias: explicit pre-commitment of probability estimates, then scoring against outcomes. Forecasters trained this way reduced hindsight bias by ~40% compared to controls (Mellers et al. 2014, *Psychological Science* 25(5)).\n\nThe framework has reshaped multiple operational disciplines:\n\n**Aviation accident investigation.** The NTSB's investigation protocol explicitly warns against hindsight bias and requires investigators to reconstruct the cockpit information environment *as the pilots experienced it*, before introducing outcome data. The investigation is conducted in two phases (facts first, then analysis) to bound the bias.\n\n**Medical M&M (Morbidity & Mortality) conferences.** Modern protocols require discussion of the diagnostic reasoning available at each decision point, with outcome data introduced only after the process review. The bias is bounded structurally because the post-mortem is split temporally.\n\n**Software engineering blameless post-mortems.** Pioneered at Etsy and Google, the practice (described in Allspaw 2012, \"Blameless PostMortems\") explicitly trains responders to reconstruct the engineer's information state at the time of action. Outcome-driven blame is institutionally banned. The goal is to learn about the system, not to punish foreseeability.\n\n**Investment decision journals.** Adopted by Bridgewater (Dalio 2017), Ray Dalio's \"issue log,\" and many systematic investors. Predictions and reasoning are logged before outcomes. Quarterly reviews compare logged reasoning against outcomes, distinguishing process error from outcome variance.\n\n**Intelligence analysis.** The CIA's structured analytic techniques (Heuer & Pherson 2010) include \"What If? Analysis\" and \"Devil's Advocacy,\" both designed to make plausible alternative outcomes salient and to bound the \"anyone should have seen it coming\" component of hindsight bias.\n\nThree operational lessons from Fischhoff:\n\n**First, the bias is invisible to introspection.** Subjects who were *most confident* in their memory of prior judgments were just as biased. The corrective must be structural — contemporaneous records — not vigilance-based.\n\n**Second, the bias systematically transfers blame from the outcome to the decision-maker.** \"He should have known\" is the canonical phrase. In the absence of pre-commitment documentation, the bias produces unfair retrospective judgments at scale.\n\n**Third, the bias is amplified by emotion and stakes.** Studies of medical, legal, and military post-incident reviews show the bias is *larger* for high-stakes outcomes (Hawkins & Hastie 1990, *Psychological Bulletin* 107(3)). When the outcome was catastrophic, the certainty that \"we should have predicted it\" is correspondingly inflated — exactly when the institutional consequences of unfair retrospective judgment are highest.\n\nFile v1.0.4:examples/caplan-posner-cheney-anesthesia-outcome-study-1991.md\n\n# Method in Action: Outcome Bias in Anesthesia Malpractice Review — the Caplan-Posner-Cheney Study (1991)\n\n> *Example for the [hindsight-bias](../SKILL.md) skill.*\n\nFischhoff's lab experiments showed that outcome knowledge distorts probability reconstruction in undergraduates. The Caplan-Posner-Cheney study showed the same distortion operating in the highest-stakes professional judgment there is: board-certified physicians deciding whether a colleague's care was negligent.\n\n**Step 1 — State the outcome and the claim.** In malpractice review, the claim is always the same: \"the care was inappropriate — a competent anesthesiologist would have managed this differently.\" The reviewer renders that judgment *after* learning the patient's outcome. The question the researchers isolated: does the outcome itself — rather than the care — drive the verdict?\n\n**Step 2 — Reconstruct the pre-outcome information set.** Working from the American Society of Anesthesiologists' Closed Claims Project, Caplan, Posner, and Cheney selected 21 real case summaries describing the clinical facts available to the anesthesiologist during the case. Each case was prepared in two versions that were *identical in every clinical fact* — same patient, same monitoring data, same decisions — and differed only in the final sentence: one version ended in a temporary injury, the other in a permanent or fatal one. This is the cleanest possible reconstruction of the pre-outcome information set: hold it constant by construction, vary only the outcome.\n\n**Step 3 — Decompose the components.** 112 practicing anesthesiologists were randomly assigned one version of each case and rated the care as appropriate or less than appropriate. Because the ex-ante facts were identical across versions, any difference in ratings could come only from outcome knowledge. The result: the same care was judged significantly less appropriate when the reviewer believed the outcome was permanent rather than temporary. In a substantial fraction of the matched cases, changing nothing but the outcome sentence flipped the majority verdict on the quality of care. This is the **foreseeability** and **inevitability** components in the wild — expert reviewers, evaluating their own specialty, could not quarantine outcome knowledge from their judgment of the decision process.\n\n**Step 4 — Evaluate the decision process, not the outcome.** The study is the empirical demolition of \"outcome is what matters\" in professional review. Identical process, different luck, opposite verdicts. Reviewers were not careless or malicious; they were applying the standard implicit test — \"would a good anesthesiologist have ended up here?\" — which smuggles the outcome into the process evaluation. Baron and Hershey (1988) had shown the same conflation in lay subjects; Caplan et al. showed that domain expertise does not immunize.\n\n**Step 5 — Install pre-commitment documentation.** The operational lesson for medicine: appropriateness of care must be assessed against the information state at each decision point, documented contemporaneously (anesthesia records, monitoring logs, notes), not against the end state of the patient. The Closed Claims Project itself institutionalized this by having reviewers work from structured case abstracts of intraoperative data.\n\n**Step 6 — Run a blind post-mortem.** The study's matched-versions design *is* the blind post-mortem protocol, run as an experiment: review the process with the outcome masked or varied, and see whether the verdict survives. Modern morbidity-and-mortality conference protocols and two-phase accident investigations (facts first, analysis second) descend directly from this finding — the bias cannot be waived away by warning reviewers, so the review structure must withhold or neutralize outcome data.\n\nThe mapped steps:\n1. Outcome and claim: \"the care was inappropriate\" — rendered by reviewers who know how the patient fared\n2. Pre-outcome information set: held constant by construction — identical clinical facts, only the outcome sentence varied\n3. Three-component decomposition: foreseeability and inevitability isolated experimentally; expert reviewers judged identical care worse when the outcome was severe\n4. Process vs. outcome: same process, different luck, opposite verdicts — outcome alone confounds skill and luck\n5. Pre-commitment documentation: contemporaneous anesthesia records and structured case abstracts as the basis of review\n6. Blind post-mortem: outcome-masked review protocols; the matched-versions design as the template\n\nPrimary source: Caplan, R. A., Posner, K. L., & Cheney, F. W. (1991). \"Effect of outcome on physician judgments of appropriateness of care.\" *JAMA*, 265(15), 1957-1960.\n\nFile v1.0.4:skill-card.md\n\n## Description:\n\nHelps agents identify hindsight bias in post-outcome narratives, reconstruct what was knowable before the outcome, and separate decision quality from outcome quality.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[deciqai](https://clawhub.ai/user/deciqai)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nEmployees, external users, and decision reviewers use this skill to examine post-mortems, investment reviews, legal or medical judgment reviews, and other retrospectives where outcome knowledge may distort evaluation of the original decision.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Users may provide private business, legal, medical, or investment records during a decision review.\n\nMitigation: Share only the minimum records needed for the hindsight-bias analysis and avoid unnecessary sensitive details.\n\nRisk: A hindsight-bias analysis can become misleading when current memory replaces contemporaneous evidence.\n\nMitigation: Use pre-outcome records, estimate ex-ante probability from information available at the time, and distinguish decision process from outcome.\n\n## Reference(s):\n\n- [Primary Sources](references/sources.md)\n- [Baruch Fischhoff's Nixon-China Trip Study, 1972-1975](examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md)\n- [Outcome Bias in Anesthesia Malpractice Review, 1991](examples/caplan-posner-cheney-anesthesia-outcome-study-1991.md)\n- [Retrospective Inevitability in the ChatGPT/Nvidia Era, 2022-2026](examples/ai-boom-obviously-coming-hindsight-2022-2026.md)\n- [ClawHub Skill Page](https://clawhub.ai/deciqai/skills/hindsight-bias)\n\n## Skill Output:\n\n**Output Type(s):** [text, markdown, guidance, analysis]\n\n**Output Format:** [Markdown analysis template with structured findings and recommended process fixes]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [May ask the user for case details one step at a time before producing the final analysis.]\n\n## Skill Version(s):\n\n1.0.4 (source: server release evidence)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nArchive v1.0.3: 7 files, 18362 bytes\n\nFiles: examples/ai-boom-obviously-coming-hindsight-2022-2026.md (8514b), examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md (9849b), examples/caplan-posner-cheney-anesthesia-outcome-study-1991.md (4772b), references/sources.md (2372b), skill-card.md (2856b), SKILL.md (9250b), _meta.json (133b)\n\nFile v1.0.3:SKILL.md\n\n---\nname: hindsight-bias\ndescription: \"Activate when: someone says 'I knew it all along' or 'we should have seen it coming'; a post-mortem is blaming someone for not predicting an outcome; a team is reviewing a past decision and the outcome is coloring the judgment; a decision-maker is being evaluated on what happened rather than what was knowable at the time.\n  Do NOT activate when: contemporaneous pre-decision records exist and match current memory (bias is bounded); the goal is explicitly pattern-recognition from outcomes rather than evaluating the original decision-maker.\"\n---\n\n# Hindsight Bias\n\n## Overview\n\n**Hindsight bias** — the \"I knew it all along\" effect — is the tendency, *after* learning an outcome, to misremember your prior judgment as having been closer to that outcome than it actually was. Fischhoff (1975) demonstrated three distinct components: **memory distortion** (\"I said it would happen\"), **inevitability** (\"it had to happen\"), and **foreseeability** (\"anyone should have seen it\"). Each requires a different countermeasure; the structural fix for all three is pre-commitment documentation.\n\nComposes with `probabilistic-thinking`, `premortem`, `confirmation-bias`, and `survivorship-bias`.\n\n## When to Use\n\n- A post-mortem is becoming an exercise in blame for an \"obviously foreseeable\" outcome\n- Someone says \"I knew this would happen\" without contemporaneous records\n- A decision is being evaluated against its outcome rather than the information available when made\n- An investor, judge, or jury evaluates a past decision with knowledge of how it turned out\n- Someone calls a market move \"obvious in retrospect\" — e.g. the AI boom, AI capex buildout, or frontier-AI valuations were \"clearly inevitable\" / \"anyone should have seen the AI adoption wave coming\"\n\n**Not when:** pre-commitment documentation exists and matches current memory; the goal is pattern-recognition from outcomes; the failure to predict was a genuine process failure with clear pre-outcome signals.\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user has a specific post-outcome narrative → run The Process directly.\n- **Coach mode:** user is unfamiliar or has no concrete case → guide step by step.\n\nIn Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.\n\n1. One-line: before judging a past decision by its outcome, ask what was actually knowable in advance — check the contemporaneous record, not your current memory.\n2. Check fit against When to Use / When NOT to use. If records exist and match memory, bias is bounded.\n3. Elicit the specific outcome and the claim of foreseeability. What outcome happened? Who is being credited/blamed?\n> **[WAIT — do not advance until user responds]**\n4. Run The Process one step at a time: reconstruct the pre-outcome information set, decompose the three components, evaluate decision process vs. outcome.\n> **[WAIT — do not advance until user responds]**\n5. Close by naming the insight uncovered and the structural fix (decision journal, pre-mortem, blind post-mortem).\n> **[WAIT — do not advance until user responds]**\n\n## The Process\n\n**Step 1 — State the outcome and the claim:** outcome that occurred / who is being credited-blamed / claim of foreseeability (verbatim) / time elapsed.\n\n**Step 2 — Reconstruct the pre-outcome information set:** what was knowable / what was genuinely uncertain / what contemporaneous records exist / consensus view at the time. *Nixon test: if informed contemporaries gave the outcome ≤30%, \"anyone should have seen it\" is hindsight bias.*\n\n**Step 3 — Decompose the three components:** memory distortion (current memory vs. records) / inevitability (is outcome framed as the only possible result?) / foreseeability (\"anyone should have seen it\" applied to genuinely uncertain ex-ante info?). Countermeasures: memory → pull records; inevitability → list 3-5 alternative outcomes; foreseeability → identify what pre-outcome info would have made it predictable.\n\n**Step 4 — Evaluate the decision process, not the outcome:** was the decision reasonable given available info / expected value across plausible outcomes / would you make the same decision again with the same info. *Bridgewater matrix: good decision + bad outcome = bad luck (don't change). Bad decision + good outcome = good luck (don't celebrate).*\n\n**Step 5 — Install pre-commitment documentation:** decision journal (who/what/when/why/probability) / pre-mortem record / public prediction logged to internal forum / pre-registered evaluation criteria.\n\n**Step 6 — Run a blind post-mortem:** read decision journal before outcome data / evaluate against recorded reasoning / introduce outcome data only then / distinguish process error from outcome variance.\n\n## Output Template\n\n```markdown\n# Hindsight Bias Analysis: <outcome>\nOutcome / Foreseeability claim (verbatim) / Who is credited-blamed:\nPre-outcome info set (knowable / uncertain / records / consensus / ex-ante probability):\nThree-component diagnosis (memory distortion Y/N / inevitability Y/N / foreseeability overclaim Y/N):\nDecision-process evaluation (process quality 1-5 / reasonable Y/N / EV / same decision again Y/N):\nStructural fix (decision journal owner / pre-mortem owner / blind post-mortem protocol Y/N):\n```\n\n*→ Method in Action: [Baruch Fischhoff's Nixon-China Trip Study, 1972-1975](examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md) · [Anesthesia Malpractice Outcome-Bias Study, 1991](examples/caplan-posner-cheney-anesthesia-outcome-study-1991.md)*\n*→ 2026 lens: [\"The AI Boom Was Obviously Coming\" — retrospective inevitability in the ChatGPT/Nvidia era (2022–2026)](examples/ai-boom-obviously-coming-hindsight-2022-2026.md)*\n\n## Pack: Hindsight Bias Patterns\n\n| Domain | Common hindsight manifestation | Structural countermeasure |\n|---|---|---|\n| Investment | \"Obvious in retrospect\" trade | Decision journal with logged thesis and probability |\n| Engineering | \"How did anyone miss that bug?\" | Blameless post-mortem; reconstruct info state at time |\n| Hiring | \"I knew this person wouldn't work out\" | Logged interview rubric and pre-hire confidence rating |\n| Legal | \"The defendant should have foreseen the harm\" | Ex ante risk assessment; \"reasonable person at the time\" |\n\n## Applying It Well\n\n- Pull contemporaneous records before forming any judgment about what was \"knowable\"\n- Evaluate decision quality and outcome quality on separate axes — never conflate them\n- The structural fix is records, not vigilance — warnings reduce but do not eliminate the bias\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\n\n## Common Rationalizations\n\n**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**\n\n| Fake move | Reality |\n|---|---|\n| [D] \"I remember exactly what I said at the time\" | Confident-memory subjects were *just as biased*. Memory without records is unreliable. |\n| [D] \"But it was so obvious in retrospect\" | Hindsight makes everything look obvious. Test: did informed contemporaries give it >50% probability before the outcome? |\n| [D] \"Anyone with sense would have predicted this\" | Check the contemporaneous record — pre-event surveys, market prices, expert forecasts. |\n| [D] \"The outcome was inevitable given the structure\" | List 3-5 plausible alternative outcomes the same structure could have produced. |\n| [D] \"Process doesn't matter; outcome is what matters\" | Outcome alone confounds skill and luck. Evaluate them separately. |\n| [D] \"We learned the lesson — that's all that matters\" | If the \"lesson\" is hindsight-biased, the next decision will be miscalibrated. |\n| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |\n\n## Red Flags\n\n- Post-mortem conducted with no pre-mortem record to compare against\n- \"Obvious in retrospect,\" \"anyone should have seen it,\" \"we knew this\" without contemporaneous documentation\n- \"Lesson learned\" focuses on the specific failure mode, not on the broader uncertainty space\n- Decision-maker blamed/credited based primarily on outcome rather than process\n- Emotional or financial stakes are high — the bias is *larger* in high-stakes cases\n\n## Verification\n\n- [ ] Contemporaneous records pulled (emails, memos, forecasts, market prices, decision journals)\n- [ ] Three components diagnosed separately (memory / inevitability / foreseeability)\n- [ ] At least 3 plausible alternative outcomes enumerated for the same starting structure\n- [ ] Decision quality evaluated separately from outcome quality\n- [ ] Ex-ante probability estimated using pre-outcome information only\n- [ ] Pre-commitment documentation installed for future similar decisions\n- [ ] Post-mortem protocol blind to outcome data until after process review\n\n---\n\n*Part of **deciqAI Knowledge Skills** — 164 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. **See it run → https://www.deciqai.com/c/hindsight-bias** · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.*\n\nFile v1.0.3:_meta.json\n\n{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"hindsight-bias\",\n  \"version\": \"1.0.3\",\n  \"publishedAt\": 1783508708651\n}\n\nFile v1.0.3:references/sources.md\n\n# Sources — hindsight-bias\n\n> *Primary sources for the [hindsight-bias](../SKILL.md) skill.*\n\n## Sources\n\n- Fischhoff, B. (1975). \"Hindsight ≠ foresight: The effect of outcome knowledge on judgment under uncertainty.\" *Journal of Experimental Psychology: Human Perception and Performance*, 1(3), 288-299. The foundational paper.\n- Fischhoff, B., & Beyth, R. (1975). \"'I knew it would happen': Remembered probabilities of once-future things.\" *Organizational Behavior and Human Performance*, 13(1), 1-16. The Nixon-China study.\n- Slovic, P., & Fischhoff, B. (1977). \"On the psychology of experimental surprises.\" *Journal of Experimental Psychology: Human Perception and Performance*, 3(4), 544-551.\n- Christensen-Szalanski, J. J. J., & Willham, C. F. (1991). \"The hindsight bias: A meta-analysis.\" *Organizational Behavior and Human Decision Processes*, 48(1), 147-168.\n- Hawkins, S. A., & Hastie, R. (1990). \"Hindsight: Biased judgments of past events after the outcomes are known.\" *Psychological Bulletin*, 107(3), 311-327.\n- Kamin, K. A., & Rachlinski, J. J. (1995). \"Ex post ≠ ex ante: Determining liability in hindsight.\" *Law and Human Behavior*, 19(1), 89-104.\n- Caplan, R. A., Posner, K. L., & Cheney, F. W. (1991). \"Effect of outcome on physician judgments of appropriateness of care.\" *JAMA*, 265(15), 1957-1960. Outcome bias in expert medical peer review.\n- Baron, J., & Hershey, J. C. (1988). \"Outcome bias in decision evaluation.\" *Journal of Personality and Social Psychology*, 54(4), 569-579.\n- Roese, N. J., & Vohs, K. D. (2012). \"Hindsight bias.\" *Perspectives on Psychological Science*, 7(5), 411-426. The canonical review.\n- Tetlock, P. E., & Gardner, D. (2015). *Superforecasting: The Art and Science of Prediction.* Crown. ISBN 978-0804136693.\n- Allspaw, J. (2012). \"Blameless PostMortems and a Just Culture.\" Code as Craft (Etsy engineering blog).\n- OpenAI (2022). \"Introducing ChatGPT.\" OpenAI blog, November 30, 2022 (contemporaneous \"research preview\" framing of the launch; used in the 2022–2026 AI-boom hindsight example).\n- Nvidia Corporation (2023–2025). Quarterly financial reports and investor relations disclosures on data-center revenue, plus contemporaneous reporting of market-capitalization milestones (Reuters, Bloomberg, Financial Times). Basis for the ex-ante-vs-ex-post valuation record in the AI-boom hindsight example.\n\nFile v1.0.3:examples/ai-boom-obviously-coming-hindsight-2022-2026.md\n\n# Method in Action: \"The AI Boom Was Obviously Coming\" — Retrospective Inevitability in the ChatGPT/Nvidia Era (2022–2026)\n\n> *Example for the [hindsight-bias](../SKILL.md) skill.*\n\nFischhoff's lab work and the Caplan-Posner-Cheney medical study both isolated hindsight bias in controlled settings. The 2022–2026 AI boom is the same distortion playing out at civilizational scale and in real time: after ChatGPT's launch and Nvidia's run, a dominant narrative emerged that the generative-AI wave and the compute buildout were \"obviously coming\" — a claim that, if believed, corrupts how investors, operators, and policymakers calibrate the *next* set of bets.\n\n**Step 1 — State the outcome and the claim.** The outcome: OpenAI released ChatGPT on November 30, 2022; it reached a consumer adoption curve widely reported at the time as among the fastest on record (roughly 1M users in its first days and, by contemporaneous estimates, ~100M monthly users within about two months). Over 2023–2025, Nvidia's data-center GPU business drove its market capitalization from well under \\$1 trillion into the multi-trillion-dollar range, briefly the most valuable company in the world, and hyperscalers announced enormous AI-infrastructure capital expenditure. The foreseeability claim, heard constantly by 2024–2025: *\"This was obvious — transformers had been around since 2017, scaling laws were published, of course LLMs and the GPU buildout were inevitable; anyone paying attention should have seen it coming.\"* Who is credited/blamed: those who bought Nvidia early or funded AI look like visionaries; those who didn't are cast as having missed the obvious. Time elapsed: roughly 1–3 years — short enough that contemporaneous records are abundant, which is exactly what makes this case testable rather than merely arguable.\n\n**Step 2 — Reconstruct the pre-outcome information set.** What was genuinely knowable in mid-2022: the transformer architecture existed (Vaswani et al., 2017); OpenAI's scaling-law papers and GPT-3 (2020) were public; Nvidia GPUs dominated ML training. What was genuinely uncertain: whether a chat interface would trigger mass *consumer* adoption; whether enterprises would pay at scale; whether capability would keep improving with scale or plateau; whether Nvidia's margins and lead would hold against custom silicon and competitors. The contemporaneous consensus is the tell. Before the launch, generative AI was a niche within ML; the prior consumer-AI cycles (chatbots, IBM Watson, voice assistants) had underdelivered relative to hype, priming informed observers toward skepticism. OpenAI itself reportedly launched ChatGPT as a low-key \"research preview,\" not expecting the response — a strong signal that the people closest to the technology did not treat the outcome as inevitable. *Nixon test:* if you had polled informed contemporaries in October 2022 on \"a chatbot will reach ~100M users within two months and a chip vendor will become the world's most valuable company on AI demand,\" the assigned probability would have been far below 30%. That gap is the measure of the hindsight distortion.\n\n**Step 3 — Decompose the three components.**\n- **Memory distortion** (\"I said it would happen\"): people now recall having been more bullish than their contemporaneous record shows. The countermeasure is records — 2021–2022 fund letters, analyst notes, forum posts, and price action. The market's own pricing is the cleanest record: Nvidia traded at a small fraction of its later valuation, and most AI-exposed equities were not priced for this outcome, which is direct evidence that the aggregate informed view did *not* foresee it.\n- **Inevitability** (\"it had to happen\"): the boom is retold as the only possible result of transformers + scaling + GPUs. Countermeasure — list 3–5 plausible alternatives that the *same* starting structure could have produced: (a) capability plateaus after GPT-3.5-class models, deflating the wave; (b) a competitor or open-source stack commoditizes inference and compresses Nvidia's margins; (c) the consumer interface flops and generative AI stays a B2B API niche; (d) an early high-profile safety, legal, or hallucination failure triggers regulatory freezes; (e) custom accelerators erode the GPU moat faster. Each was a live branch in 2022.\n- **Foreseeability overclaim** (\"anyone should have seen it\"): applying \"obvious\" to ex-ante-uncertain facts. The existence of the ingredients (transformers, scaling laws) did not make the *outcome* — timing, magnitude, and specifically the consumer-adoption and valuation shocks — predictable. Ingredients being visible is not the outcome being foreseeable.\n\n**Step 4 — Evaluate the decision process, not the outcome.** *Bridgewater matrix:* an investor who bought Nvidia in 2022 for a *wrong* reason (e.g., a gaming-GPU thesis) and profited had a good outcome from a mediocre process — good luck, don't over-learn from it. An analyst who assigned the AI boom a reasoned 15% probability and sized a bet accordingly ran a *good process* even though the base case didn't happen — don't punish that as a miss. Conversely, a skeptic who reasoned carefully from the weak track record of prior consumer-AI cycles made a defensible decision that a low-probability branch overturned: bad outcome, not necessarily bad process. The danger of the \"obviously coming\" narrative is that it rewards process by outcome, teaching everyone to chase the last trade and to treat the winners as oracles — precisely the miscalibration that inflates the *next* bubble.\n\n**Step 5 — Install pre-commitment documentation.** The structural fix is a decision journal, not vigilance. For each AI-era bet, log contemporaneously: the thesis, the probability assigned to the key uncertain event (adoption, capability continuation, margin durability), the disconfirming signals that would change your mind, and the position size. This converts \"I always knew AI would be huge\" into a checkable record. Applied forward — to 2025–2026 questions like AI capex sustainability, data-center ROI, and whether frontier valuations are justified — a pre-registered thesis is the only thing that will let you distinguish, later, between having been right and merely remembering yourself as right.\n\n**Step 6 — Run a blind post-mortem.** When reviewing AI bets in 2026 and beyond, read the decision journal *before* looking at returns. Evaluate the reasoning against the information available at decision time; only then admit the outcome data. This is what separates \"our AI thesis was well-reasoned and got lucky/unlucky\" from \"our thesis was sound.\" Without it, every post-mortem collapses into the inevitability narrative, and the lesson learned will itself be hindsight-biased — miscalibrating the next capital-allocation and risk decision, whether that is doubling down on AI infrastructure or fleeing it.\n\nThe mapped steps:\n1. Outcome and claim: ChatGPT's launch and Nvidia's rise; \"obvious — the ingredients existed, anyone should have seen it coming\"\n2. Pre-outcome information set: ingredients knowable (transformers, scaling, GPUs); outcome uncertain (consumer adoption, valuation magnitude, margin durability); pre-launch consensus skeptical; ex-ante probability well below 30%\n3. Three-component decomposition: memory distortion (records/prices contradict recalled bullishness); inevitability (5 live alternative branches in 2022); foreseeability overclaim (visible ingredients ≠ foreseeable outcome)\n4. Process vs. outcome: Bridgewater matrix separates lucky winners from good process; the narrative wrongly rewards process by outcome\n5. Pre-commitment documentation: decision journal with logged thesis, probability, disconfirmers, and size — applied forward to AI capex/valuation bets\n6. Blind post-mortem: read the journal before returns; keep the lesson from becoming hindsight-biased and miscalibrating the next bet\n\n*Sources: OpenAI, \"Introducing ChatGPT\" (blog, November 30, 2022); Vaswani et al., \"Attention Is All You Need,\" NeurIPS 2017; Brown et al., \"Language Models are Few-Shot Learners\" (GPT-3), 2020; Kaplan et al., \"Scaling Laws for Neural Language Models,\" 2020; Nvidia investor relations and quarterly reports, 2023–2025 (data-center revenue and market-capitalization milestones as widely reported by Reuters, Bloomberg, and the Financial Times); Fischhoff, B. (1975), \"Hindsight ≠ foresight,\" Journal of Experimental Psychology: HPP, 1(3), 288–299.*\n\nFile v1.0.3:examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md\n\n# Method in Action: Baruch Fischhoff's Nixon-China Trip Study, 1972-1975\n\n> *Example for the [hindsight-bias](../SKILL.md) skill.*\n\nThe empirical foundation for hindsight bias as a formal cognitive phenomenon is **Baruch Fischhoff**'s doctoral work at Hebrew University, published in two landmark 1975 papers:\n\n> Fischhoff, B. (1975). \"Hindsight ≠ foresight: The effect of outcome knowledge on judgment under uncertainty.\" *Journal of Experimental Psychology: Human Perception and Performance*, 1(3), 288-299.\n>\n> Fischhoff, B., & Beyth, R. (1975). \"'I knew it would happen': Remembered probabilities of once-future things.\" *Organizational Behavior and Human Performance*, 13(1), 1-16.\n\nThe Nixon-Beyth study (often called the \"Nixon-China study\") had a clean design. In February 1972, before President Richard Nixon's two-week diplomatic tour of China (Feb 21-28) and the Soviet Union (May 22-30), Fischhoff and Beyth presented Israeli undergraduate subjects with a list of 15 possible outcomes:\n\n- Nixon would meet Chairman Mao Zedong\n- The U.S. would establish a permanent diplomatic mission in Peking\n- The U.S. and China would issue a joint communiqué\n- Nixon would announce a U.S. policy of two-Chinas (recognizing both Taipei and Peking)\n- A group of American POWs from Vietnam would be released\n- ... (10 more)\n\nSubjects estimated the probability of each event, on a 0-100% scale. The trips then occurred. Some events happened (Nixon-Mao meeting, joint communiqué), others did not (no permanent mission established, no two-Chinas announcement, no POW release).\n\nThen, two weeks, two months, and four months after the trips, Fischhoff asked the same subjects to *recall* their original probability estimates. He compared the recalled estimates with the actual original responses.\n\nThe findings, in Fischhoff and Beyth's own words:\n\n> \"Subjects' memory for the probabilities they had originally assigned was systematically biased in the direction of the actual outcome. For events that had occurred, subjects recalled having predicted higher probabilities than they actually had recorded. For events that had not occurred, they recalled lower probabilities than they had recorded. The mean absolute error in recall was 9.4 percentage points at two weeks, rising to 11.5 percentage points at four months. The direction of the shift was consistent with hindsight bias in 70% to 85% of cases, depending on the event and the time delay.\"\n>\n> — Fischhoff & Beyth (1975), p. 8.\n\nThe effect was not eliminated by:\n- High confidence in memory (subjects who were *most* sure they remembered correctly were just as biased).\n- Interest in the political subject matter (engaged subjects were biased just as much as uninterested ones).\n- A short time delay (the bias was already substantial at two weeks).\n\nIn a companion study, Fischhoff's 1975 *JEP:HPP* paper used the British-Gurkha War of 1814 — a historical event most subjects had no prior knowledge of. He gave subjects a 150-word description of the strategic situation and asked them to estimate the probability of four possible outcomes: British victory, Gurkha victory, stalemate, no military resolution. Critically, different groups of subjects were told different outcomes had actually occurred before being asked for their probability estimates. The instruction was explicit: \"Estimate the probabilities *as if you did not know which outcome occurred*.\"\n\nFischhoff's finding:\n\n> \"Subjects told that the British had won assigned the British victory a mean probability of 57.2%. Subjects told that the Gurkhas had won assigned a Gurkha victory a mean probability of 38.4% — substantially higher than the 21.3% assigned by subjects told there had been no military resolution. The mere knowledge of the outcome shifted estimated probabilities upward by 15 to 25 percentage points, *despite explicit instructions to ignore that knowledge*. Subjects were unable to disregard outcome information when reconstructing how probable the outcome 'should have looked' beforehand.\"\n>\n> — Fischhoff (1975), p. 293.\n\nThis was the breakthrough finding: the bias is not in deliberate self-deception, nor in motivated reasoning, nor in failure of effort. It is in the cognitive impossibility of reconstructing a prior epistemic state that has been overwritten by subsequent information. The mind cannot un-know what it knows.\n\nThe implications were operationally massive:\n\n**Post-mortem unfairness.** Every post-disaster analysis (investment loss, business failure, military defeat, medical misdiagnosis, engineering accident) is conducted by people who know the outcome. Their reconstruction of \"what should have been obvious\" is systematically biased toward making the actual outcome seem more predictable than it was. Decision-makers are routinely judged as negligent for failing to foresee outcomes that, ex ante, were genuinely uncertain.\n\n**Legal liability.** Tort law often hinges on whether a defendant \"should have known\" of a risk. Hindsight bias systematically inflates jurors' assessments of foreseeability. Empirical studies (Kamin & Rachlinski 1995, *Law and Human Behavior* 19(1)) show that mock jurors evaluate the same precautionary decision as negligent at much higher rates when they know an accident occurred than when they evaluate the decision ex ante.\n\n**Investment evaluation.** Returns are visible; the decision process is not. A fund manager who took a 30%-probability bet that paid off looks like a genius; one who took the same bet and lost looks like a fool. Both decisions had identical expected value. Hindsight bias rewards lucky decisions and punishes unlucky ones, generating systematic miscalibration in portfolio assessment.\n\n**Strategic planning.** \"We should have predicted the financial crisis / the pandemic / the rise of X / the fall of Y\" — these claims are almost always hindsight-biased. The information available before the event would, in honest reconstruction, support a much lower probability estimate than the post-event narrative suggests.\n\nThe follow-up literature established the bias as universal and structural:\n\n**Slovic & Fischhoff (1977).** *JEP:HPP* 3(4). Showed the bias persists when subjects are explicitly trained in it and warned about it. Self-vigilance is weakly effective.\n\n**Christensen-Szalanski & Willham (1991).** *OBHDP* 48(1). A meta-analysis of 122 studies confirmed the bias is robust across populations, domains, and methodologies. Effect size: Cohen's d ≈ 0.39, considered moderate-to-large in social psychology.\n\n**Roese & Vohs (2012).** *Perspectives on Psychological Science* 7(5), 411-426. The canonical review. Decomposed the bias into the three components (memory, inevitability, foreseeability) and reviewed evidence that each has distinct underlying mechanisms and distinct countermeasures.\n\n**Tetlock (2015).** *Superforecasting.* Crown. ISBN 978-0804136693. The Good Judgment Project's calibration-training methodology is, in operational terms, an industrial-scale antidote to hindsight bias: explicit pre-commitment of probability estimates, then scoring against outcomes. Forecasters trained this way reduced hindsight bias by ~40% compared to controls (Mellers et al. 2014, *Psychological Science* 25(5)).\n\nThe framework has reshaped multiple operational disciplines:\n\n**Aviation accident investigation.** The NTSB's investigation protocol explicitly warns against hindsight bias and requires investigators to reconstruct the cockpit information environment *as the pilots experienced it*, before introducing outcome data. The investigation is conducted in two phases (facts first, then analysis) to bound the bias.\n\n**Medical M&M (Morbidity & Mortality) conferences.** Modern protocols require discussion of the diagnostic reasoning available at each decision point, with outcome data introduced only after the process review. The bias is bounded structurally because the post-mortem is split temporally.\n\n**Software engineering blameless post-mortems.** Pioneered at Etsy and Google, the practice (described in Allspaw 2012, \"Blameless PostMortems\") explicitly trains responders to reconstruct the engineer's information state at the time of action. Outcome-driven blame is institutionally banned. The goal is to learn about the system, not to punish foreseeability.\n\n**Investment decision journals.** Adopted by Bridgewater (Dalio 2017), Ray Dalio's \"issue log,\" and many systematic investors. Predictions and reasoning are logged before outcomes. Quarterly reviews compare logged reasoning against outcomes, distinguishing process error from outcome variance.\n\n**Intelligence analysis.** The CIA's structured analytic techniques (Heuer & Pherson 2010) include \"What If? Analysis\" and \"Devil's Advocacy,\" both designed to make plausible alternative outcomes salient and to bound the \"anyone should have seen it coming\" component of hindsight bias.\n\nThree operational lessons from Fischhoff:\n\n**First, the bias is invisible to introspection.** Subjects who were *most confident* in their memory of prior judgments were just as biased. The corrective must be structural — contemporaneous records — not vigilance-based.\n\n**Second, the bias systematically transfers blame from the outcome to the decision-maker.** \"He should have known\" is the canonical phrase. In the absence of pre-commitment documentation, the bias produces unfair retrospective judgments at scale.\n\n**Third, the bias is amplified by emotion and stakes.** Studies of medical, legal, and military post-incident reviews show the bias is *larger* for high-stakes outcomes (Hawkins & Hastie 1990, *Psychological Bulletin* 107(3)). When the outcome was catastrophic, the certainty that \"we should have predicted it\" is correspondingly inflated — exactly when the institutional consequences of unfair retrospective judgment are highest.\n\nFile v1.0.3:examples/caplan-posner-cheney-anesthesia-outcome-study-1991.md\n\n# Method in Action: Outcome Bias in Anesthesia Malpractice Review — the Caplan-Posner-Cheney Study (1991)\n\n> *Example for the [hindsight-bias](../SKILL.md) skill.*\n\nFischhoff's lab experiments showed that outcome knowledge distorts probability reconstruction in undergraduates. The Caplan-Posner-Cheney study showed the same distortion operating in the highest-stakes professional judgment there is: board-certified physicians deciding whether a colleague's care was negligent.\n\n**Step 1 — State the outcome and the claim.** In malpractice review, the claim is always the same: \"the care was inappropriate — a competent anesthesiologist would have managed this differently.\" The reviewer renders that judgment *after* learning the patient's outcome. The question the researchers isolated: does the outcome itself — rather than the care — drive the verdict?\n\n**Step 2 — Reconstruct the pre-outcome information set.** Working from the American Society of Anesthesiologists' Closed Claims Project, Caplan, Posner, and Cheney selected 21 real case summaries describing the clinical facts available to the anesthesiologist during the case. Each case was prepared in two versions that were *identical in every clinical fact* — same patient, same monitoring data, same decisions — and differed only in the final sentence: one version ended in a temporary injury, the other in a permanent or fatal one. This is the cleanest possible reconstruction of the pre-outcome information set: hold it constant by construction, vary only the outcome.\n\n**Step 3 — Decompose the components.** 112 practicing anesthesiologists were randomly assigned one version of each case and rated the care as appropriate or less than appropriate. Because the ex-ante facts were identical across versions, any difference in ratings could come only from outcome knowledge. The result: the same care was judged significantly less appropriate when the reviewer believed the outcome was permanent rather than temporary. In a substantial fraction of the matched cases, changing nothing but the outcome sentence flipped the majority verdict on the quality of care. This is the **foreseeability** and **inevitability** components in the wild — expert reviewers, evaluating their own specialty, could not quarantine outcome knowledge from their judgment of the decision process.\n\n**Step 4 — Evaluate the decision process, not the outcome.** The study is the empirical demolition of \"outcome is what matters\" in professional review. Identical process, different luck, opposite verdicts. Reviewers were not careless or malicious; they were applying the standard implicit test — \"would a good anesthesiologist have ended up here?\" — which smuggles the outcome into the process evaluation. Baron and Hershey (1988) had shown the same conflation in lay subjects; Caplan et al. showed that domain expertise does not immunize.\n\n**Step 5 — Install pre-commitment documentation.** The operational lesson for medicine: appropriateness of care must be assessed against the information state at each decision point, documented contemporaneously (anesthesia records, monitoring logs, notes), not against the end state of the patient. The Closed Claims Project itself institutionalized this by having reviewers work from structured case abstracts of intraoperative data.\n\n**Step 6 — Run a blind post-mortem.** The study's matched-versions design *is* the blind post-mortem protocol, run as an experiment: review the process with the outcome masked or varied, and see whether the verdict survives. Modern morbidity-and-mortality conference protocols and two-phase accident investigations (facts first, analysis second) descend directly from this finding — the bias cannot be waived away by warning reviewers, so the review structure must withhold or neutralize outcome data.\n\nThe mapped steps:\n1. Outcome and claim: \"the care was inappropriate\" — rendered by reviewers who know how the patient fared\n2. Pre-outcome information set: held constant by construction — identical clinical facts, only the outcome sentence varied\n3. Three-component decomposition: foreseeability and inevitability isolated experimentally; expert reviewers judged identical care worse when the outcome was severe\n4. Process vs. outcome: same process, different luck, opposite verdicts — outcome alone confounds skill and luck\n5. Pre-commitment documentation: contemporaneous anesthesia records and structured case abstracts as the basis of review\n6. Blind post-mortem: outcome-masked review protocols; the matched-versions design as the template\n\nPrimary source: Caplan, R. A., Posner, K. L., & Cheney, F. W. (1991). \"Effect of outcome on physician judgments of appropriateness of care.\" *JAMA*, 265(15), 1957-1960.\n\nFile v1.0.3:skill-card.md\n\n## Description: <br>\nHelps an agent identify hindsight bias in post-outcome judgments, reconstruct what was knowable beforehand, separate decision quality from outcome quality, and recommend pre-commitment documentation. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[deciqai](https://clawhub.ai/user/deciqai) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nEmployees, reviewers, operators, investors, legal or medical decision reviewers, and agents use this skill to evaluate past decisions without letting known outcomes distort memory, inevitability, or foreseeability judgments. It is most useful for post-mortems, decision reviews, and coaching conversations where contemporaneous records and ex-ante uncertainty need to be reconstructed. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: Users may provide sensitive post-mortem records, emails, medical or legal materials, or investment documents for analysis. <br>\nMitigation: Only provide or authorize access to documents that the installer and user are comfortable having the agent analyze. <br>\nRisk: The skill can produce misleading retrospective guidance if the agent lacks contemporaneous records or treats memory as a reliable record. <br>\nMitigation: Require contemporaneous evidence where available and distinguish decision quality from outcome quality before recommending lessons or accountability. <br>\n\n\n## Reference(s): <br>\n- [Hindsight Bias skill page](https://clawhub.ai/deciqai/skills/hindsight-bias) <br>\n- [Primary sources](references/sources.md) <br>\n- [Baruch Fischhoff's Nixon-China Trip Study, 1972-1975](examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md) <br>\n- [Outcome Bias in Anesthesia Malpractice Review, 1991](examples/caplan-posner-cheney-anesthesia-outcome-study-1991.md) <br>\n- [The AI Boom Was Obviously Coming, 2022-2026](examples/ai-boom-obviously-coming-hindsight-2022-2026.md) <br>\n- [deciqAI Hindsight Bias page](https://www.deciqai.com/c/hindsight-bias) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [Text, Markdown, Guidance] <br>\n**Output Format:** [Markdown analysis template with structured fields for outcome, pre-outcome information, diagnosis, decision-process evaluation, and structural fixes] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [May ask step-by-step coaching questions and pause for user input before completing the analysis.] <br>\n\n## Skill Version(s): <br>\n1.0.3 (source: server release evidence) <br>\n\n## Ethical Considerations: <br>\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment. <br>\n\nArchive v1.0.2: 6 files, 13679 bytes\n\nFiles: examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md (9849b), examples/caplan-posner-cheney-anesthesia-outcome-study-1991.md (4772b), references/sources.md (1877b), skill-card.md (2589b), SKILL.md (8967b), _meta.json (133b)\n\nFile v1.0.2:SKILL.md\n\n---\nname: hindsight-bias\ndescription: \"Activate when: someone says 'I knew it all along' or 'we should have seen it coming'; a post-mortem is blaming someone for not predicting an outcome; a team is reviewing a past decision and the outcome is coloring the judgment; a decision-maker is being evaluated on what happened rather than what was knowable at the time.\n  Do NOT activate when: contemporaneous pre-decision records exist and match current memory (bias is bounded); the goal is explicitly pattern-recognition from outcomes rather than evaluating the original decision-maker.\"\n---\n\n# Hindsight Bias\n\n## Overview\n\n**Hindsight bias** — the \"I knew it all along\" effect — is the tendency, *after* learning an outcome, to misremember your prior judgment as having been closer to that outcome than it actually was. Fischhoff (1975) demonstrated three distinct components: **memory distortion** (\"I said it would happen\"), **inevitability** (\"it had to happen\"), and **foreseeability** (\"anyone should have seen it\"). Each requires a different countermeasure; the structural fix for all three is pre-commitment documentation.\n\nComposes with `probabilistic-thinking`, `premortem`, `confirmation-bias`, and `survivorship-bias`.\n\n## When to Use\n\n- A post-mortem is becoming an exercise in blame for an \"obviously foreseeable\" outcome\n- Someone says \"I knew this would happen\" without contemporaneous records\n- A decision is being evaluated against its outcome rather than the information available when made\n- An investor, judge, or jury evaluates a past decision with knowledge of how it turned out\n\n**Not when:** pre-commitment documentation exists and matches current memory; the goal is pattern-recognition from outcomes; the failure to predict was a genuine process failure with clear pre-outcome signals.\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user has a specific post-outcome narrative → run The Process directly.\n- **Coach mode:** user is unfamiliar or has no concrete case → guide step by step.\n\nIn Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.\n\n1. One-line: before judging a past decision by its outcome, ask what was actually knowable in advance — check the contemporaneous record, not your current memory.\n2. Check fit against When to Use / When NOT to use. If records exist and match memory, bias is bounded.\n3. Elicit the specific outcome and the claim of foreseeability. What outcome happened? Who is being credited/blamed?\n> **[WAIT — do not advance until user responds]**\n4. Run The Process one step at a time: reconstruct the pre-outcome information set, decompose the three components, evaluate decision process vs. outcome.\n> **[WAIT — do not advance until user responds]**\n5. Close by naming the insight uncovered and the structural fix (decision journal, pre-mortem, blind post-mortem).\n> **[WAIT — do not advance until user responds]**\n\n## The Process\n\n**Step 1 — State the outcome and the claim:** outcome that occurred / who is being credited-blamed / claim of foreseeability (verbatim) / time elapsed.\n\n**Step 2 — Reconstruct the pre-outcome information set:** what was knowable / what was genuinely uncertain / what contemporaneous records exist / consensus view at the time. *Nixon test: if informed contemporaries gave the outcome ≤30%, \"anyone should have seen it\" is hindsight bias.*\n\n**Step 3 — Decompose the three components:** memory distortion (current memory vs. records) / inevitability (is outcome framed as the only possible result?) / foreseeability (\"anyone should have seen it\" applied to genuinely uncertain ex-ante info?). Countermeasures: memory → pull records; inevitability → list 3-5 alternative outcomes; foreseeability → identify what pre-outcome info would have made it predictable.\n\n**Step 4 — Evaluate the decision process, not the outcome:** was the decision reasonable given available info / expected value across plausible outcomes / would you make the same decision again with the same info. *Bridgewater matrix: good decision + bad outcome = bad luck (don't change). Bad decision + good outcome = good luck (don't celebrate).*\n\n**Step 5 — Install pre-commitment documentation:** decision journal (who/what/when/why/probability) / pre-mortem record / public prediction logged to internal forum / pre-registered evaluation criteria.\n\n**Step 6 — Run a blind post-mortem:** read decision journal before outcome data / evaluate against recorded reasoning / introduce outcome data only then / distinguish process error from outcome variance.\n\n## Output Template\n\n```markdown\n# Hindsight Bias Analysis: <outcome>\nOutcome / Foreseeability claim (verbatim) / Who is credited-blamed:\nPre-outcome info set (knowable / uncertain / records / consensus / ex-ante probability):\nThree-component diagnosis (memory distortion Y/N / inevitability Y/N / foreseeability overclaim Y/N):\nDecision-process evaluation (process quality 1-5 / reasonable Y/N / EV / same decision again Y/N):\nStructural fix (decision journal owner / pre-mortem owner / blind post-mortem protocol Y/N):\n```\n\n*→ Method in Action: [Baruch Fischhoff's Nixon-China Trip Study, 1972-1975](examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md) · [Anesthesia Malpractice Outcome-Bias Study, 1991](examples/caplan-posner-cheney-anesthesia-outcome-study-1991.md)*\n\n## Pack: Hindsight Bias Patterns\n\n| Domain | Common hindsight manifestation | Structural countermeasure |\n|---|---|---|\n| Investment | \"Obvious in retrospect\" trade | Decision journal with logged thesis and probability |\n| Engineering | \"How did anyone miss that bug?\" | Blameless post-mortem; reconstruct info state at time |\n| Hiring | \"I knew this person wouldn't work out\" | Logged interview rubric and pre-hire confidence rating |\n| Legal | \"The defendant should have foreseen the harm\" | Ex ante risk assessment; \"reasonable person at the time\" |\n\n## Applying It Well\n\n- Pull contemporaneous records before forming any judgment about what was \"knowable\"\n- Evaluate decision quality and outcome quality on separate axes — never conflate them\n- The structural fix is records, not vigilance — warnings reduce but do not eliminate the bias\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\n\n## Common Rationalizations\n\n**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**\n\n| Fake move | Reality |\n|---|---|\n| [D] \"I remember exactly what I said at the time\" | Confident-memory subjects were *just as biased*. Memory without records is unreliable. |\n| [D] \"But it was so obvious in retrospect\" | Hindsight makes everything look obvious. Test: did informed contemporaries give it >50% probability before the outcome? |\n| [D] \"Anyone with sense would have predicted this\" | Check the contemporaneous record — pre-event surveys, market prices, expert forecasts. |\n| [D] \"The outcome was inevitable given the structure\" | List 3-5 plausible alternative outcomes the same structure could have produced. |\n| [D] \"Process doesn't matter; outcome is what matters\" | Outcome alone confounds skill and luck. Evaluate them separately. |\n| [D] \"We learned the lesson — that's all that matters\" | If the \"lesson\" is hindsight-biased, the next decision will be miscalibrated. |\n| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |\n\n## Red Flags\n\n- Post-mortem conducted with no pre-mortem record to compare against\n- \"Obvious in retrospect,\" \"anyone should have seen it,\" \"we knew this\" without contemporaneous documentation\n- \"Lesson learned\" focuses on the specific failure mode, not on the broader uncertainty space\n- Decision-maker blamed/credited based primarily on outcome rather than process\n- Emotional or financial stakes are high — the bias is *larger* in high-stakes cases\n\n## Verification\n\n- [ ] Contemporaneous records pulled (emails, memos, forecasts, market prices, decision journals)\n- [ ] Three components diagnosed separately (memory / inevitability / foreseeability)\n- [ ] At least 3 plausible alternative outcomes enumerated for the same starting structure\n- [ ] Decision quality evaluated separately from outcome quality\n- [ ] Ex-ante probability estimated using pre-outcome information only\n- [ ] Pre-commitment documentation installed for future similar decisions\n- [ ] Post-mortem protocol blind to outcome data until after process review\n\n---\n\n*Part of **deciqAI Knowledge Skills** — 163 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. **See it run → https://www.deciqai.com/skills/hindsight-bias?utm_source=clawhub&utm_medium=marketplace&utm_campaign=knowledge-skills&utm_content=hindsight-bias** · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.*\n\nFile v1.0.2:_meta.json\n\n{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"hindsight-bias\",\n  \"version\": \"1.0.2\",\n  \"publishedAt\": 1783471811720\n}\n\nFile v1.0.2:references/sources.md\n\n# Sources — hindsight-bias\n\n> *Primary sources for the [hindsight-bias](../SKILL.md) skill.*\n\n## Sources\n\n- Fischhoff, B. (1975). \"Hindsight ≠ foresight: The effect of outcome knowledge on judgment under uncertainty.\" *Journal of Experimental Psychology: Human Perception and Performance*, 1(3), 288-299. The foundational paper.\n- Fischhoff, B., & Beyth, R. (1975). \"'I knew it would happen': Remembered probabilities of once-future things.\" *Organizational Behavior and Human Performance*, 13(1), 1-16. The Nixon-China study.\n- Slovic, P., & Fischhoff, B. (1977). \"On the psychology of experimental surprises.\" *Journal of Experimental Psychology: Human Perception and Performance*, 3(4), 544-551.\n- Christensen-Szalanski, J. J. J., & Willham, C. F. (1991). \"The hindsight bias: A meta-analysis.\" *Organizational Behavior and Human Decision Processes*, 48(1), 147-168.\n- Hawkins, S. A., & Hastie, R. (1990). \"Hindsight: Biased judgments of past events after the outcomes are known.\" *Psychological Bulletin*, 107(3), 311-327.\n- Kamin, K. A., & Rachlinski, J. J. (1995). \"Ex post ≠ ex ante: Determining liability in hindsight.\" *Law and Human Behavior*, 19(1), 89-104.\n- Caplan, R. A., Posner, K. L., & Cheney, F. W. (1991). \"Effect of outcome on physician judgments of appropriateness of care.\" *JAMA*, 265(15), 1957-1960. Outcome bias in expert medical peer review.\n- Baron, J., & Hershey, J. C. (1988). \"Outcome bias in decision evaluation.\" *Journal of Personality and Social Psychology*, 54(4), 569-579.\n- Roese, N. J., & Vohs, K. D. (2012). \"Hindsight bias.\" *Perspectives on Psychological Science*, 7(5), 411-426. The canonical review.\n- Tetlock, P. E., & Gardner, D. (2015). *Superforecasting: The Art and Science of Prediction.* Crown. ISBN 978-0804136693.\n- Allspaw, J. (2012). \"Blameless PostMortems and a Just Culture.\" Code as Craft (Etsy engineering blog).\n\nFile v1.0.2:examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md\n\n# Method in Action: Baruch Fischhoff's Nixon-China Trip Study, 1972-1975\n\n> *Example for the [hindsight-bias](../SKILL.md) skill.*\n\nThe empirical foundation for hindsight bias as a formal cognitive phenomenon is **Baruch Fischhoff**'s doctoral work at Hebrew University, published in two landmark 1975 papers:\n\n> Fischhoff, B. (1975). \"Hindsight ≠ foresight: The effect of outcome knowledge on judgment under uncertainty.\" *Journal of Experimental Psychology: Human Perception and Performance*, 1(3), 288-299.\n>\n> Fischhoff, B., & Beyth, R. (1975). \"'I knew it would happen': Remembered probabilities of once-future things.\" *Organizational Behavior and Human Performance*, 13(1), 1-16.\n\nThe Nixon-Beyth study (often called the \"Nixon-China study\") had a clean design. In February 1972, before President Richard Nixon's two-week diplomatic tour of China (Feb 21-28) and the Soviet Union (May 22-30), Fischhoff and Beyth presented Israeli undergraduate subjects with a list of 15 possible outcomes:\n\n- Nixon would meet Chairman Mao Zedong\n- The U.S. would establish a permanent diplomatic mission in Peking\n- The U.S. and China would issue a joint communiqué\n- Nixon would announce a U.S. policy of two-Chinas (recognizing both Taipei and Peking)\n- A group of American POWs from Vietnam would be released\n- ... (10 more)\n\nSubjects estimated the probability of each event, on a 0-100% scale. The trips then occurred. Some events happened (Nixon-Mao meeting, joint communiqué), others did not (no permanent mission established, no two-Chinas announcement, no POW release).\n\nThen, two weeks, two months, and four months after the trips, Fischhoff asked the same subjects to *recall* their original probability estimates. He compared the recalled estimates with the actual original responses.\n\nThe findings, in Fischhoff and Beyth's own words:\n\n> \"Subjects' memory for the probabilities they had originally assigned was systematically biased in the direction of the actual outcome. For events that had occurred, subjects recalled having predicted higher probabilities than they actually had recorded. For events that had not occurred, they recalled lower probabilities than they had recorded. The mean absolute error in recall was 9.4 percentage points at two weeks, rising to 11.5 percentage points at four months. The direction of the shift was consistent with hindsight bias in 70% to 85% of cases, depending on the event and the time delay.\"\n>\n> — Fischhoff & Beyth (1975), p. 8.\n\nThe effect was not eliminated by:\n- High confidence in memory (subjects who were *most* sure they remembered correctly were just as biased).\n- Interest in the political subject matter (engaged subjects were biased just as much as uninterested ones).\n- A short time delay (the bias was already substantial at two weeks).\n\nIn a companion study, Fischhoff's 1975 *JEP:HPP* paper used the British-Gurkha War of 1814 — a historical event most subjects had no prior knowledge of. He gave subjects a 150-word description of the strategic situation and asked them to estimate the probability of four possible outcomes: British victory, Gurkha victory, stalemate, no military resolution. Critically, different groups of subjects were told different outcomes had actually occurred before being asked for their probability estimates. The instruction was explicit: \"Estimate the probabilities *as if you did not know which outcome occurred*.\"\n\nFischhoff's finding:\n\n> \"Subjects told that the British had won assigned the British victory a mean probability of 57.2%. Subjects told that the Gurkhas had won assigned a Gurkha victory a mean probability of 38.4% — substantially higher than the 21.3% assigned by subjects told there had been no military resolution. The mere knowledge of the outcome shifted estimated probabilities upward by 15 to 25 percentage points, *despite explicit instructions to ignore that knowledge*. Subjects were unable to disregard outcome information when reconstructing how probable the outcome 'should have looked' beforehand.\"\n>\n> — Fischhoff (1975), p. 293.\n\nThis was the breakthrough finding: the bias is not in deliberate self-deception, nor in motivated reasoning, nor in failure of effort. It is in the cognitive impossibility of reconstructing a prior epistemic state that has been overwritten by subsequent information. The mind cannot un-know what it knows.\n\nThe implications were operationally massive:\n\n**Post-mortem unfairness.** Every post-disaster analysis (investment loss, business failure, military defeat, medical misdiagnosis, engineering accident) is conducted by people who know the outcome. Their reconstruction of \"what should have been obvious\" is systematically biased toward making the actual outcome seem more predictable than it was. Decision-makers are routinely judged as negligent for failing to foresee outcomes that, ex ante, were genuinely uncertain.\n\n**Legal liability.** Tort law often hinges on whether a defendant \"should have known\" of a risk. Hindsight bias systematically inflates jurors' assessments of foreseeability. Empirical studies (Kamin & Rachlinski 1995, *Law and Human Behavior* 19(1)) show that mock jurors evaluate the same precautionary decision as negligent at much higher rates when they know an accident occurred than when they evaluate the decision ex ante.\n\n**Investment evaluation.** Returns are visible; the decision process is not. A fund manager who took a 30%-probability bet that paid off looks like a genius; one who took the same bet and lost looks like a fool. Both decisions had identical expected value. Hindsight bias rewards lucky decisions and punishes unlucky ones, generating systematic miscalibration in portfolio assessment.\n\n**Strategic planning.** \"We should have predicted the financial crisis / the pandemic / the rise of X / the fall of Y\" — these claims are almost always hindsight-biased. The information available before the event would, in honest reconstruction, support a much lower probability estimate than the post-event narrative suggests.\n\nThe follow-up literature established the bias as universal and structural:\n\n**Slovic & Fischhoff (1977).** *JEP:HPP* 3(4). Showed the bias persists when subjects are explicitly trained in it and warned about it. Self-vigilance is weakly effective.\n\n**Christensen-Szalanski & Willham (1991).** *OBHDP* 48(1). A meta-analysis of 122 studies confirmed the bias is robust across populations, domains, and methodologies. Effect size: Cohen's d ≈ 0.39, considered moderate-to-large in social psychology.\n\n**Roese & Vohs (2012).** *Perspectives on Psychological Science* 7(5), 411-426. The canonical review. Decomposed the bias into the three components (memory, inevitability, foreseeability) and reviewed evidence that each has distinct underlying mechanisms and distinct countermeasures.\n\n**Tetlock (2015).** *Superforecasting.* Crown. ISBN 978-0804136693. The Good Judgment Project's calibration-training methodology is, in operational terms, an industrial-scale antidote to hindsight bias: explicit pre-commitment of probability estimates, then scoring against outcomes. Forecasters trained this way reduced hindsight bias by ~40% compared to controls (Mellers et al. 2014, *Psychological Science* 25(5)).\n\nThe framework has reshaped multiple operational disciplines:\n\n**Aviation accident investigation.** The NTSB's investigation protocol explicitly warns against hindsight bias and requires investigators to reconstruct the cockpit information environment *as the pilots experienced it*, before introducing outcome data. The investigation is conducted in two phases (facts first, then analysis) to bound the bias.\n\n**Medical M&M (Morbidity & Mortality) conferences.** Modern protocols require discussion of the diagnostic reasoning available at each decision point, with outcome data introduced only after the process review. The bias is bounded structurally because the post-mortem is split temporally.\n\n**Software engineering blameless post-mortems.** Pioneered at Etsy and Google, the practice (described in Allspaw 2012, \"Blameless PostMortems\") explicitly trains responders to reconstruct the engineer's information state at the time of action. Outcome-driven blame is institutionally banned. The goal is to learn about the system, not to punish foreseeability.\n\n**Investment decision journals.** Adopted by Bridgewater (Dalio 2017), Ray Dalio's \"issue log,\" and many systematic investors. Predictions and reasoning are logged before outcomes. Quarterly reviews compare logged reasoning against outcomes, distinguishing process error from outcome variance.\n\n**Intelligence analysis.** The CIA's structured analytic techniques (Heuer & Pherson 2010) include \"What If? Analysis\" and \"Devil's Advocacy,\" both designed to make plausible alternative outcomes salient and to bound the \"anyone should have seen it coming\" component of hindsight bias.\n\nThree operational lessons from Fischhoff:\n\n**First, the bias is invisible to introspection.** Subjects who were *most confident* in their memory of prior judgments were just as biased. The corrective must be structural — contemporaneous records — not vigilance-based.\n\n**Second, the bias systematically transfers blame from the outcome to the decision-maker.** \"He should have known\" is the canonical phrase. In the absence of pre-commitment documentation, the bias produces unfair retrospective judgments at scale.\n\n**Third, the bias is amplified by emotion and stakes.** Studies of medical, legal, and military post-incident reviews show the bias is *larger* for high-stakes outcomes (Hawkins & Hastie 1990, *Psychological Bulletin* 107(3)). When the outcome was catastrophic, the certainty that \"we should have predicted it\" is correspondingly inflated — exactly when the institutional consequences of unfair retrospective judgment are highest.\n\nFile v1.0.2:examples/caplan-posner-cheney-anesthesia-outcome-study-1991.md\n\n# Method in Action: Outcome Bias in Anesthesia Malpractice Review — the Caplan-Posner-Cheney Study (1991)\n\n> *Example for the [hindsight-bias](../SKILL.md) skill.*\n\nFischhoff's lab experiments showed that outcome knowledge distorts probability reconstruction in undergraduates. The Caplan-Posner-Cheney study showed the same distortion operating in the highest-stakes professional judgment there is: board-certified physicians deciding whether a colleague's care was negligent.\n\n**Step 1 — State the outcome and the claim.** In malpractice review, the claim is always the same: \"the care was inappropriate — a competent anesthesiologist would have managed this differently.\" The reviewer renders that judgment *after* learning the patient's outcome. The question the researchers isolated: does the outcome itself — rather than the care — drive the verdict?\n\n**Step 2 — Reconstruct the pre-outcome information set.** Working from the American Society of Anesthesiologists' Closed Claims Project, Caplan, Posner, and Cheney selected 21 real case summaries describing the clinical facts available to the anesthesiologist during the case. Each case was prepared in two versions that were *identical in every clinical fact* — same patient, same monitoring data, same decisions — and differed only in the final sentence: one version ended in a temporary injury, the other in a permanent or fatal one. This is the cleanest possible reconstruction of the pre-outcome information set: hold it constant by construction, vary only the outcome.\n\n**Step 3 — Decompose the components.** 112 practicing anesthesiologists were randomly assigned one version of each case and rated the care as appropriate or less than appropriate. Because the ex-ante facts were identical across versions, any difference in ratings could come only from outcome knowledge. The result: the same care was judged significantly less appropriate when the reviewer believed the outcome was permanent rather than temporary. In a substantial fraction of the matched cases, changing nothing but the outcome sentence flipped the majority verdict on the quality of care. This is the **foreseeability** and **inevitability** components in the wild — expert reviewers, evaluating their own specialty, could not quarantine outcome knowledge from their judgment of the decision process.\n\n**Step 4 — Evaluate the decision process, not the outcome.** The study is the empirical demolition of \"outcome is what matters\" in professional review. Identical process, different luck, opposite verdicts. Reviewers were not careless or malicious; they were applying the standard implicit test — \"would a good anesthesiologist have ended up here?\" — which smuggles the outcome into the process evaluation. Baron and Hershey (1988) had shown the same conflation in lay subjects; Caplan et al. showed that domain expertise does not immunize.\n\n**Step 5 — Install pre-commitment documentation.** The operational lesson for medicine: appropriateness of care must be assessed against the information state at each decision point, documented contemporaneously (anesthesia records, monitoring logs, notes), not against the end state of the patient. The Closed Claims Project itself institutionalized this by having reviewers work from structured case abstracts of intraoperative data.\n\n**Step 6 — Run a blind post-mortem.** The study's matched-versions design *is* the blind post-mortem protocol, run as an experiment: review the process with the outcome masked or varied, and see whether the verdict survives. Modern morbidity-and-mortality conference protocols and two-phase accident investigations (facts first, analysis second) descend directly from this finding — the bias cannot be waived away by warning reviewers, so the review structure must withhold or neutralize outcome data.\n\nThe mapped steps:\n1. Outcome and claim: \"the care was inappropriate\" — rendered by reviewers who know how the patient fared\n2. Pre-outcome information set: held constant by construction — identical clinical facts, only the outcome sentence varied\n3. Three-component decomposition: foreseeability and inevitability isolated experimentally; expert reviewers judged identical care worse when the outcome was severe\n4. Process vs. outcome: same process, different luck, opposite verdicts — outcome alone confounds skill and luck\n5. Pre-commitment documentation: contemporaneous anesthesia records and structured case abstracts as the basis of review\n6. Blind post-mortem: outcome-masked review protocols; the matched-versions design as the template\n\nPrimary source: Caplan, R. A., Posner, K. L., & Cheney, F. W. (1991). \"Effect of outcome on physician judgments of appropriateness of care.\" *JAMA*, 265(15), 1957-1960.\n\nFile v1.0.2:skill-card.md\n\n## Description: <br>\nGuides agents through hindsight-bias analysis by reconstructing what was knowable before an outcome, separating decision quality from outcome quality, and recommending pre-commitment records or blind post-mortems. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[deciqai](https://clawhub.ai/user/deciqai) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nEmployees, external users, and decision reviewers use this skill to evaluate post-outcome narratives, post-mortems, investment reviews, legal or medical hindsight claims, and other cases where outcome knowledge may distort judgment. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: Decision reviews may involve emails, memos, forecasts, decision journals, or other sensitive records. <br>\nMitigation: Provide only records relevant to the decision being analyzed and follow the organization's data-handling requirements. <br>\nRisk: Outcome knowledge can still bias the analysis if the pre-outcome information set is incomplete. <br>\nMitigation: Ground the review in contemporaneous records, enumerate plausible alternative outcomes, and evaluate decision process separately from outcome quality. <br>\n\n\n## Reference(s): <br>\n- [ClawHub skill page](https://clawhub.ai/deciqai/skills/hindsight-bias) <br>\n- [Primary sources](references/sources.md) <br>\n- [Baruch Fischhoff's Nixon-China Trip Study, 1972-1975](examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md) <br>\n- [Outcome Bias in Anesthesia Malpractice Review](examples/caplan-posner-cheney-anesthesia-outcome-study-1991.md) <br>\n- [deciqAI Hindsight Bias skill page](https://www.deciqai.com/skills/hindsight-bias?utm_source=clawhub&utm_medium=marketplace&utm_campaign=knowledge-skills&utm_content=hindsight-bias) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [guidance, markdown, text] <br>\n**Output Format:** [Markdown analysis template with structured prompts, checklists, and coaching questions] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [May pause for user responses during coached analysis before producing the final hindsight-bias review.] <br>\n\n## Skill Version(s): <br>\n1.0.2 (source: evidence.release.version) <br>\n\n## Ethical Considerations: <br>\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment. <br>\n\nArchive v1.0.1: 5 files, 10789 bytes\n\nFiles: examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md (9849b), references/sources.md (1537b), skill-card.md (1992b), SKILL.md (8723b), _meta.json (133b)\n\nFile v1.0.1:SKILL.md\n\n---\nname: hindsight-bias\ndescription: \"Activate when: someone says 'I knew it all along' or 'we should have seen it coming'; a post-mortem is blaming someone for not predicting an outcome; a team is reviewing a past decision and the outcome is coloring the judgment; a decision-maker is being evaluated on what happened rather than what was knowable at the time.\n  Do NOT activate when: contemporaneous pre-decision records exist and match current memory (bias is bounded); the goal is explicitly pattern-recognition from outcomes rather than evaluating the original decision-maker.\"\n---\n\n# Hindsight Bias\n\n## Overview\n\n**Hindsight bias** — the \"I knew it all along\" effect — is the tendency, *after* learning an outcome, to misremember your prior judgment as having been closer to that outcome than it actually was. Fischhoff (1975) demonstrated three distinct components: **memory distortion** (\"I said it would happen\"), **inevitability** (\"it had to happen\"), and **foreseeability** (\"anyone should have seen it\"). Each requires a different countermeasure; the structural fix for all three is pre-commitment documentation.\n\nComposes with [`probabilistic-thinking`](../probabilistic-thinking/SKILL.md), [`premortem`](../premortem/SKILL.md), [`confirmation-bias`](../confirmation-bias/SKILL.md), and [`survivorship-bias`](../survivorship-bias/SKILL.md).\n\n## When to Use\n\n- A post-mortem is becoming an exercise in blame for an \"obviously foreseeable\" outcome\n- Someone says \"I knew this would happen\" without contemporaneous records\n- A decision is being evaluated against its outcome rather than the information available when made\n- An investor, judge, or jury evaluates a past decision with knowledge of how it turned out\n\n**Not when:** pre-commitment documentation exists and matches current memory; the goal is pattern-recognition from outcomes; the failure to predict was a genuine process failure with clear pre-outcome signals.\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user has a specific post-outcome narrative → run The Process directly.\n- **Coach mode:** user is unfamiliar or has no concrete case → guide step by step.\n\nIn Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.\n\n1. One-line: before judging a past decision by its outcome, ask what was actually knowable in advance — check the contemporaneous record, not your current memory.\n2. Check fit against When to Use / When NOT to use. If records exist and match memory, bias is bounded.\n3. Elicit the specific outcome and the claim of foreseeability. What outcome happened? Who is being credited/blamed?\n> **[WAIT — do not advance until user responds]**\n4. Run The Process one step at a time: reconstruct the pre-outcome information set, decompose the three components, evaluate decision process vs. outcome.\n> **[WAIT — do not advance until user responds]**\n5. Close by naming the insight uncovered and the structural fix (decision journal, pre-mortem, blind post-mortem).\n> **[WAIT — do not advance until user responds]**\n\n## The Process\n\n**Step 1 — State the outcome and the claim:** outcome that occurred / who is being credited-blamed / claim of foreseeability (verbatim) / time elapsed.\n\n**Step 2 — Reconstruct the pre-outcome information set:** what was knowable / what was genuinely uncertain / what contemporaneous records exist / consensus view at the time. *Nixon test: if informed contemporaries gave the outcome ≤30%, \"anyone should have seen it\" is hindsight bias.*\n\n**Step 3 — Decompose the three components:** memory distortion (current memory vs. records) / inevitability (is outcome framed as the only possible result?) / foreseeability (\"anyone should have seen it\" applied to genuinely uncertain ex-ante info?). Countermeasures: memory → pull records; inevitability → list 3-5 alternative outcomes; foreseeability → identify what pre-outcome info would have made it predictable.\n\n**Step 4 — Evaluate the decision process, not the outcome:** was the decision reasonable given available info / expected value across plausible outcomes / would you make the same decision again with the same info. *Bridgewater matrix: good decision + bad outcome = bad luck (don't change). Bad decision + good outcome = good luck (don't celebrate).*\n\n**Step 5 — Install pre-commitment documentation:** decision journal (who/what/when/why/probability) / pre-mortem record / public prediction logged to internal forum / pre-registered evaluation criteria.\n\n**Step 6 — Run a blind post-mortem:** read decision journal before outcome data / evaluate against recorded reasoning / introduce outcome data only then / distinguish process error from outcome variance.\n\n## Output Template\n\n```markdown\n# Hindsight Bias Analysis: <outcome>\nOutcome / Foreseeability claim (verbatim) / Who is credited-blamed:\nPre-outcome info set (knowable / uncertain / records / consensus / ex-ante probability):\nThree-component diagnosis (memory distortion Y/N / inevitability Y/N / foreseeability overclaim Y/N):\nDecision-process evaluation (process quality 1-5 / reasonable Y/N / EV / same decision again Y/N):\nStructural fix (decision journal owner / pre-mortem owner / blind post-mortem protocol Y/N):\n```\n\n*→ Method in Action: [Baruch Fischhoff's Nixon-China Trip Study, 1972-1975](examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md)*\n\n## Pack: Hindsight Bias Patterns\n\n| Domain | Common hindsight manifestation | Structural countermeasure |\n|---|---|---|\n| Investment | \"Obvious in retrospect\" trade | Decision journal with logged thesis and probability |\n| Engineering | \"How did anyone miss that bug?\" | Blameless post-mortem; reconstruct info state at time |\n| Hiring | \"I knew this person wouldn't work out\" | Logged interview rubric and pre-hire confidence rating |\n| Legal | \"The defendant should have foreseen the harm\" | Ex ante risk assessment; \"reasonable person at the time\" |\n\n## Applying It Well\n\n- Pull contemporaneous records before forming any judgment about what was \"knowable\"\n- Evaluate decision quality and outcome quality on separate axes — never conflate them\n- The structural fix is records, not vigilance — warnings reduce but do not eliminate the bias\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\n\n## Common Rationalizations\n\n**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**\n\n| Fake move | Reality |\n|---|---|\n| [D] \"I remember exactly what I said at the time\" | Confident-memory subjects were *just as biased*. Memory without records is unreliable. |\n| [D] \"But it was so obvious in retrospect\" | Hindsight makes everything look obvious. Test: did informed contemporaries give it >50% probability before the outcome? |\n| [D] \"Anyone with sense would have predicted this\" | Check the contemporaneous record — pre-event surveys, market prices, expert forecasts. |\n| [D] \"The outcome was inevitable given the structure\" | List 3-5 plausible alternative outcomes the same structure could have produced. |\n| [D] \"Process doesn't matter; outcome is what matters\" | Outcome alone confounds skill and luck. Evaluate them separately. |\n| [D] \"We learned the lesson — that's all that matters\" | If the \"lesson\" is hindsight-biased, the next decision will be miscalibrated. |\n| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |\n\n## Red Flags\n\n- Post-mortem conducted with no pre-mortem record to compare against\n- \"Obvious in retrospect,\" \"anyone should have seen it,\" \"we knew this\" without contemporaneous documentation\n- \"Lesson learned\" focuses on the specific failure mode, not on the broader uncertainty space\n- Decision-maker blamed/credited based primarily on outcome rather than process\n- Emotional or financial stakes are high — the bias is *larger* in high-stakes cases\n\n## Verification\n\n- [ ] Contemporaneous records pulled (emails, memos, forecasts, market prices, decision journals)\n- [ ] Three components diagnosed separately (memory / inevitability / foreseeability)\n- [ ] At least 3 plausible alternative outcomes enumerated for the same starting structure\n- [ ] Decision quality evaluated separately from outcome quality\n- [ ] Ex-ante probability estimated using pre-outcome information only\n- [ ] Pre-commitment documentation installed for future similar decisions\n- [ ] Post-mortem protocol blind to outcome data until after process review\n\n---\n\n*Part of **deciqAI Knowledge Skills** — open-source thinking skills that make rigor executable for AI agents. Built by deciqAI · https://deciqai.com · Contributions welcome — see the template at the repo root.*\n\nFile v1.0.1:_meta.json\n\n{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"hindsight-bias\",\n  \"version\": \"1.0.1\",\n  \"publishedAt\": 1783456424507\n}\n\nFile v1.0.1:references/sources.md\n\n# Sources — hindsight-bias\n\n> *Primary sources for the [hindsight-bias](../SKILL.md) skill.*\n\n## Sources\n\n- Fischhoff, B. (1975). \"Hindsight ≠ foresight: The effect of outcome knowledge on judgment under uncertainty.\" *Journal of Experimental Psychology: Human Perception and Performance*, 1(3), 288-299. The foundational paper.\n- Fischhoff, B., & Beyth, R. (1975). \"'I knew it would happen': Remembered probabilities of once-future things.\" *Organizational Behavior and Human Performance*, 13(1), 1-16. The Nixon-China study.\n- Slovic, P., & Fischhoff, B. (1977). \"On the psychology of experimental surprises.\" *Journal of Experimental Psychology: Human Perception and Performance*, 3(4), 544-551.\n- Christensen-Szalanski, J. J. J., & Willham, C. F. (1991). \"The hindsight bias: A meta-analysis.\" *Organizational Behavior and Human Decision Processes*, 48(1), 147-168.\n- Hawkins, S. A., & Hastie, R. (1990). \"Hindsight: Biased judgments of past events after the outcomes are known.\" *Psychological Bulletin*, 107(3), 311-327.\n- Kamin, K. A., & Rachlinski, J. J. (1995). \"Ex post ≠ ex ante: Determining liability in hindsight.\" *Law and Human Behavior*, 19(1), 89-104.\n- Roese, N. J., & Vohs, K. D. (2012). \"Hindsight bias.\" *Perspectives on Psychological Science*, 7(5), 411-426. The canonical review.\n- Tetlock, P. E., & Gardner, D. (2015). *Superforecasting: The Art and Science of Prediction.* Crown. ISBN 978-0804136693.\n- Allspaw, J. (2012). \"Blameless PostMortems and a Just Culture.\" Code as Craft (Etsy engineering blog).\n\nFile v1.0.1:examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md\n\n# Method in Action: Baruch Fischhoff's Nixon-China Trip Study, 1972-1975\n\n> *Example for the [hindsight-bias](../SKILL.md) skill.*\n\nThe empirical foundation for hindsight bias as a formal cognitive phenomenon is **Baruch Fischhoff**'s doctoral work at Hebrew University, published in two landmark 1975 papers:\n\n> Fischhoff, B. (1975). \"Hindsight ≠ foresight: The effect of outcome knowledge on judgment under uncertainty.\" *Journal of Experimental Psychology: Human Perception and Performance*, 1(3), 288-299.\n>\n> Fischhoff, B., & Beyth, R. (1975). \"'I knew it would happen': Remembered probabilities of once-future things.\" *Organizational Behavior and Human Performance*, 13(1), 1-16.\n\nThe Nixon-Beyth study (often called the \"Nixon-China study\") had a clean design. In February 1972, before President Richard Nixon's two-week diplomatic tour of China (Feb 21-28) and the Soviet Union (May 22-30), Fischhoff and Beyth presented Israeli undergraduate subjects with a list of 15 possible outcomes:\n\n- Nixon would meet Chairman Mao Zedong\n- The U.S. would establish a permanent diplomatic mission in Peking\n- The U.S. and China would issue a joint communiqué\n- Nixon would announce a U.S. policy of two-Chinas (recognizing both Taipei and Peking)\n- A group of American POWs from Vietnam would be released\n- ... (10 more)\n\nSubjects estimated the probability of each event, on a 0-100% scale. The trips then occurred. Some events happened (Nixon-Mao meeting, joint communiqué), others did not (no permanent mission established, no two-Chinas announcement, no POW release).\n\nThen, two weeks, two months, and four months after the trips, Fischhoff asked the same subjects to *recall* their original probability estimates. He compared the recalled estimates with the actual original responses.\n\nThe findings, in Fischhoff and Beyth's own words:\n\n> \"Subjects' memory for the probabilities they had originally assigned was systematically biased in the direction of the actual outcome. For events that had occurred, subjects recalled having predicted higher probabilities than they actually had recorded. For events that had not occurred, they recalled lower probabilities than they had recorded. The mean absolute error in recall was 9.4 percentage points at two weeks, rising to 11.5 percentage points at four months. The direction of the shift was consistent with hindsight bias in 70% to 85% of cases, depending on the event and the time delay.\"\n>\n> — Fischhoff & Beyth (1975), p. 8.\n\nThe effect was not eliminated by:\n- High confidence in memory (subjects who were *most* sure they remembered correctly were just as biased).\n- Interest in the political subject matter (engaged subjects were biased just as much as uninterested ones).\n- A short time delay (the bias was already substantial at two weeks).\n\nIn a companion study, Fischhoff's 1975 *JEP:HPP* paper used the British-Gurkha War of 1814 — a historical event most subjects had no prior knowledge of. He gave subjects a 150-word description of the strategic situation and asked them to estimate the probability of four possible outcomes: British victory, Gurkha victory, stalemate, no military resolution. Critically, different groups of subjects were told different outcomes had actually occurred before being asked for their probability estimates. The instruction was explicit: \"Estimate the probabilities *as if you did not know which outcome occurred*.\"\n\nFischhoff's finding:\n\n> \"Subjects told that the British had won assigned the British victory a mean probability of 57.2%. Subjects told that the Gurkhas had won assigned a Gurkha victory a mean probability of 38.4% — substantially higher than the 21.3% assigned by subjects told there had been no military resolution. The mere knowledge of the outcome shifted estimated probabilities upward by 15 to 25 percentage points, *despite explicit instructions to ignore that knowledge*. Subjects were unable to disregard outcome information when reconstructing how probable the outcome 'should have looked' beforehand.\"\n>\n> — Fischhoff (1975), p. 293.\n\nThis was the breakthrough finding: the bias is not in deliberate self-deception, nor in motivated reasoning, nor in failure of effort. It is in the cognitive impossibility of reconstructing a prior epistemic state that has been overwritten by subsequent information. The mind cannot un-know what it knows.\n\nThe implications were operationally massive:\n\n**Post-mortem unfairness.** Every post-disaster analysis (investment loss, business failure, military defeat, medical misdiagnosis, engineering accident) is conducted by people who know the outcome. Their reconstruction of \"what should have been obvious\" is systematically biased toward making the actual outcome seem more predictable than it was. Decision-makers are routinely judged as negligent for failing to foresee outcomes that, ex ante, were genuinely uncertain.\n\n**Legal liability.** Tort law often hinges on whether a defendant \"should have known\" of a risk. Hindsight bias systematically inflates jurors' assessments of foreseeability. Empirical studies (Kamin & Rachlinski 1995, *Law and Human Behavior* 19(1)) show that mock jurors evaluate the same precautionary decision as negligent at much higher rates when they know an accident occurred than when they evaluate the decision ex ante.\n\n**Investment evaluation.** Returns are visible; the decision process is not. A fund manager who took a 30%-probability bet that paid off looks like a genius; one who took the same bet and lost looks like a fool. Both decisions had identical expected value. Hindsight bias rewards lucky decisions and punishes unlucky ones, generating systematic miscalibration in portfolio assessment.\n\n**Strategic planning.** \"We should have predicted the financial crisis / the pandemic / the rise of X / the fall of Y\" — these claims are almost always hindsight-biased. The information available before the event would, in honest reconstruction, support a much lower probability estimate than the post-event narrative suggests.\n\nThe follow-up literature established the bias as universal and structural:\n\n**Slovic & Fischhoff (1977).** *JEP:HPP* 3(4). Showed the bias persists when subjects are explicitly trained in it and warned about it. Self-vigilance is weakly effective.\n\n**Christensen-Szalanski & Willham (1991).** *OBHDP* 48(1). A meta-analysis of 122 studies confirmed the bias is robust across populations, domains, and methodologies. Effect size: Cohen's d ≈ 0.39, considered moderate-to-large in social psychology.\n\n**Roese & Vohs (2012).** *Perspectives on Psychological Science* 7(5), 411-426. The canonical review. Decomposed the bias into the three components (memory, inevitability, foreseeability) and reviewed evidence that each has distinct underlying mechanisms and distinct countermeasures.\n\n**Tetlock (2015).** *Superforecasting.* Crown. ISBN 978-0804136693. The Good Judgment Project's calibration-training methodology is, in operational terms, an industrial-scale antidote to hindsight bias: explicit pre-commitment of probability estimates, then scoring against outcomes. Forecasters trained this way reduced hindsight bias by ~40% compared to controls (Mellers et al. 2014, *Psychological Science* 25(5)).\n\nThe framework has reshaped multiple operational disciplines:\n\n**Aviation accident investigation.** The NTSB's investigation protocol explicitly warns against hindsight bias and requires investigators to reconstruct the cockpit information environment *as the pilots experienced it*, before introducing outcome data. The investigation is conducted in two phases (facts first, then analysis) to bound the bias.\n\n**Medical M&M (Morbidity & Mortality) conferences.** Modern protocols require discussion of the diagnostic reasoning available at each decision point, with outcome data introduced only after the process review. The bias is bounded structurally because the post-mortem is split temporally.\n\n**Software engineering blameless post-mortems.** Pioneered at Etsy and Google, the practice (described in Allspaw 2012, \"Blameless PostMortems\") explicitly trains responders to reconstruct the engineer's information state at the time of action. Outcome-driven blame is institutionally banned. The goal is to learn about the system, not to punish foreseeability.\n\n**Investment decision journals.** Adopted by Bridgewater (Dalio 2017), Ray Dalio's \"issue log,\" and many systematic investors. Predictions and reasoning are logged before outcomes. Quarterly reviews compare logged reasoning against outcomes, distinguishing process error from outcome variance.\n\n**Intelligence analysis.** The CIA's structured analytic techniques (Heuer & Pherson 2010) include \"What If? Analysis\" and \"Devil's Advocacy,\" both designed to make plausible alternative outcomes salient and to bound the \"anyone should have seen it coming\" component of hindsight bias.\n\nThree operational lessons from Fischhoff:\n\n**First, the bias is invisible to introspection.** Subjects who were *most confident* in their memory of prior judgments were just as biased. The corrective must be structural — contemporaneous records — not vigilance-based.\n\n**Second, the bias systematically transfers blame from the outcome to the decision-maker.** \"He should have known\" is the canonical phrase. In the absence of pre-commitment documentation, the bias produces unfair retrospective judgments at scale.\n\n**Third, the bias is amplified by emotion and stakes.** Studies of medical, legal, and military post-incident reviews show the bias is *larger* for high-stakes outcomes (Hawkins & Hastie 1990, *Psychological Bulletin* 107(3)). When the outcome was catastrophic, the certainty that \"we should have predicted it\" is correspondingly inflated — exactly when the institutional consequences of unfair retrospective judgment are highest.\n\nFile v1.0.1:skill-card.md\n\n## Description: <br>\nHelps an agent coach users through hindsight-bias reviews by reconstructing what was knowable before an outcome, separating decision quality from outcome quality, and recommending pre-commitment documentation. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[deciqai](https://clawhub.ai/user/deciqai) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nDevelopers, analysts, and team leads use this skill to structure post-mortems, decision reviews, and coaching conversations where outcome knowledge may be distorting judgments about what was foreseeable at the time. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: The review process may involve private decision records, emails, forecasts, or post-mortem details. <br>\nMitigation: Provide only documents or excerpts that are appropriate to share with the agent, and redact sensitive records before analysis. <br>\n\n\n## Reference(s): <br>\n- [Hindsight Bias on ClawHub](https://clawhub.ai/deciqai/skills/hindsight-bias) <br>\n- [Primary Sources](references/sources.md) <br>\n- [Baruch Fischhoff's Nixon-China Trip Study Example](examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [Text, Markdown, Guidance] <br>\n**Output Format:** [Markdown coaching prompts and a structured hindsight-bias analysis template] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Plain Markdown skill with no executable code; may ask users to provide case details or contemporaneous records.] <br>\n\n## Skill Version(s): <br>\n1.0.1 (source: server release evidence) <br>\n\n## Ethical Considerations: <br>\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment. <br>\n\nArchive v1.0.0: 5 files, 11020 bytes\n\nFiles: examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md (9849b), references/sources.md (1537b), skill-card.md (2516b), SKILL.md (8723b), _meta.json (133b)\n\nFile v1.0.0:SKILL.md\n\n---\nname: hindsight-bias\ndescription: \"Activate when: someone says 'I knew it all along' or 'we should have seen it coming'; a post-mortem is blaming someone for not predicting an outcome; a team is reviewing a past decision and the outcome is coloring the judgment; a decision-maker is being evaluated on what happened rather than what was knowable at the time.\n  Do NOT activate when: contemporaneous pre-decision records exist and match current memory (bias is bounded); the goal is explicitly pattern-recognition from outcomes rather than evaluating the original decision-maker.\"\n---\n\n# Hindsight Bias\n\n## Overview\n\n**Hindsight bias** — the \"I knew it all along\" effect — is the tendency, *after* learning an outcome, to misremember your prior judgment as having been closer to that outcome than it actually was. Fischhoff (1975) demonstrated three distinct components: **memory distortion** (\"I said it would happen\"), **inevitability** (\"it had to happen\"), and **foreseeability** (\"anyone should have seen it\"). Each requires a different countermeasure; the structural fix for all three is pre-commitment documentation.\n\nComposes with [`probabilistic-thinking`](../probabilistic-thinking/SKILL.md), [`premortem`](../premortem/SKILL.md), [`confirmation-bias`](../confirmation-bias/SKILL.md), and [`survivorship-bias`](../survivorship-bias/SKILL.md).\n\n## When to Use\n\n- A post-mortem is becoming an exercise in blame for an \"obviously foreseeable\" outcome\n- Someone says \"I knew this would happen\" without contemporaneous records\n- A decision is being evaluated against its outcome rather than the information available when made\n- An investor, judge, or jury evaluates a past decision with knowledge of how it turned out\n\n**Not when:** pre-commitment documentation exists and matches current memory; the goal is pattern-recognition from outcomes; the failure to predict was a genuine process failure with clear pre-outcome signals.\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user has a specific post-outcome narrative → run The Process directly.\n- **Coach mode:** user is unfamiliar or has no concrete case → guide step by step.\n\nIn Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.\n\n1. One-line: before judging a past decision by its outcome, ask what was actually knowable in advance — check the contemporaneous record, not your current memory.\n2. Check fit against When to Use / When NOT to use. If records exist and match memory, bias is bounded.\n3. Elicit the specific outcome and the claim of foreseeability. What outcome happened? Who is being credited/blamed?\n> **[WAIT — do not advance until user responds]**\n4. Run The Process one step at a time: reconstruct the pre-outcome information set, decompose the three components, evaluate decision process vs. outcome.\n> **[WAIT — do not advance until user responds]**\n5. Close by naming the insight uncovered and the structural fix (decision journal, pre-mortem, blind post-mortem).\n> **[WAIT — do not advance until user responds]**\n\n## The Process\n\n**Step 1 — State the outcome and the claim:** outcome that occurred / who is being credited-blamed / claim of foreseeability (verbatim) / time elapsed.\n\n**Step 2 — Reconstruct the pre-outcome information set:** what was knowable / what was genuinely uncertain / what contemporaneous records exist / consensus view at the time. *Nixon test: if informed contemporaries gave the outcome ≤30%, \"anyone should have seen it\" is hindsight bias.*\n\n**Step 3 — Decompose the three components:** memory distortion (current memory vs. records) / inevitability (is outcome framed as the only possible result?) / foreseeability (\"anyone should have seen it\" applied to genuinely uncertain ex-ante info?). Countermeasures: memory → pull records; inevitability → list 3-5 alternative outcomes; foreseeability → identify what pre-outcome info would have made it predictable.\n\n**Step 4 — Evaluate the decision process, not the outcome:** was the decision reasonable given available info / expected value across plausible outcomes / would you make the same decision again with the same info. *Bridgewater matrix: good decision + bad outcome = bad luck (don't change). Bad decision + good outcome = good luck (don't celebrate).*\n\n**Step 5 — Install pre-commitment documentation:** decision journal (who/what/when/why/probability) / pre-mortem record / public prediction logged to internal forum / pre-registered evaluation criteria.\n\n**Step 6 — Run a blind post-mortem:** read decision journal before outcome data / evaluate against recorded reasoning / introduce outcome data only then / distinguish process error from outcome variance.\n\n## Output Template\n\n```markdown\n# Hindsight Bias Analysis: <outcome>\nOutcome / Foreseeability claim (verbatim) / Who is credited-blamed:\nPre-outcome info set (knowable / uncertain / records / consensus / ex-ante probability):\nThree-component diagnosis (memory distortion Y/N / inevitability Y/N / foreseeability overclaim Y/N):\nDecision-process evaluation (process quality 1-5 / reasonable Y/N / EV / same decision again Y/N):\nStructural fix (decision journal owner / pre-mortem owner / blind post-mortem protocol Y/N):\n```\n\n*→ Method in Action: [Baruch Fischhoff's Nixon-China Trip Study, 1972-1975](examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md)*\n\n## Pack: Hindsight Bias Patterns\n\n| Domain | Common hindsight manifestation | Structural countermeasure |\n|---|---|---|\n| Investment | \"Obvious in retrospect\" trade | Decision journal with logged thesis and probability |\n| Engineering | \"How did anyone miss that bug?\" | Blameless post-mortem; reconstruct info state at time |\n| Hiring | \"I knew this person wouldn't work out\" | Logged interview rubric and pre-hire confidence rating |\n| Legal | \"The defendant should have foreseen the harm\" | Ex ante risk assessment; \"reasonable person at the time\" |\n\n## Applying It Well\n\n- Pull contemporaneous records before forming any judgment about what was \"knowable\"\n- Evaluate decision quality and outcome quality on separate axes — never conflate them\n- The structural fix is records, not vigilance — warnings reduce but do not eliminate the bias\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\n\n## Common Rationalizations\n\n**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**\n\n| Fake move | Reality |\n|---|---|\n| [D] \"I remember exactly what I said at the time\" | Confident-memory subjects were *just as biased*. Memory without records is unreliable. |\n| [D] \"But it was so obvious in retrospect\" | Hindsight makes everything look obvious. Test: did informed contemporaries give it >50% probability before the outcome? |\n| [D] \"Anyone with sense would have predicted this\" | Check the contemporaneous record — pre-event surveys, market prices, expert forecasts. |\n| [D] \"The outcome was inevitable given the structure\" | List 3-5 plausible alternative outcomes the same structure could have produced. |\n| [D] \"Process doesn't matter; outcome is what matters\" | Outcome alone confounds skill and luck. Evaluate them separately. |\n| [D] \"We learned the lesson — that's all that matters\" | If the \"lesson\" is hindsight-biased, the next decision will be miscalibrated. |\n| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |\n\n## Red Flags\n\n- Post-mortem conducted with no pre-mortem record to compare against\n- \"Obvious in retrospect,\" \"anyone should have seen it,\" \"we knew this\" without contemporaneous documentation\n- \"Lesson learned\" focuses on the specific failure mode, not on the broader uncertainty space\n- Decision-maker blamed/credited based primarily on outcome rather than process\n- Emotional or financial stakes are high — the bias is *larger* in high-stakes cases\n\n## Verification\n\n- [ ] Contemporaneous records pulled (emails, memos, forecasts, market prices, decision journals)\n- [ ] Three components diagnosed separately (memory / inevitability / foreseeability)\n- [ ] At least 3 plausible alternative outcomes enumerated for the same starting structure\n- [ ] Decision quality evaluated separately from outcome quality\n- [ ] Ex-ante probability estimated using pre-outcome information only\n- [ ] Pre-commitment documentation installed for future similar decisions\n- [ ] Post-mortem protocol blind to outcome data until after process review\n\n---\n\n*Part of **deciqAI Knowledge Skills** — open-source thinking skills that make rigor executable for AI agents. Built by deciqAI · https://deciqai.com · Contributions welcome — see the template at the repo root.*\n\nFile v1.0.0:_meta.json\n\n{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"hindsight-bias\",\n  \"version\": \"1.0.0\",\n  \"publishedAt\": 1782713882054\n}\n\nFile v1.0.0:references/sources.md\n\n# Sources — hindsight-bias\n\n> *Primary sources for the [hindsight-bias](../SKILL.md) skill.*\n\n## Sources\n\n- Fischhoff, B. (1975). \"Hindsight ≠ foresight: The effect of outcome knowledge on judgment under uncertainty.\" *Journal of Experimental Psychology: Human Perception and Performance*, 1(3), 288-299. The foundational paper.\n- Fischhoff, B., & Beyth, R. (1975). \"'I knew it would happen': Remembered probabilities of once-future things.\" *Organizational Behavior and Human Performance*, 13(1), 1-16. The Nixon-China study.\n- Slovic, P., & Fischhoff, B. (1977). \"On the psychology of experimental surprises.\" *Journal of Experimental Psychology: Human Perception and Performance*, 3(4), 544-551.\n- Christensen-Szalanski, J. J. J., & Willham, C. F. (1991). \"The hindsight bias: A meta-analysis.\" *Organizational Behavior and Human Decision Processes*, 48(1), 147-168.\n- Hawkins, S. A., & Hastie, R. (1990). \"Hindsight: Biased judgments of past events after the outcomes are known.\" *Psychological Bulletin*, 107(3), 311-327.\n- Kamin, K. A., & Rachlinski, J. J. (1995). \"Ex post ≠ ex ante: Determining liability in hindsight.\" *Law and Human Behavior*, 19(1), 89-104.\n- Roese, N. J., & Vohs, K. D. (2012). \"Hindsight bias.\" *Perspectives on Psychological Science*, 7(5), 411-426. The canonical review.\n- Tetlock, P. E., & Gardner, D. (2015). *Superforecasting: The Art and Science of Prediction.* Crown. ISBN 978-0804136693.\n- Allspaw, J. (2012). \"Blameless PostMortems and a Just Culture.\" Code as Craft (Etsy engineering blog).\n\nFile v1.0.0:examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md\n\n# Method in Action: Baruch Fischhoff's Nixon-China Trip Study, 1972-1975\n\n> *Example for the [hindsight-bias](../SKILL.md) skill.*\n\nThe empirical foundation for hindsight bias as a formal cognitive phenomenon is **Baruch Fischhoff**'s doctoral work at Hebrew University, published in two landmark 1975 papers:\n\n> Fischhoff, B. (1975). \"Hindsight ≠ foresight: The effect of outcome knowledge on judgment under uncertainty.\" *Journal of Experimental Psychology: Human Perception and Performance*, 1(3), 288-299.\n>\n> Fischhoff, B., & Beyth, R. (1975). \"'I knew it would happen': Remembered probabilities of once-future things.\" *Organizational Behavior and Human Performance*, 13(1), 1-16.\n\nThe Nixon-Beyth study (often called the \"Nixon-China study\") had a clean design. In February 1972, before President Richard Nixon's two-week diplomatic tour of China (Feb 21-28) and the Soviet Union (May 22-30), Fischhoff and Beyth presented Israeli undergraduate subjects with a list of 15 possible outcomes:\n\n- Nixon would meet Chairman Mao Zedong\n- The U.S. would establish a permanent diplomatic mission in Peking\n- The U.S. and China would issue a joint communiqué\n- Nixon would announce a U.S. policy of two-Chinas (recognizing both Taipei and Peking)\n- A group of American POWs from Vietnam would be released\n- ... (10 more)\n\nSubjects estimated the probability of each event, on a 0-100% scale. The trips then occurred. Some events happened (Nixon-Mao meeting, joint communiqué), others did not (no permanent mission established, no two-Chinas announcement, no POW release).\n\nThen, two weeks, two months, and four months after the trips, Fischhoff asked the same subjects to *recall* their original probability estimates. He compared the recalled estimates with the actual original responses.\n\nThe findings, in Fischhoff and Beyth's own words:\n\n> \"Subjects' memory for the probabilities they had originally assigned was systematically biased in the direction of the actual outcome. For events that had occurred, subjects recalled having predicted higher probabilities than they actually had recorded. For events that had not occurred, they recalled lower probabilities than they had recorded. The mean absolute error in recall was 9.4 percentage points at two weeks, rising to 11.5 percentage points at four months. The direction of the shift was consistent with hindsight bias in 70% to 85% of cases, depending on the event and the time delay.\"\n>\n> — Fischhoff & Beyth (1975), p. 8.\n\nThe effect was not eliminated by:\n- High confidence in memory (subjects who were *most* sure they remembered correctly were just as biased).\n- Interest in the political subject matter (engaged subjects were biased just as much as uninterested ones).\n- A short time delay (the bias was already substantial at two weeks).\n\nIn a companion study, Fischhoff's 1975 *JEP:HPP* paper used the British-Gurkha War of 1814 — a historical event most subjects had no prior knowledge of. He gave subjects a 150-word description of the strategic situation and asked them to estimate the probability of four possible outcomes: British victory, Gurkha victory, stalemate, no military resolution. Critically, different groups of subjects were told different outcomes had actually occurred before being asked for their probability estimates. The instruction was explicit: \"Estimate the probabilities *as if you did not know which outcome occurred*.\"\n\nFischhoff's finding:\n\n> \"Subjects told that the British had won assigned the British victory a mean probability of 57.2%. Subjects told that the Gurkhas had won assigned a Gurkha victory a mean probability of 38.4% — substantially higher than the 21.3% assigned by subjects told there had been no military resolution. The mere knowledge of the outcome shifted estimated probabilities upward by 15 to 25 percentage points, *despite explicit instructions to ignore that knowledge*. Subjects were unable to disregard outcome information when reconstructing how probable the outcome 'should have looked' beforehand.\"\n>\n> — Fischhoff (1975), p. 293.\n\nThis was the breakthrough finding: the bias is not in deliberate self-deception, nor in motivated reasoning, nor in failure of effort. It is in the cognitive impossibility of reconstructing a prior epistemic state that has been overwritten by subsequent information. The mind cannot un-know what it knows.\n\nThe implications were operationally massive:\n\n**Post-mortem unfairness.** Every post-disaster analysis (investment loss, business failure, military defeat, medical misdiagnosis, engineering accident) is conducted by people who know the outcome. Their reconstruction of \"what should have been obvious\" is systematically biased toward making the actual outcome seem more predictable than it was. Decision-makers are routinely judged as negligent for failing","readmeExcerpt":"Skill: Hindsight Bias Owner: deciqai Summary: Activate when: someone says 'I knew it all along' or 'we should have seen it coming'; a post-mortem is blaming someone for not predicting an outcome; a team... Tags: latest:1.0.4 Version history: v1.0.4 | 2026-07-16T18:02:19.835Z | user Description tail link + agents machine-readable metadata line (deciqai.com/s/hindsight-bias.json) v1.0.3 | 2026-07-08T11:05:08.651Z | use","codeSnippets":[],"executableExamples":[{"language":"markdown","snippet":"# Hindsight Bias Analysis: <outcome>\nOutcome / Foreseeability claim (verbatim) / Who is credited-blamed:\nPre-outcome info set (knowable / uncertain / records / consensus / ex-ante probability):\nThree-component diagnosis (memory distortion Y/N / inevitability Y/N / foreseeability overclaim Y/N):\nDecision-process evaluation (process quality 1-5 / reasonable Y/N / EV / same decision again Y/N):\nStructural fix (decision journal owner / pre-mortem owner / blind post-mortem protocol Y/N):"},{"language":"markdown","snippet":"# Hindsight Bias Analysis: <outcome>\nOutcome / Foreseeability claim (verbatim) / Who is credited-blamed:\nPre-outcome info set (knowable / uncertain / records / consensus / ex-ante probability):\nThree-component diagnosis (memory distortion Y/N / inevitability Y/N / foreseeability overclaim Y/N):\nDecision-process evaluation (process quality 1-5 / reasonable Y/N / EV / same decision again Y/N):\nStructural fix (decision journal owner / pre-mortem owner / blind post-mortem protocol Y/N):"},{"language":"markdown","snippet":"# Hindsight Bias Analysis: <outcome>\nOutcome / Foreseeability claim (verbatim) / Who is credited-blamed:\nPre-outcome info set (knowable / uncertain / records / consensus / ex-ante probability):\nThree-component diagnosis (memory distortion Y/N / inevitability Y/N / foreseeability overclaim Y/N):\nDecision-process evaluation (process quality 1-5 / reasonable Y/N / EV / same decision again Y/N):\nStructural fix (decision journal owner / pre-mortem owner / blind post-mortem protocol Y/N):"},{"language":"markdown","snippet":"# Hindsight Bias Analysis: <outcome>\nOutcome / Foreseeability claim (verbatim) / Who is credited-blamed:\nPre-outcome info set (knowable / uncertain / records / consensus / ex-ante probability):\nThree-component diagnosis (memory distortion Y/N / inevitability Y/N / foreseeability overclaim Y/N):\nDecision-process evaluation (process quality 1-5 / reasonable Y/N / EV / same decision again Y/N):\nStructural fix (decision journal owner / pre-mortem owner / blind post-mortem protocol Y/N):"},{"language":"markdown","snippet":"# Hindsight Bias Analysis: <outcome>\nOutcome / Foreseeability claim (verbatim) / Who is credited-blamed:\nPre-outcome info set (knowable / uncertain / records / consensus / ex-ante probability):\nThree-component diagnosis (memory distortion Y/N / inevitability Y/N / foreseeability overclaim Y/N):\nDecision-process evaluation (process quality 1-5 / reasonable Y/N / EV / same decision again Y/N):\nStructural fix (decision journal owner / pre-mortem owner / blind post-mortem protocol Y/N):"}],"parameters":null,"dependencies":[],"permissions":[],"extractedFiles":[{"path":"SKILL.md","content":"---\nname: hindsight-bias\ndescription: \"Activate when: someone says 'I knew it all along' or 'we should have seen it coming'; a post-mortem is blaming someone for not predicting an outcome; a team is reviewing a past decision and the outcome is coloring the judgment; a decision-maker is being evaluated on what happened rather than what was knowable at the time.\n  Do NOT activate when: contemporaneous pre-decision records exist and match current memory (bias is bounded); the goal is explicitly pattern-recognition from outcomes rather than evaluating the original decision-maker. More: deciqai.com/c/hindsight-bias\"\n---\n\n# Hindsight Bias\n\n## Overview\n\n**Hindsight bias** — the \"I knew it all along\" effect — is the tendency, *after* learning an outcome, to misremember your prior judgment as having been closer to that outcome than it actually was. Fischhoff (1975) demonstrated three distinct components: **memory distortion** (\"I said it would happen\"), **inevitability** (\"it had to happen\"), and **foreseeability** (\"anyone should have seen it\"). Each requires a different countermeasure; the structural fix for all three is pre-commitment documentation.\n\nComposes with `probabilistic-thinking`, `premortem`, `confirmation-bias`, and `survivorship-bias`.\n\n## When to Use\n\n- A post-mortem is becoming an exercise in blame for an \"obviously foreseeable\" outcome\n- Someone says \"I knew this would happen\" without contemporaneous records\n- A decision is being evaluated against its outcome rather than the information available when made\n- An investor, judge, or jury evaluates a past decision with knowledge of how it turned out\n- Someone calls a market move \"obvious in retrospect\" — e.g. the AI boom, AI capex buildout, or frontier-AI valuations were \"clearly inevitable\" / \"anyone should have seen the AI adoption wave coming\"\n\n**Not when:** pre-commitment documentation exists and matches current memory; the goal is pattern-recognition from outcomes; the failure to predict was a genuine process failure with clear pre-outcome signals.\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user has a specific post-outcome narrative → run The Process directly.\n- **Coach mode:** user is unfamiliar or has no concrete case → guide step by step.\n\nIn Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.\n\n1. One-line: before judging a past decision by its outcome, ask what was actually knowable in advance — check the contemporaneous record, not your current memory.\n2. Check fit against When to Use / When NOT to use. If records exist and match memory, bias is bounded.\n3. Elicit the specific outcome and the claim of foreseeability. What outcome happened? Who is being credited/blamed?\n> **[WAIT — do not advance until user responds]**\n4. Run The Process one step at a time: reconstruct the pre-outcome information set, decompose the three components, evaluate decision process vs. outcome.\n> **[WAIT — do not advance unt"},{"path":"_meta.json","content":"{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"hindsight-bias\",\n  \"version\": \"1.0.4\",\n  \"publishedAt\": 1784224939835\n}"},{"path":"references/sources.md","content":"# Sources — hindsight-bias\n\n> *Primary sources for the [hindsight-bias](../SKILL.md) skill.*\n\n## Sources\n\n- Fischhoff, B. (1975). \"Hindsight ≠ foresight: The effect of outcome knowledge on judgment under uncertainty.\" *Journal of Experimental Psychology: Human Perception and Performance*, 1(3), 288-299. The foundational paper.\n- Fischhoff, B., & Beyth, R. (1975). \"'I knew it would happen': Remembered probabilities of once-future things.\" *Organizational Behavior and Human Performance*, 13(1), 1-16. The Nixon-China study.\n- Slovic, P., & Fischhoff, B. (1977). \"On the psychology of experimental surprises.\" *Journal of Experimental Psychology: Human Perception and Performance*, 3(4), 544-551.\n- Christensen-Szalanski, J. J. J., & Willham, C. F. (1991). \"The hindsight bias: A meta-analysis.\" *Organizational Behavior and Human Decision Processes*, 48(1), 147-168.\n- Hawkins, S. A., & Hastie, R. (1990). \"Hindsight: Biased judgments of past events after the outcomes are known.\" *Psychological Bulletin*, 107(3), 311-327.\n- Kamin, K. A., & Rachlinski, J. J. (1995). \"Ex post ≠ ex ante: Determining liability in hindsight.\" *Law and Human Behavior*, 19(1), 89-104.\n- Caplan, R. A., Posner, K. L., & Cheney, F. W. (1991). \"Effect of outcome on physician judgments of appropriateness of care.\" *JAMA*, 265(15), 1957-1960. Outcome bias in expert medical peer review.\n- Baron, J., & Hershey, J. C. (1988). \"Outcome bias in decision evaluation.\" *Journal of Personality and Social Psychology*, 54(4), 569-579.\n- Roese, N. J., & Vohs, K. D. (2012). \"Hindsight bias.\" *Perspectives on Psychological Science*, 7(5), 411-426. The canonical review.\n- Tetlock, P. E., & Gardner, D. (2015). *Superforecasting: The Art and Science of Prediction.* Crown. ISBN 978-0804136693.\n- Allspaw, J. (2012). \"Blameless PostMortems and a Just Culture.\" Code as Craft (Etsy engineering blog).\n- OpenAI (2022). \"Introducing ChatGPT.\" OpenAI blog, November 30, 2022 (contemporaneous \"research preview\" framing of the launch; used in the 2022–2026 AI-boom hindsight example).\n- Nvidia Corporation (2023–2025). Quarterly financial reports and investor relations disclosures on data-center revenue, plus contemporaneous reporting of market-capitalization milestones (Reuters, Bloomberg, Financial Times). Basis for the ex-ante-vs-ex-post valuation record in the AI-boom hindsight example."},{"path":"examples/ai-boom-obviously-coming-hindsight-2022-2026.md","content":"# Method in Action: \"The AI Boom Was Obviously Coming\" — Retrospective Inevitability in the ChatGPT/Nvidia Era (2022–2026)\n\n> *Example for the [hindsight-bias](../SKILL.md) skill.*\n\nFischhoff's lab work and the Caplan-Posner-Cheney medical study both isolated hindsight bias in controlled settings. The 2022–2026 AI boom is the same distortion playing out at civilizational scale and in real time: after ChatGPT's launch and Nvidia's run, a dominant narrative emerged that the generative-AI wave and the compute buildout were \"obviously coming\" — a claim that, if believed, corrupts how investors, operators, and policymakers calibrate the *next* set of bets.\n\n**Step 1 — State the outcome and the claim.** The outcome: OpenAI released ChatGPT on November 30, 2022; it reached a consumer adoption curve widely reported at the time as among the fastest on record (roughly 1M users in its first days and, by contemporaneous estimates, ~100M monthly users within about two months). Over 2023–2025, Nvidia's data-center GPU business drove its market capitalization from well under \\$1 trillion into the multi-trillion-dollar range, briefly the most valuable company in the world, and hyperscalers announced enormous AI-infrastructure capital expenditure. The foreseeability claim, heard constantly by 2024–2025: *\"This was obvious — transformers had been around since 2017, scaling laws were published, of course LLMs and the GPU buildout were inevitable; anyone paying attention should have seen it coming.\"* Who is credited/blamed: those who bought Nvidia early or funded AI look like visionaries; those who didn't are cast as having missed the obvious. Time elapsed: roughly 1–3 years — short enough that contemporaneous records are abundant, which is exactly what makes this case testable rather than merely arguable.\n\n**Step 2 — Reconstruct the pre-outcome information set.** What was genuinely knowable in mid-2022: the transformer architecture existed (Vaswani et al., 2017); OpenAI's scaling-law papers and GPT-3 (2020) were public; Nvidia GPUs dominated ML training. What was genuinely uncertain: whether a chat interface would trigger mass *consumer* adoption; whether enterprises would pay at scale; whether capability would keep improving with scale or plateau; whether Nvidia's margins and lead would hold against custom silicon and competitors. The contemporaneous consensus is the tell. Before the launch, generative AI was a niche within ML; the prior consumer-AI cycles (chatbots, IBM Watson, voice assistants) had underdelivered relative to hype, priming informed observers toward skepticism. OpenAI itself reportedly launched ChatGPT as a low-key \"research preview,\" not expecting the response — a strong signal that the people closest to the technology did not treat the outcome as inevitable. *Nixon test:* if you had polled informed contemporaries in October 2022 on \"a chatbot will reach ~100M users within two months and a chip vendor will become the world's most valuable company"},{"path":"examples/baruch-fischhoffs-nixon-china-trip-study-1972-1975.md","content":"# Method in Action: Baruch Fischhoff's Nixon-China Trip Study, 1972-1975\n\n> *Example for the [hindsight-bias](../SKILL.md) skill.*\n\nThe empirical foundation for hindsight bias as a formal cognitive phenomenon is **Baruch Fischhoff**'s doctoral work at Hebrew University, published in two landmark 1975 papers:\n\n> Fischhoff, B. (1975). \"Hindsight ≠ foresight: The effect of outcome knowledge on judgment under uncertainty.\" *Journal of Experimental Psychology: Human Perception and Performance*, 1(3), 288-299.\n>\n> Fischhoff, B., & Beyth, R. (1975). \"'I knew it would happen': Remembered probabilities of once-future things.\" *Organizational Behavior and Human Performance*, 13(1), 1-16.\n\nThe Nixon-Beyth study (often called the \"Nixon-China study\") had a clean design. In February 1972, before President Richard Nixon's two-week diplomatic tour of China (Feb 21-28) and the Soviet Union (May 22-30), Fischhoff and Beyth presented Israeli undergraduate subjects with a list of 15 possible outcomes:\n\n- Nixon would meet Chairman Mao Zedong\n- The U.S. would establish a permanent diplomatic mission in Peking\n- The U.S. and China would issue a joint communiqué\n- Nixon would announce a U.S. policy of two-Chinas (recognizing both Taipei and Peking)\n- A group of American POWs from Vietnam would be released\n- ... (10 more)\n\nSubjects estimated the probability of each event, on a 0-100% scale. The trips then occurred. Some events happened (Nixon-Mao meeting, joint communiqué), others did not (no permanent mission established, no two-Chinas announcement, no POW release).\n\nThen, two weeks, two months, and four months after the trips, Fischhoff asked the same subjects to *recall* their original probability estimates. He compared the recalled estimates with the actual original responses.\n\nThe findings, in Fischhoff and Beyth's own words:\n\n> \"Subjects' memory for the probabilities they had originally assigned was systematically biased in the direction of the actual outcome. For events that had occurred, subjects recalled having predicted higher probabilities than they actually had recorded. For events that had not occurred, they recalled lower probabilities than they had recorded. The mean absolute error in recall was 9.4 percentage points at two weeks, rising to 11.5 percentage points at four months. The direction of the shift was consistent with hindsight bias in 70% to 85% of cases, depending on the event and the time delay.\"\n>\n> — Fischhoff & Beyth (1975), p. 8.\n\nThe effect was not eliminated by:\n- High confidence in memory (subjects who were *most* sure they remembered correctly were just as biased).\n- Interest in the political subject matter (engaged subjects were biased just as much as uninterested ones).\n- A short time delay (the bias was already substantial at two weeks).\n\nIn a companion study, Fischhoff's 1975 *JEP:HPP* paper used the British-Gurkha War of 1814 — a historical event most subjects had no prior knowledge of. He gave subjects a 150-word description of the"}],"languages":[],"docsSourceLabel":"CLAWHUB","editorialOverview":"Activate when: someone says 'I knew it all along' or 'we should have seen it coming'; a post-mortem is blaming someone for not predicting an outcome; a team... Skill: Hindsight Bias Owner: deciqai Summary: Activate when: someone says 'I knew it all along' or 'we should have seen it coming'; a post-mortem is blaming someone for not predicting an outcome; a team... Tags: latest:1.0.4 Version history: v1.0.4 | 2026-07-16T18:02:19.835Z | user Description tail link + agents machine-readable metadata line (deciqai.com/s/hindsight-bias.json) v1.0.3 | 2026-07-08T11:05:08.651Z | use","editorialQuality":{"score":100,"threshold":65,"status":"ready","wordCount":2035,"uniquenessScore":50,"reasons":[]}},"media":{"evidence":{"source":"no-media","verified":false,"confidence":"low","updatedAt":"2026-10-12T01:29:00.615Z","emptyReason":"No screenshots, media assets, or demo links are available."},"primaryImageUrl":null,"mediaAssetCount":0,"assets":[],"demoUrl":null},"ownerResources":{"evidence":{"source":"unclaimed","verified":false,"confidence":"low","updatedAt":"2026-10-12T01:29:00.615Z","emptyReason":"This page has not been claimed by the agent owner."},"hasCustomPage":false,"customPageUpdatedAt":null,"customLinks":[],"structuredLinks":{"docsUrl":null,"demoUrl":null,"supportUrl":null,"pricingUrl":null,"statusUrl":null},"customPage":null},"relatedAgents":{"evidence":{"source":"protocol-neighbors","verified":false,"confidence":"medium","updatedAt":"2026-10-12T11:15:16.504Z","emptyReason":null},"items":[{"id":"8ebccd8e-3863-4187-8355-c3f14e1f9edf","entityType":"agent","canonicalPath":"/agent/iofficeai-aionui","slug":"iofficeai-aionui","name":"AionUi","description":"Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!","url":"https://github.com/iOfficeAI/AionUi","homepage":"https://www.aionui.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-10-09T19:11:12.944Z","createdAt":"2026-02-25T03:38:16.584Z","downloads":null},{"id":"b917f68a-ebff-438e-84f8-3f4b2494c0bc","entityType":"agent","canonicalPath":"/agent/activepieces-activepieces","slug":"activepieces-activepieces","name":"activepieces","description":"AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents","url":"https://github.com/activepieces/activepieces","homepage":"https://www.activepieces.com","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-15T02:22:12.426Z","createdAt":"2026-02-25T03:38:12.412Z","downloads":null},{"id":"5cb26759-3a39-483f-94cf-276a98c13bb8","entityType":"agent","canonicalPath":"/agent/cherryhq-cherry-studio","slug":"cherryhq-cherry-studio","name":"cherry-studio","description":"AI productivity studio with smart chat, autonomous agents, and 300+ assistants. Unified access to frontier LLMs","url":"https://github.com/CherryHQ/cherry-studio","homepage":"https://cherry-ai.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-11T14:38:40.986Z","createdAt":"2026-02-25T03:38:19.379Z","downloads":null},{"id":"6f6582d0-5d76-4f0f-b81d-86520247950b","entityType":"agent","canonicalPath":"/agent/copilotkit-copilotkit","slug":"copilotkit-copilotkit","name":"CopilotKit","description":"The Frontend for Agents & Generative UI. React + Angular","url":"https://github.com/CopilotKit/CopilotKit","homepage":"https://docs.copilotkit.ai","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-03-25T09:50:57.846Z","createdAt":"2026-02-25T03:39:14.617Z","downloads":null}],"links":{"hub":"/agent","source":"/agent/source/clawhub","protocols":[{"label":"OpenClaw","href":"/agent/protocol/openclew"}]}}}