{"id":"8e91bbec-bbd2-428c-98c1-fa66e9baf696","entityType":"agent","slug":"clawhub-deciqai-regression-to-the-mean","name":"Regression to the Mean","canonicalUrl":"https://www.xpersona.co/agent/clawhub-deciqai-regression-to-the-mean","canonicalPath":"/agent/clawhub-deciqai-regression-to-the-mean","generatedAt":"2026-10-13T03:30:00.400Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-12T02:14:17.149Z","emptyReason":null},"description":"Activate when: someone says 'things always bounce back,' wonders why an intervention seemed to work on a struggling team, is surprised their star performer r... Skill: Regression to the Mean Owner: deciqai Summary: Activate when: someone says 'things always bounce back,' wonders why an intervention seemed to work on a struggling team, is surprised their star performer r... Tags: latest:1.0.5 Version history: v1.0.5 | 2026-07-16T18:13:20.862Z | user Description tail link + agents machine-readable metadata line (deciqai.com/s/regression-to-the-mean.json) v1.0.4 | 2026-07-10T10","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 956 downloads reported by the source. Last updated 10/12/2026.","installCommand":"clawhub skill install s17a4mqcnk515kvaca5ze55d0x88pfpx:regression-to-the-mean","sourceUrl":"https://clawhub.ai/deciqai/regression-to-the-mean","homepage":"https://clawhub.ai/deciqai/skills/regression-to-the-mean","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/deciqai/regression-to-the-mean","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/deciqai/skills/regression-to-the-mean","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":60,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"Activate when: someone says 'things always bounce back,' wonders why an intervention seemed to work on a struggling team, is surprised their star performer r..."},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-12T02:14:17.149Z","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":"2026-10-12T02:14:17.149Z","emptyReason":null},"stars":null,"forks":null,"downloads":956,"likes":null,"task":null,"library":null,"packageName":null,"latestVersion":"1.0.5","tractionLabel":"956 downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-12T02:14:17.136Z","emptyReason":null},"lastUpdatedAt":"2026-10-12T02:14:17.149Z","lastCrawledAt":"2026-10-12T02:14:17.136Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-13T02:14:17.136Z","lastVerifiedAt":null,"highlights":[{"version":"1.0.5","createdAt":"2026-07-16T18:13:20.862Z","changelog":"Description tail link + agents machine-readable metadata line (deciqai.com/s/regression-to-the-mean.json)","fileCount":6,"zipByteSize":13752},{"version":"1.0.4","createdAt":"2026-07-10T10:27:51.365Z","changelog":"Add 2024-2026 AI-era worked example + updated sources","fileCount":6,"zipByteSize":13662},{"version":"1.0.3","createdAt":"2026-07-08T11:16:37.998Z","changelog":"Footer now uses /c/<slug> short link (fixes UTM truncation when SKILL.md is read in a terminal)","fileCount":5,"zipByteSize":8958},{"version":"1.0.2","createdAt":"2026-07-08T01:01:20.944Z","changelog":"Refreshed content + GitHub star link in footer","fileCount":5,"zipByteSize":8996},{"version":"1.0.1","createdAt":"2026-07-07T22:32:45.333Z","changelog":"Add catalog categories and topics","fileCount":5,"zipByteSize":8990},{"version":"1.0.0","createdAt":"2026-07-01T14:18:17.300Z","changelog":"Initial publish","fileCount":5,"zipByteSize":9015}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s17a4mqcnk515kvaca5ze55d0x88pfpx:regression-to-the-mean","setupComplexity":"low","setupSteps":["Setup complexity is classified as HIGH. You must provision dedicated cloud infrastructure or an isolated VM. Do not run this directly on your local workstation.","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-regression-to-the-mean/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-regression-to-the-mean/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-regression-to-the-mean/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-regression-to-the-mean/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-regression-to-the-mean/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-regression-to-the-mean/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-13T03:30:00.397Z"}},"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-regression-to-the-mean/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-regression-to-the-mean/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-regression-to-the-mean/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-regression-to-the-mean/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":"2026-10-12T02:14:17.149Z","emptyReason":null},"readme":"Skill: Regression to the Mean\n\nOwner: deciqai\n\nSummary: Activate when: someone says 'things always bounce back,' wonders why an intervention seemed to work on a struggling team, is surprised their star performer r...\n\nTags: latest:1.0.5\n\nVersion history:\n\nv1.0.5 | 2026-07-16T18:13:20.862Z | user\n\nDescription tail link + agents machine-readable metadata line (deciqai.com/s/regression-to-the-mean.json)\n\nv1.0.4 | 2026-07-10T10:27:51.365Z | user\n\nAdd 2024-2026 AI-era worked example + updated sources\n\nv1.0.3 | 2026-07-08T11:16:37.998Z | user\n\nFooter now uses /c/<slug> short link (fixes UTM truncation when SKILL.md is read in a terminal)\n\nv1.0.2 | 2026-07-08T01:01:20.944Z | user\n\nRefreshed content + GitHub star link in footer\n\nv1.0.1 | 2026-07-07T22:32:45.333Z | user\n\nAdd catalog categories and topics\n\nv1.0.0 | 2026-07-01T14:18:17.300Z | user\n\nInitial publish\n\nArchive index:\n\nArchive v1.0.5: 6 files, 13752 bytes\n\nFiles: examples/ai-hot-streak-benchmark-spike-2023-2026.md (8628b), examples/galton-1886-kahneman-2011-israeli-air-force.md (6308b), references/sources.md (1716b), skill-card.md (2549b), SKILL.md (8595b), _meta.json (141b)\n\nFile v1.0.5:SKILL.md\n\n---\nname: regression-to-the-mean\ndescription: \"Activate when: someone says 'things always bounce back,' wonders why an intervention seemed to work on a struggling team, is surprised their star performer regressed, is evaluating whether a coaching/bonus/firing had an effect, or is interpreting before-vs-after performance changes without a control group. Do NOT activate when: the measurement is genuinely noise-free (e.g., deterministic process outputs); the underlying signal has demonstrably changed due to a structural shift (new product launch, technology change). More: deciqai.com/c/regression-to-the-mean\"\n---\n\n# Regression to the Mean\n\n## Overview\n\n**Regression to the mean** is the statistical regularity that any noisy measurement producing an extreme value tends to be followed on retest by a less-extreme value — because the extreme portion was partly driven by non-repeating random noise. There is no \"force pulling back to average\"; it is a mathematical consequence of signal + noise structure.\n\nNamed by Francis Galton (1886) studying parent-child height: tall parents have tall children, but slightly shorter; short parents have short children, but slightly taller. Kahneman's Israeli Air Force example (2011, Ch. 17) is the most-cited operational case — flight instructors concluded punishment works and praise doesn't, but were observing regression, not causation.\n\nComposes with `survivorship-bias` (extreme survivors regress), `probabilistic-thinking` (regression is probabilistic), `narrative-fallacy` (regression drives post-hoc narratives), `fundamental-attribution-error` (attributing regression to character/intervention is FAE).\n\n## When to Use\n\n- Evaluating the effect of an intervention on extreme performers (struggling teams, top sales reps, low-rated branches)\n- Designing or interpreting A/B tests or pilot programs\n- Reviewing year-over-year performance changes\n- Hiring or promoting top performers\n- Evaluating investment fund performance\n- Analyzing acquisition outcomes\n- Building or critiquing causal claims about training, coaching, or feedback\n- Judging whether an AI startup's viral quarter, a fund's AI hot streak, or a model's benchmark spike is a durable trend or an outlier reverting toward average (AI hype extrapolation)\n- Someone says \"regression to the mean,\" \"things will average out,\" \"they always come back\"\n\n**Not when:** the measurement is noise-free (rare in business); the underlying signal is genuinely changing (e.g., the business model fundamentally improved); the intervention is so substantial that no plausible regression can explain the effect.\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user brings a specific performance change → run The Process directly.\n- **Coach mode:** user is new to the framework → 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:** when an extreme performance is followed by a less-extreme one, regression to the mean is the default explanation — not the intervention.\n2. **Check fit.** Did we observe an extreme value followed by retest? If yes, regression applies.\n3. **Elicit the structure.** What was the extreme observation? What was the follow-up? What intervention happened in between? Is there a control or baseline?\n> **[WAIT — do not advance until user responds]**\n4. **One question at a time:** how much regression would we expect without the intervention? Does the observed change exceed it? What's the noise level in the measurement?\n> **[WAIT — do not advance until user responds]**\n5. **Close:** regression vs. causal effect separated + control or baseline identified + claim calibrated.\n> **[WAIT — do not advance until user responds]**\n\n## The Process\n\n**1. Identify the extreme observation** — value, subject, how extreme, any intervention.\n**2. Identify the retest** — follow-up measurement, period, direction (toward mean?).\n**3. Estimate expected regression** — noise level, historical variance; formula: `regression ≈ (1 − reliability) × distance from mean`.\n**4. Compare to control** — untreated extreme performers; intervention effect = treatment change − control change. Without control, regression cannot be ruled out.\n**5. Calibrate causal claim** — did the change exceed the regression baseline? By how much? Confidence?\n**6. Adjust action** — credit/blame intervention only if it exceeds regression; recommend control structure for future tests.\n\n## Output: Regression Analysis\n\n```markdown\n# Regression Analysis: <observed change>\nExtreme observation — Subject: | Value: | Period:\nIntervention — What: | When: | By whom:\nRetest — Follow-up value: | Toward mean (Y/N): | Magnitude:\nExpected regression — Noise level: | Predicted magnitude:\nControl — Group: | Change: | Intervention effect = treatment − control:\nCausal calibration — % regression: | % intervention: | Confidence:\nAdjusted action — Credit/blame verdict + future evaluation process:\n```\n\n*→ Method in Action: [Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force](examples/galton-1886-kahneman-2011-israeli-air-force.md)*\n*→ 2026 lens: [The AI Hot Streak — viral quarters, fund streaks & benchmark spikes (2023–2026)](examples/ai-hot-streak-benchmark-spike-2023-2026.md)*\n\n## Pack: Regression Across Domains\n\n| Domain | Extreme obs. | Regression artifact | Discipline |\n|---|---|---|---|\n| Sales | Top/worst quarter | Regresses next quarter | 4-quarter averages + control |\n| Athlete | Breakout season | Sophomore slump | Multi-season average |\n| Stock fund | Hot year | Underperforms next year | 5-year track record |\n| Customer sat. | Low quarter | Improvement next quarter | Control group |\n| Manager intervention | Struggling team improves | Mostly regression | Compare untreated peer |\n\n## Applying It Well\n\n- Use longer time windows to reduce noise (yearly vs. quarterly)\n- Always ask: \"what would regression alone predict?\" before crediting an intervention\n- Control groups are not optional when evaluating interventions on extreme performers\n- \"Praise hurts, punishment helps\" is the signature cognitive error of regression blindness\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] \"The bonus motivated her / criticism got through to him\" | Exceptional quarter → regresses; bad quarter → improves. Both are regression, not intervention effect. |\n| [D] \"Our turnaround plan / new CEO is working\" | Trough is partly noise. Compare to untreated peers before crediting the plan or leader. |\n| [D] \"This pilot is working on our struggling teams\" | Extreme performers are the worst pilot group — regression is the null. Use a control. |\n| [D] \"Star performers will keep starring / sophomore slump is real\" | Single-period peaks include noise; regression is the default on retest, not a slump. |\n| [D] \"After we fired him the team improved\" | Could be regression. Disentangle by comparing to teams with no change. |\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- An intervention is being evaluated on extreme performers without a control group\n- A single-period change is being treated as evidence of a stable trend\n- Causal credit/blame is being assigned to interventions that coincided with regression\n- \"The intervention worked\" claims lack baseline regression estimates\n- Promotion / firing decisions are based on single-period extreme performance\n- \"Comeback\" narratives credit decisive action, not regression\n\n## Verification\n\n- [ ] Extreme observation identified; retest compared to it\n- [ ] Noise level and regression magnitude estimated\n- [ ] Control group (or baseline regression expectation) used\n- [ ] Causal effect separated from regression; claim calibrated\n- [ ] Long-window measurements used where possible; future evaluation process set\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\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/regression-to-the-mean** · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.*\n\n*Agents: latest version & machine-readable metadata → https://www.deciqai.com/s/regression-to-the-mean.json*\n\nFile v1.0.5:_meta.json\n\n{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"regression-to-the-mean\",\n  \"version\": \"1.0.5\",\n  \"publishedAt\": 1784225600862\n}\n\nFile v1.0.5:references/sources.md\n\n# Sources — regression-to-the-mean\n\n> *Primary sources for the [regression-to-the-mean](../SKILL.md) skill.*\n\n- Galton, F. (1886). \"Regression towards Mediocrity in Hereditary Stature.\" *Journal of the Anthropological Institute*, 15, 246–263. The original paper.\n- Kahneman, D. (2011). *Thinking, Fast and Slow.* Farrar, Straus and Giroux. ISBN 978-0374533557. Chapter 17, \"Regression to the Mean.\"\n- Stigler, S. M. (1986). *The History of Statistics: The Measurement of Uncertainty before 1900.* Belknap. ISBN 978-0674403413. Historical context for Galton's work.\n- Senn, S. J. (2011). \"Francis Galton and regression to the mean.\" *Significance*, 8(3), 124–126. Modern statistical interpretation.\n- Mlodinow, L. (2008). *The Drunkard's Walk: How Randomness Rules Our Lives.* Pantheon. ISBN 978-0307275172. Popular treatment.\n- Tversky, A., & Kahneman, D. (1974). \"Judgment under Uncertainty: Heuristics and Biases.\" *Science*, 185, 1124–1131. The cognitive-bias treatment.\n- Campbell, D. T., & Kenny, D. A. (1999). *A Primer on Regression Artifacts.* Guilford. ISBN 978-1572304826.\n- LMSYS Org — Chatbot Arena Leaderboard (2023–). https://lmarena.ai/ — public, frequently-updated ranking of frontier LLMs; illustrates the rapid turnover of benchmark leaders that makes any single-release \"#1\" spike a poor durable-superiority signal.\n- Sainz, O., et al. (2023). \"NLP Evaluation in Trouble: On the Need to Measure LLM Data Contamination for each Benchmark.\" *Findings of EMNLP 2023.* https://aclanthology.org/2023.findings-emnlp.722/ — documents benchmark data-contamination/overfitting, one reason a single-eval spike overstates true capability and should be expected to regress on broader testing.\n\nFile v1.0.5:examples/ai-hot-streak-benchmark-spike-2023-2026.md\n\n# Method in Action: The AI Hot Streak — Viral Quarters, Fund Streaks, and Benchmark Spikes (2023–2026)\n\n> *Example for the [regression-to-the-mean](../SKILL.md) skill.*\n\nBetween 2023 and 2026, the generative-AI boom produced a steady stream of extreme single-period observations: an app that went viral in one quarter, a venture fund riding a hot streak, a model that spiked to the top of one benchmark leaderboard. The temptation in each case is identical — extrapolate the outlier run into a permanent trend and act on it (raise at the peak valuation, pour capital into the \"proven\" fund, declare a model the new state of the art). Regression to the mean is the default explanation the excitement crowds out.\n\nThis example walks the AI hot-streak pattern through the skill's six Process steps. Rather than lean on one fragile figure, it treats three recurring, well-documented AI-era patterns as instances of the same statistical structure.\n\n## 1. Identify the extreme observation\n\nThree archetypal extreme observations from the 2023–2026 window:\n\n- **The viral quarter.** A consumer AI app posts an explosive stretch — ChatGPT's launch in late 2022 famously reached an enormous user base faster than prior consumer software, and a wave of AI-wrapper apps subsequently posted breakout months of downloads and revenue. The extreme value is a single period (a quarter, a launch window) of growth or engagement far above the app's own baseline.\n- **The fund's hot streak.** A venture or hedge fund posts a standout period of returns during the AI run-up, often concentrated in a handful of AI bets. The extreme value is one fund's short-window performance ranking near the top of its peer set.\n- **The benchmark spike.** A newly released model tops a specific leaderboard — a coding benchmark, a math eval, a chatbot arena — by a notable margin at release. The extreme value is one score on one benchmark at one point in time.\n\nIn all three, the subject is a noisy performer, the value is extreme relative to its own history and its peers, and the \"intervention\" that observers want to credit is some narrative of durable superiority: better product, better fund, better model.\n\n## 2. Identify the retest\n\nThe retest is simply the next comparable measurement period:\n\n- **Viral app → next quarter's retention and growth.** Many breakout AI apps saw growth decelerate and retention settle well below launch-window peaks as novelty faded — a widely observed pattern for consumer apps generally, and repeatedly reported for AI-wrapper products in 2023–2025.\n- **Hot fund → next period's returns.** The fund's subsequent-period performance, and its ranking relative to peers.\n- **Spiking model → the next benchmark, or the same benchmark once competitors respond.** Across the 2023–2025 window, frontier leaderboards changed hands frequently; a model at the top at release was often surpassed within months, and margins on any single eval tended to compress as rivals shipped.\n\nIn each case, the retest tends to land closer to the relevant average than the extreme observation did. That directional move toward the mean is exactly what the skill predicts.\n\n## 3. Estimate expected regression\n\nThe skill's rule of thumb: `regression ≈ (1 − reliability) × distance from mean`. The lower the reliability (the higher the noise share) of a single-period measurement, the more regression to expect.\n\n- **Viral quarter:** low reliability. A launch quarter is dominated by one-time novelty, press cycles, and curiosity traffic — a large transient noise component. Distance from the app's own steady-state baseline is huge. Expected regression is large: most of the launch spike should be presumed non-repeating until proven otherwise.\n- **Fund hot streak:** short-window fund returns are notoriously low-reliability signals. A few concentrated bets in one hot sector inflate a short-window number; the skill's `stock fund` row in the Pack table flags exactly this — a hot year regularly underperforms the next. Expected regression is substantial.\n- **Benchmark spike:** a single eval is a noisy proxy for general capability — vulnerable to benchmark-specific tuning, prompt sensitivity, and (documented in the period) train/test contamination concerns. Expected regression on *that specific number*, and on the ranking, is meaningful.\n\nThe honest move is to state the expected regression *before* crediting any durable-superiority story.\n\n## 4. Compare to control\n\nThe control question: how did comparable, untreated peers move over the same window?\n\n- If **most** breakout consumer apps decelerate after a launch quarter, then a given AI app's deceleration is the baseline, not evidence its product \"cooled off\" for idiosyncratic reasons.\n- If **most** top-decile funds in one period fail to stay top-decile the next, then a hot AI fund reverting is the peer baseline, not proof the manager lost their touch.\n- If **most** benchmark leaders get surpassed within months as competitors ship, then one model losing its #1 spot is the field baseline, not a sign it \"got worse.\"\n\nIntervention effect = the subject's change minus the peer-group change. In the AI hot-streak cases, once you subtract the peer-group regression baseline, the residual \"durable edge\" is usually far smaller than the launch-day narrative implied — and sometimes indistinguishable from zero.\n\n## 5. Calibrate the causal claim\n\nNow separate signal from noise for each:\n\n- **Viral quarter:** How much of the launch spike exceeds what pure novelty-driven regression predicts? The durable signal is *retained* users and *repeat* revenue after novelty decays — not the peak. Calibrate the valuation and the growth trend to the post-decay baseline, with wide confidence bands.\n- **Fund hot streak:** Does the fund's edge persist across multiple periods and diversified bets, or is it one hot sector in one window? A durable edge shows up over a multi-year track record (the skill's prescribed `5-year track record`), not one AI-fueled quarter.\n- **Benchmark spike:** Does the model lead across many independent evals and real user tasks, or spike on one? A durable capability signal is broad, replicated superiority — not a single leaderboard peak that inverts when the next model ships.\n\nIn every case, the calibrated claim is narrower than the hype: \"extreme single-period observation, expect reversion toward the mean, durable edge unproven pending multi-period / multi-benchmark confirmation.\"\n\n## 6. Adjust action\n\n- **Do not extrapolate the outlier run into a permanent trend.** Model the app's trajectory off its post-novelty baseline, not its launch spike, before committing capital or headcount at a peak valuation.\n- **Credit the fund's \"skill\" only across a long window and diversified bets** — not one hot AI quarter. Chasing last period's top fund is a documented way to buy the peak and hold through the reversion.\n- **Judge models on broad, replicated, contamination-resistant evaluation and real-task performance**, not one leaderboard number at release. Treat a single benchmark spike as a hypothesis, not a coronation.\n- **Build the control in upfront.** For any AI bet justified by a breakout number, write down the peer-group baseline and the expected regression *first*, then invest only in the residual edge that survives.\n\nThe 2023–2026 AI cycle did not repeal statistics. Viral quarters, hot streaks, and benchmark spikes are, structurally, the flight instructor's cadet who nailed one maneuver: praise the peak all you like — the next attempt still regresses. The discipline is to size your commitment to the durable signal, not the noise-inflated extreme.\n\n*Sources: Kahneman, D. (2011), *Thinking, Fast and Slow*, Ch. 17 (regression to the mean; the framework applied here). Galton, F. (1886), \"Regression towards Mediocrity in Hereditary Stature,\" *Journal of the Anthropological Institute*, 15, 246–263 (origin of the concept). On the AI-era patterns: ChatGPT's rapid post-launch user growth was widely reported (e.g., Reuters/UBS coverage, early 2023); the frequent turnover of frontier-model benchmark leaders and concerns about benchmark contamination/overfitting were widely documented across 2023–2025 (e.g., LMSYS Chatbot Arena leaderboard dynamics; academic work on benchmark data contamination). Consumer-app novelty decay and the non-persistence of short-window fund performance are long-established, well-documented general patterns (see SPIVA persistence studies). Specific figures for individual apps/funds/models are deliberately omitted where they cannot be verified precisely.*\n\nFile v1.0.5:examples/galton-1886-kahneman-2011-israeli-air-force.md\n\n# Method in Action: Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force\n\n> *Example for the [regression-to-the-mean](../SKILL.md) skill.*\n\n**Francis Galton** (1822–1911) was a Victorian polymath: cousin of Charles Darwin, founder of the field of biometrics, and (less savorily) the founder of eugenics. His 1886 paper documenting the regression phenomenon was a methodological breakthrough. He had collected height data on 928 adult children and 205 sets of parents at the International Health Exhibition in London (1884) — a dataset that was, for its time, extraordinarily large and well-controlled.\n\nGalton's empirical finding was the contraction toward the mean across generations. Tall parents had children averaging shorter than themselves but still taller than the population mean. The contraction was systematic. Galton's framing — \"regression towards mediocrity\" — has stuck, though his interpretation (that a \"force\" was pulling toward mediocrity) is now understood to be wrong.\n\n> \"The law of Regression towards Mediocrity that I have proposed is precisely this: Every peculiarity in a man is shared by his kindred, but on the average in a less degree. ... I find that the height of an adult offspring deviates from the mid-parent height by a fraction (about two-thirds) of the deviation of the mid-parents from the average.\"\n>\n> — Galton (1886), pp. 247, 252.\n\nThe correct interpretation, developed in the 20th century, is that any noisy measurement of a stable underlying signal will produce extreme observations partly via lucky/unlucky noise. Retest will not repeat the noise; the underlying signal is preserved but the noise component regresses.\n\n**Kahneman's Israeli Air Force example (2011, chapter 17).** This is the most influential operational illustration of the regression artifact. Kahneman was teaching Israeli Air Force flight instructors about behavioral psychology. He emphasized that positive reinforcement (praise) is more effective than punishment in training, citing established psychological research.\n\nAn instructor (whom Kahneman described as senior and respected) pushed back:\n\n> \"On many occasions I have praised flight cadets for clean execution of some aerobatic maneuver. The next time they try the same maneuver they usually do worse. On the other hand, I have often screamed into a cadet's earphone for bad execution, and in general he does better on his next try. So please don't tell us that reinforcement works and punishment does not, because the opposite is the case.\"\n>\n> — quoted in Kahneman (2011), p. 175.\n\nKahneman's response was the breakthrough: the instructor was observing regression to the mean, mistaken for a causal effect of praise/criticism.\n\nWhen a cadet executes a maneuver exceptionally well, that performance includes a positive noise component (lucky day, favorable winds, instructor's attention, etc.). On the next attempt, the noise component will be different (likely less favorable); the underlying skill is the same; the next performance is statistically expected to be worse — regardless of whether the instructor praised or did nothing.\n\nWhen a cadet executes badly, the performance includes a negative noise component. On the next attempt, the noise is different; the next performance is statistically expected to be better — regardless of whether the instructor screamed or did nothing.\n\nThe instructor was reading: praise → worse performance, criticism → better performance. The truth was: extreme observations → regression to the mean. The instructor's causal inference was driven by a statistical artifact.\n\nKahneman's framing:\n\n> \"Because we tend to reward others when they do well and punish them when they do badly, and because there is regression to the mean, it is part of the human condition that we are statistically punished for rewarding others and rewarded for punishing them.\"\n>\n> — Kahneman (2011), p. 175.\n\nThis is generalizable. Many domains exhibit the same pattern:\n\n**Sales performance.** A salesperson who blew out a quarter gets a bonus. Next quarter they perform less well; manager wonders if the bonus de-motivated them. A salesperson who had a terrible quarter gets a stern talk. Next quarter they perform better; manager thinks the talk worked. Both inferences are likely regression.\n\n**Sports coaching.** A breakout rookie season gets praise; sophomore year is worse — coaches wonder what went wrong. A struggling player gets benched; on return they perform better — coaches credit the benching. Both partly regression.\n\n**Investment fund manager evaluation.** A fund manager who beat the market gets capital inflows; next period they underperform — LPs feel duped. A fund manager who underperformed gets fired; the replacement does better — board credits the replacement. Both partly regression.\n\n**Acquisition targets.** Companies are typically acquired near peak performance. Post-acquisition, performance regresses. The acquirer is disappointed; narrative blames integration. Often: partly regression.\n\n**Patient improvement after medical intervention.** Patients seek treatment when symptoms are most extreme. Symptoms regress regardless of treatment. The patient credits the treatment.\n\n**Customer satisfaction interventions.** A team with low satisfaction gets training; satisfaction improves. Without a control, the improvement could be mostly regression.\n\nThree operational lessons:\n\n**First, use control groups or baseline regression estimates.** When evaluating interventions, the question is not \"did the metric change?\" but \"did the metric change more than it would have changed by regression alone?\" The control group reveals the regression baseline.\n\n**Second, watch for \"praise reduces performance, criticism improves it\" patterns.** When you find yourself thinking that punishment works and praise doesn't, you are almost certainly observing regression artifact. Update your hiring, management, and intervention conclusions accordingly.\n\n**Third, beware celebrating or punishing single-period extreme performance.** A single-quarter top performer is partly noise; promoting them based on one quarter will partly regret. A single-quarter struggler is partly noise; firing them based on one quarter will partly regret. Use longer windows to reduce noise.\n\nFile v1.0.5:skill-card.md\n\n## Description:\n\nHelps agents analyze extreme before-and-after performance changes by separating regression to the mean from plausible causal effects of interventions.\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 agents use this skill to evaluate whether a surprising performance change is likely regression to the mean or evidence of a real intervention effect. It is most useful for coaching, hiring, investment, A/B testing, model evaluation, and other decisions involving noisy extreme observations.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Users may treat examples, current-event claims, or cited statistics as automatically verified facts for financial, medical, or other high-stakes decisions.\n\nMitigation: Use the skill as a reasoning aid, verify factual claims independently, and require appropriate controls or baseline estimates before acting.\n\nRisk: Users may over-attribute an observed change to regression and miss a real structural shift or intervention effect.\n\nMitigation: Check whether the underlying signal changed, compare against a control group or baseline, and calibrate any causal claim to the effect that remains after expected regression.\n\n## Reference(s):\n\n- [Sources - regression-to-the-mean](references/sources.md)\n- [Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force](examples/galton-1886-kahneman-2011-israeli-air-force.md)\n- [The AI Hot Streak - Viral Quarters, Fund Streaks, and Benchmark Spikes (2023-2026)](examples/ai-hot-streak-benchmark-spike-2023-2026.md)\n- [LMSYS Org - Chatbot Arena Leaderboard](https://lmarena.ai/)\n- [NLP Evaluation in Trouble: On the Need to Measure LLM Data Contamination for each Benchmark](https://aclanthology.org/2023.findings-emnlp.722/)\n\n## Skill Output:\n\n**Output Type(s):** [Analysis, Markdown, Guidance]\n\n**Output Format:** [Markdown regression analysis with structured fields and concise recommendations]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [May pause for user input in coach mode before completing the analysis.]\n\n## Skill Version(s):\n\n1.0.5 (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 v1.0.4: 6 files, 13662 bytes\n\nFiles: examples/ai-hot-streak-benchmark-spike-2023-2026.md (8628b), examples/galton-1886-kahneman-2011-israeli-air-force.md (6308b), references/sources.md (1716b), skill-card.md (2593b), SKILL.md (8440b), _meta.json (141b)\n\nFile v1.0.4:SKILL.md\n\n---\nname: regression-to-the-mean\ndescription: \"Activate when: someone says 'things always bounce back,' wonders why an intervention seemed to work on a struggling team, is surprised their star performer regressed, is evaluating whether a coaching/bonus/firing had an effect, or is interpreting before-vs-after performance changes without a control group. Do NOT activate when: the measurement is genuinely noise-free (e.g., deterministic process outputs); the underlying signal has demonstrably changed due to a structural shift (new product launch, technology change).\"\n---\n\n# Regression to the Mean\n\n## Overview\n\n**Regression to the mean** is the statistical regularity that any noisy measurement producing an extreme value tends to be followed on retest by a less-extreme value — because the extreme portion was partly driven by non-repeating random noise. There is no \"force pulling back to average\"; it is a mathematical consequence of signal + noise structure.\n\nNamed by Francis Galton (1886) studying parent-child height: tall parents have tall children, but slightly shorter; short parents have short children, but slightly taller. Kahneman's Israeli Air Force example (2011, Ch. 17) is the most-cited operational case — flight instructors concluded punishment works and praise doesn't, but were observing regression, not causation.\n\nComposes with `survivorship-bias` (extreme survivors regress), `probabilistic-thinking` (regression is probabilistic), `narrative-fallacy` (regression drives post-hoc narratives), `fundamental-attribution-error` (attributing regression to character/intervention is FAE).\n\n## When to Use\n\n- Evaluating the effect of an intervention on extreme performers (struggling teams, top sales reps, low-rated branches)\n- Designing or interpreting A/B tests or pilot programs\n- Reviewing year-over-year performance changes\n- Hiring or promoting top performers\n- Evaluating investment fund performance\n- Analyzing acquisition outcomes\n- Building or critiquing causal claims about training, coaching, or feedback\n- Judging whether an AI startup's viral quarter, a fund's AI hot streak, or a model's benchmark spike is a durable trend or an outlier reverting toward average (AI hype extrapolation)\n- Someone says \"regression to the mean,\" \"things will average out,\" \"they always come back\"\n\n**Not when:** the measurement is noise-free (rare in business); the underlying signal is genuinely changing (e.g., the business model fundamentally improved); the intervention is so substantial that no plausible regression can explain the effect.\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user brings a specific performance change → run The Process directly.\n- **Coach mode:** user is new to the framework → 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:** when an extreme performance is followed by a less-extreme one, regression to the mean is the default explanation — not the intervention.\n2. **Check fit.** Did we observe an extreme value followed by retest? If yes, regression applies.\n3. **Elicit the structure.** What was the extreme observation? What was the follow-up? What intervention happened in between? Is there a control or baseline?\n> **[WAIT — do not advance until user responds]**\n4. **One question at a time:** how much regression would we expect without the intervention? Does the observed change exceed it? What's the noise level in the measurement?\n> **[WAIT — do not advance until user responds]**\n5. **Close:** regression vs. causal effect separated + control or baseline identified + claim calibrated.\n> **[WAIT — do not advance until user responds]**\n\n## The Process\n\n**1. Identify the extreme observation** — value, subject, how extreme, any intervention.\n**2. Identify the retest** — follow-up measurement, period, direction (toward mean?).\n**3. Estimate expected regression** — noise level, historical variance; formula: `regression ≈ (1 − reliability) × distance from mean`.\n**4. Compare to control** — untreated extreme performers; intervention effect = treatment change − control change. Without control, regression cannot be ruled out.\n**5. Calibrate causal claim** — did the change exceed the regression baseline? By how much? Confidence?\n**6. Adjust action** — credit/blame intervention only if it exceeds regression; recommend control structure for future tests.\n\n## Output: Regression Analysis\n\n```markdown\n# Regression Analysis: <observed change>\nExtreme observation — Subject: | Value: | Period:\nIntervention — What: | When: | By whom:\nRetest — Follow-up value: | Toward mean (Y/N): | Magnitude:\nExpected regression — Noise level: | Predicted magnitude:\nControl — Group: | Change: | Intervention effect = treatment − control:\nCausal calibration — % regression: | % intervention: | Confidence:\nAdjusted action — Credit/blame verdict + future evaluation process:\n```\n\n*→ Method in Action: [Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force](examples/galton-1886-kahneman-2011-israeli-air-force.md)*\n*→ 2026 lens: [The AI Hot Streak — viral quarters, fund streaks & benchmark spikes (2023–2026)](examples/ai-hot-streak-benchmark-spike-2023-2026.md)*\n\n## Pack: Regression Across Domains\n\n| Domain | Extreme obs. | Regression artifact | Discipline |\n|---|---|---|---|\n| Sales | Top/worst quarter | Regresses next quarter | 4-quarter averages + control |\n| Athlete | Breakout season | Sophomore slump | Multi-season average |\n| Stock fund | Hot year | Underperforms next year | 5-year track record |\n| Customer sat. | Low quarter | Improvement next quarter | Control group |\n| Manager intervention | Struggling team improves | Mostly regression | Compare untreated peer |\n\n## Applying It Well\n\n- Use longer time windows to reduce noise (yearly vs. quarterly)\n- Always ask: \"what would regression alone predict?\" before crediting an intervention\n- Control groups are not optional when evaluating interventions on extreme performers\n- \"Praise hurts, punishment helps\" is the signature cognitive error of regression blindness\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] \"The bonus motivated her / criticism got through to him\" | Exceptional quarter → regresses; bad quarter → improves. Both are regression, not intervention effect. |\n| [D] \"Our turnaround plan / new CEO is working\" | Trough is partly noise. Compare to untreated peers before crediting the plan or leader. |\n| [D] \"This pilot is working on our struggling teams\" | Extreme performers are the worst pilot group — regression is the null. Use a control. |\n| [D] \"Star performers will keep starring / sophomore slump is real\" | Single-period peaks include noise; regression is the default on retest, not a slump. |\n| [D] \"After we fired him the team improved\" | Could be regression. Disentangle by comparing to teams with no change. |\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- An intervention is being evaluated on extreme performers without a control group\n- A single-period change is being treated as evidence of a stable trend\n- Causal credit/blame is being assigned to interventions that coincided with regression\n- \"The intervention worked\" claims lack baseline regression estimates\n- Promotion / firing decisions are based on single-period extreme performance\n- \"Comeback\" narratives credit decisive action, not regression\n\n## Verification\n\n- [ ] Extreme observation identified; retest compared to it\n- [ ] Noise level and regression magnitude estimated\n- [ ] Control group (or baseline regression expectation) used\n- [ ] Causal effect separated from regression; claim calibrated\n- [ ] Long-window measurements used where possible; future evaluation process set\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\n\n---\n\n*Part of **deciqAI Knowledge Skills** — 223 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/regression-to-the-mean** · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.*\n\nFile v1.0.4:_meta.json\n\n{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"regression-to-the-mean\",\n  \"version\": \"1.0.4\",\n  \"publishedAt\": 1783679271365\n}\n\nFile v1.0.4:references/sources.md\n\n# Sources — regression-to-the-mean\n\n> *Primary sources for the [regression-to-the-mean](../SKILL.md) skill.*\n\n- Galton, F. (1886). \"Regression towards Mediocrity in Hereditary Stature.\" *Journal of the Anthropological Institute*, 15, 246–263. The original paper.\n- Kahneman, D. (2011). *Thinking, Fast and Slow.* Farrar, Straus and Giroux. ISBN 978-0374533557. Chapter 17, \"Regression to the Mean.\"\n- Stigler, S. M. (1986). *The History of Statistics: The Measurement of Uncertainty before 1900.* Belknap. ISBN 978-0674403413. Historical context for Galton's work.\n- Senn, S. J. (2011). \"Francis Galton and regression to the mean.\" *Significance*, 8(3), 124–126. Modern statistical interpretation.\n- Mlodinow, L. (2008). *The Drunkard's Walk: How Randomness Rules Our Lives.* Pantheon. ISBN 978-0307275172. Popular treatment.\n- Tversky, A., & Kahneman, D. (1974). \"Judgment under Uncertainty: Heuristics and Biases.\" *Science*, 185, 1124–1131. The cognitive-bias treatment.\n- Campbell, D. T., & Kenny, D. A. (1999). *A Primer on Regression Artifacts.* Guilford. ISBN 978-1572304826.\n- LMSYS Org — Chatbot Arena Leaderboard (2023–). https://lmarena.ai/ — public, frequently-updated ranking of frontier LLMs; illustrates the rapid turnover of benchmark leaders that makes any single-release \"#1\" spike a poor durable-superiority signal.\n- Sainz, O., et al. (2023). \"NLP Evaluation in Trouble: On the Need to Measure LLM Data Contamination for each Benchmark.\" *Findings of EMNLP 2023.* https://aclanthology.org/2023.findings-emnlp.722/ — documents benchmark data-contamination/overfitting, one reason a single-eval spike overstates true capability and should be expected to regress on broader testing.\n\nFile v1.0.4:examples/ai-hot-streak-benchmark-spike-2023-2026.md\n\n# Method in Action: The AI Hot Streak — Viral Quarters, Fund Streaks, and Benchmark Spikes (2023–2026)\n\n> *Example for the [regression-to-the-mean](../SKILL.md) skill.*\n\nBetween 2023 and 2026, the generative-AI boom produced a steady stream of extreme single-period observations: an app that went viral in one quarter, a venture fund riding a hot streak, a model that spiked to the top of one benchmark leaderboard. The temptation in each case is identical — extrapolate the outlier run into a permanent trend and act on it (raise at the peak valuation, pour capital into the \"proven\" fund, declare a model the new state of the art). Regression to the mean is the default explanation the excitement crowds out.\n\nThis example walks the AI hot-streak pattern through the skill's six Process steps. Rather than lean on one fragile figure, it treats three recurring, well-documented AI-era patterns as instances of the same statistical structure.\n\n## 1. Identify the extreme observation\n\nThree archetypal extreme observations from the 2023–2026 window:\n\n- **The viral quarter.** A consumer AI app posts an explosive stretch — ChatGPT's launch in late 2022 famously reached an enormous user base faster than prior consumer software, and a wave of AI-wrapper apps subsequently posted breakout months of downloads and revenue. The extreme value is a single period (a quarter, a launch window) of growth or engagement far above the app's own baseline.\n- **The fund's hot streak.** A venture or hedge fund posts a standout period of returns during the AI run-up, often concentrated in a handful of AI bets. The extreme value is one fund's short-window performance ranking near the top of its peer set.\n- **The benchmark spike.** A newly released model tops a specific leaderboard — a coding benchmark, a math eval, a chatbot arena — by a notable margin at release. The extreme value is one score on one benchmark at one point in time.\n\nIn all three, the subject is a noisy performer, the value is extreme relative to its own history and its peers, and the \"intervention\" that observers want to credit is some narrative of durable superiority: better product, better fund, better model.\n\n## 2. Identify the retest\n\nThe retest is simply the next comparable measurement period:\n\n- **Viral app → next quarter's retention and growth.** Many breakout AI apps saw growth decelerate and retention settle well below launch-window peaks as novelty faded — a widely observed pattern for consumer apps generally, and repeatedly reported for AI-wrapper products in 2023–2025.\n- **Hot fund → next period's returns.** The fund's subsequent-period performance, and its ranking relative to peers.\n- **Spiking model → the next benchmark, or the same benchmark once competitors respond.** Across the 2023–2025 window, frontier leaderboards changed hands frequently; a model at the top at release was often surpassed within months, and margins on any single eval tended to compress as rivals shipped.\n\nIn each case, the retest tends to land closer to the relevant average than the extreme observation did. That directional move toward the mean is exactly what the skill predicts.\n\n## 3. Estimate expected regression\n\nThe skill's rule of thumb: `regression ≈ (1 − reliability) × distance from mean`. The lower the reliability (the higher the noise share) of a single-period measurement, the more regression to expect.\n\n- **Viral quarter:** low reliability. A launch quarter is dominated by one-time novelty, press cycles, and curiosity traffic — a large transient noise component. Distance from the app's own steady-state baseline is huge. Expected regression is large: most of the launch spike should be presumed non-repeating until proven otherwise.\n- **Fund hot streak:** short-window fund returns are notoriously low-reliability signals. A few concentrated bets in one hot sector inflate a short-window number; the skill's `stock fund` row in the Pack table flags exactly this — a hot year regularly underperforms the next. Expected regression is substantial.\n- **Benchmark spike:** a single eval is a noisy proxy for general capability — vulnerable to benchmark-specific tuning, prompt sensitivity, and (documented in the period) train/test contamination concerns. Expected regression on *that specific number*, and on the ranking, is meaningful.\n\nThe honest move is to state the expected regression *before* crediting any durable-superiority story.\n\n## 4. Compare to control\n\nThe control question: how did comparable, untreated peers move over the same window?\n\n- If **most** breakout consumer apps decelerate after a launch quarter, then a given AI app's deceleration is the baseline, not evidence its product \"cooled off\" for idiosyncratic reasons.\n- If **most** top-decile funds in one period fail to stay top-decile the next, then a hot AI fund reverting is the peer baseline, not proof the manager lost their touch.\n- If **most** benchmark leaders get surpassed within months as competitors ship, then one model losing its #1 spot is the field baseline, not a sign it \"got worse.\"\n\nIntervention effect = the subject's change minus the peer-group change. In the AI hot-streak cases, once you subtract the peer-group regression baseline, the residual \"durable edge\" is usually far smaller than the launch-day narrative implied — and sometimes indistinguishable from zero.\n\n## 5. Calibrate the causal claim\n\nNow separate signal from noise for each:\n\n- **Viral quarter:** How much of the launch spike exceeds what pure novelty-driven regression predicts? The durable signal is *retained* users and *repeat* revenue after novelty decays — not the peak. Calibrate the valuation and the growth trend to the post-decay baseline, with wide confidence bands.\n- **Fund hot streak:** Does the fund's edge persist across multiple periods and diversified bets, or is it one hot sector in one window? A durable edge shows up over a multi-year track record (the skill's prescribed `5-year track record`), not one AI-fueled quarter.\n- **Benchmark spike:** Does the model lead across many independent evals and real user tasks, or spike on one? A durable capability signal is broad, replicated superiority — not a single leaderboard peak that inverts when the next model ships.\n\nIn every case, the calibrated claim is narrower than the hype: \"extreme single-period observation, expect reversion toward the mean, durable edge unproven pending multi-period / multi-benchmark confirmation.\"\n\n## 6. Adjust action\n\n- **Do not extrapolate the outlier run into a permanent trend.** Model the app's trajectory off its post-novelty baseline, not its launch spike, before committing capital or headcount at a peak valuation.\n- **Credit the fund's \"skill\" only across a long window and diversified bets** — not one hot AI quarter. Chasing last period's top fund is a documented way to buy the peak and hold through the reversion.\n- **Judge models on broad, replicated, contamination-resistant evaluation and real-task performance**, not one leaderboard number at release. Treat a single benchmark spike as a hypothesis, not a coronation.\n- **Build the control in upfront.** For any AI bet justified by a breakout number, write down the peer-group baseline and the expected regression *first*, then invest only in the residual edge that survives.\n\nThe 2023–2026 AI cycle did not repeal statistics. Viral quarters, hot streaks, and benchmark spikes are, structurally, the flight instructor's cadet who nailed one maneuver: praise the peak all you like — the next attempt still regresses. The discipline is to size your commitment to the durable signal, not the noise-inflated extreme.\n\n*Sources: Kahneman, D. (2011), *Thinking, Fast and Slow*, Ch. 17 (regression to the mean; the framework applied here). Galton, F. (1886), \"Regression towards Mediocrity in Hereditary Stature,\" *Journal of the Anthropological Institute*, 15, 246–263 (origin of the concept). On the AI-era patterns: ChatGPT's rapid post-launch user growth was widely reported (e.g., Reuters/UBS coverage, early 2023); the frequent turnover of frontier-model benchmark leaders and concerns about benchmark contamination/overfitting were widely documented across 2023–2025 (e.g., LMSYS Chatbot Arena leaderboard dynamics; academic work on benchmark data contamination). Consumer-app novelty decay and the non-persistence of short-window fund performance are long-established, well-documented general patterns (see SPIVA persistence studies). Specific figures for individual apps/funds/models are deliberately omitted where they cannot be verified precisely.*\n\nFile v1.0.4:examples/galton-1886-kahneman-2011-israeli-air-force.md\n\n# Method in Action: Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force\n\n> *Example for the [regression-to-the-mean](../SKILL.md) skill.*\n\n**Francis Galton** (1822–1911) was a Victorian polymath: cousin of Charles Darwin, founder of the field of biometrics, and (less savorily) the founder of eugenics. His 1886 paper documenting the regression phenomenon was a methodological breakthrough. He had collected height data on 928 adult children and 205 sets of parents at the International Health Exhibition in London (1884) — a dataset that was, for its time, extraordinarily large and well-controlled.\n\nGalton's empirical finding was the contraction toward the mean across generations. Tall parents had children averaging shorter than themselves but still taller than the population mean. The contraction was systematic. Galton's framing — \"regression towards mediocrity\" — has stuck, though his interpretation (that a \"force\" was pulling toward mediocrity) is now understood to be wrong.\n\n> \"The law of Regression towards Mediocrity that I have proposed is precisely this: Every peculiarity in a man is shared by his kindred, but on the average in a less degree. ... I find that the height of an adult offspring deviates from the mid-parent height by a fraction (about two-thirds) of the deviation of the mid-parents from the average.\"\n>\n> — Galton (1886), pp. 247, 252.\n\nThe correct interpretation, developed in the 20th century, is that any noisy measurement of a stable underlying signal will produce extreme observations partly via lucky/unlucky noise. Retest will not repeat the noise; the underlying signal is preserved but the noise component regresses.\n\n**Kahneman's Israeli Air Force example (2011, chapter 17).** This is the most influential operational illustration of the regression artifact. Kahneman was teaching Israeli Air Force flight instructors about behavioral psychology. He emphasized that positive reinforcement (praise) is more effective than punishment in training, citing established psychological research.\n\nAn instructor (whom Kahneman described as senior and respected) pushed back:\n\n> \"On many occasions I have praised flight cadets for clean execution of some aerobatic maneuver. The next time they try the same maneuver they usually do worse. On the other hand, I have often screamed into a cadet's earphone for bad execution, and in general he does better on his next try. So please don't tell us that reinforcement works and punishment does not, because the opposite is the case.\"\n>\n> — quoted in Kahneman (2011), p. 175.\n\nKahneman's response was the breakthrough: the instructor was observing regression to the mean, mistaken for a causal effect of praise/criticism.\n\nWhen a cadet executes a maneuver exceptionally well, that performance includes a positive noise component (lucky day, favorable winds, instructor's attention, etc.). On the next attempt, the noise component will be different (likely less favorable); the underlying skill is the same; the next performance is statistically expected to be worse — regardless of whether the instructor praised or did nothing.\n\nWhen a cadet executes badly, the performance includes a negative noise component. On the next attempt, the noise is different; the next performance is statistically expected to be better — regardless of whether the instructor screamed or did nothing.\n\nThe instructor was reading: praise → worse performance, criticism → better performance. The truth was: extreme observations → regression to the mean. The instructor's causal inference was driven by a statistical artifact.\n\nKahneman's framing:\n\n> \"Because we tend to reward others when they do well and punish them when they do badly, and because there is regression to the mean, it is part of the human condition that we are statistically punished for rewarding others and rewarded for punishing them.\"\n>\n> — Kahneman (2011), p. 175.\n\nThis is generalizable. Many domains exhibit the same pattern:\n\n**Sales performance.** A salesperson who blew out a quarter gets a bonus. Next quarter they perform less well; manager wonders if the bonus de-motivated them. A salesperson who had a terrible quarter gets a stern talk. Next quarter they perform better; manager thinks the talk worked. Both inferences are likely regression.\n\n**Sports coaching.** A breakout rookie season gets praise; sophomore year is worse — coaches wonder what went wrong. A struggling player gets benched; on return they perform better — coaches credit the benching. Both partly regression.\n\n**Investment fund manager evaluation.** A fund manager who beat the market gets capital inflows; next period they underperform — LPs feel duped. A fund manager who underperformed gets fired; the replacement does better — board credits the replacement. Both partly regression.\n\n**Acquisition targets.** Companies are typically acquired near peak performance. Post-acquisition, performance regresses. The acquirer is disappointed; narrative blames integration. Often: partly regression.\n\n**Patient improvement after medical intervention.** Patients seek treatment when symptoms are most extreme. Symptoms regress regardless of treatment. The patient credits the treatment.\n\n**Customer satisfaction interventions.** A team with low satisfaction gets training; satisfaction improves. Without a control, the improvement could be mostly regression.\n\nThree operational lessons:\n\n**First, use control groups or baseline regression estimates.** When evaluating interventions, the question is not \"did the metric change?\" but \"did the metric change more than it would have changed by regression alone?\" The control group reveals the regression baseline.\n\n**Second, watch for \"praise reduces performance, criticism improves it\" patterns.** When you find yourself thinking that punishment works and praise doesn't, you are almost certainly observing regression artifact. Update your hiring, management, and intervention conclusions accordingly.\n\n**Third, beware celebrating or punishing single-period extreme performance.** A single-quarter top performer is partly noise; promoting them based on one quarter will partly regret. A single-quarter struggler is partly noise; firing them based on one quarter will partly regret. Use longer windows to reduce noise.\n\nFile v1.0.4:skill-card.md\n\n## Description: <br>\nHelps agents recognize when extreme before-and-after performance changes may reflect regression to the mean rather than a causal effect from coaching, bonuses, firings, pilots, or other interventions. <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>\nBusiness operators, analysts, managers, investors, and AI agents use this skill to evaluate noisy performance claims, separate regression artifacts from causal effects, and design better control or baseline comparisons. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: The skill may activate on broad everyday language about things bouncing back and over-apply a regression-to-the-mean frame. <br>\nMitigation: Use the fit check in the skill: confirm there was an extreme noisy observation, a retest, and a meaningful baseline or control before relying on the analysis. <br>\nRisk: Users may mistake regression analysis for proof that an intervention had no causal effect. <br>\nMitigation: Treat the output as causal calibration: compare against untreated peers or a baseline regression estimate, then state uncertainty and confidence explicitly. <br>\n\n\n## Reference(s): <br>\n- [Sources - regression-to-the-mean](references/sources.md) <br>\n- [Galton 1886 + Kahneman 2011 Israeli Air Force example](examples/galton-1886-kahneman-2011-israeli-air-force.md) <br>\n- [AI hot-streak benchmark-spike example](examples/ai-hot-streak-benchmark-spike-2023-2026.md) <br>\n- [LMSYS Chatbot Arena Leaderboard](https://lmarena.ai/) <br>\n- [NLP Evaluation in Trouble](https://aclanthology.org/2023.findings-emnlp.722/) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [text, markdown, guidance] <br>\n**Output Format:** [Markdown regression analysis with structured fields for the extreme observation, retest, expected regression, control comparison, causal calibration, and adjusted action.] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [No executable code, shell commands, credentials, persistence, or hidden system actions are produced by the skill.] <br>\n\n## Skill Version(s): <br>\n1.0.4 (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 v1.0.3: 5 files, 8958 bytes\n\nFiles: examples/galton-1886-kahneman-2011-israeli-air-force.md (6308b), references/sources.md (1091b), skill-card.md (2289b), SKILL.md (8100b), _meta.json (141b)\n\nFile v1.0.3:SKILL.md\n\n---\nname: regression-to-the-mean\ndescription: \"Activate when: someone says 'things always bounce back,' wonders why an intervention seemed to work on a struggling team, is surprised their star performer regressed, is evaluating whether a coaching/bonus/firing had an effect, or is interpreting before-vs-after performance changes without a control group. Do NOT activate when: the measurement is genuinely noise-free (e.g., deterministic process outputs); the underlying signal has demonstrably changed due to a structural shift (new product launch, technology change).\"\n---\n\n# Regression to the Mean\n\n## Overview\n\n**Regression to the mean** is the statistical regularity that any noisy measurement producing an extreme value tends to be followed on retest by a less-extreme value — because the extreme portion was partly driven by non-repeating random noise. There is no \"force pulling back to average\"; it is a mathematical consequence of signal + noise structure.\n\nNamed by Francis Galton (1886) studying parent-child height: tall parents have tall children, but slightly shorter; short parents have short children, but slightly taller. Kahneman's Israeli Air Force example (2011, Ch. 17) is the most-cited operational case — flight instructors concluded punishment works and praise doesn't, but were observing regression, not causation.\n\nComposes with `survivorship-bias` (extreme survivors regress), `probabilistic-thinking` (regression is probabilistic), `narrative-fallacy` (regression drives post-hoc narratives), `fundamental-attribution-error` (attributing regression to character/intervention is FAE).\n\n## When to Use\n\n- Evaluating the effect of an intervention on extreme performers (struggling teams, top sales reps, low-rated branches)\n- Designing or interpreting A/B tests or pilot programs\n- Reviewing year-over-year performance changes\n- Hiring or promoting top performers\n- Evaluating investment fund performance\n- Analyzing acquisition outcomes\n- Building or critiquing causal claims about training, coaching, or feedback\n- Someone says \"regression to the mean,\" \"things will average out,\" \"they always come back\"\n\n**Not when:** the measurement is noise-free (rare in business); the underlying signal is genuinely changing (e.g., the business model fundamentally improved); the intervention is so substantial that no plausible regression can explain the effect.\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user brings a specific performance change → run The Process directly.\n- **Coach mode:** user is new to the framework → 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:** when an extreme performance is followed by a less-extreme one, regression to the mean is the default explanation — not the intervention.\n2. **Check fit.** Did we observe an extreme value followed by retest? If yes, regression applies.\n3. **Elicit the structure.** What was the extreme observation? What was the follow-up? What intervention happened in between? Is there a control or baseline?\n> **[WAIT — do not advance until user responds]**\n4. **One question at a time:** how much regression would we expect without the intervention? Does the observed change exceed it? What's the noise level in the measurement?\n> **[WAIT — do not advance until user responds]**\n5. **Close:** regression vs. causal effect separated + control or baseline identified + claim calibrated.\n> **[WAIT — do not advance until user responds]**\n\n## The Process\n\n**1. Identify the extreme observation** — value, subject, how extreme, any intervention.\n**2. Identify the retest** — follow-up measurement, period, direction (toward mean?).\n**3. Estimate expected regression** — noise level, historical variance; formula: `regression ≈ (1 − reliability) × distance from mean`.\n**4. Compare to control** — untreated extreme performers; intervention effect = treatment change − control change. Without control, regression cannot be ruled out.\n**5. Calibrate causal claim** — did the change exceed the regression baseline? By how much? Confidence?\n**6. Adjust action** — credit/blame intervention only if it exceeds regression; recommend control structure for future tests.\n\n## Output: Regression Analysis\n\n```markdown\n# Regression Analysis: <observed change>\nExtreme observation — Subject: | Value: | Period:\nIntervention — What: | When: | By whom:\nRetest — Follow-up value: | Toward mean (Y/N): | Magnitude:\nExpected regression — Noise level: | Predicted magnitude:\nControl — Group: | Change: | Intervention effect = treatment − control:\nCausal calibration — % regression: | % intervention: | Confidence:\nAdjusted action — Credit/blame verdict + future evaluation process:\n```\n\n*→ Method in Action: [Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force](examples/galton-1886-kahneman-2011-israeli-air-force.md)*\n\n## Pack: Regression Across Domains\n\n| Domain | Extreme obs. | Regression artifact | Discipline |\n|---|---|---|---|\n| Sales | Top/worst quarter | Regresses next quarter | 4-quarter averages + control |\n| Athlete | Breakout season | Sophomore slump | Multi-season average |\n| Stock fund | Hot year | Underperforms next year | 5-year track record |\n| Customer sat. | Low quarter | Improvement next quarter | Control group |\n| Manager intervention | Struggling team improves | Mostly regression | Compare untreated peer |\n\n## Applying It Well\n\n- Use longer time windows to reduce noise (yearly vs. quarterly)\n- Always ask: \"what would regression alone predict?\" before crediting an intervention\n- Control groups are not optional when evaluating interventions on extreme performers\n- \"Praise hurts, punishment helps\" is the signature cognitive error of regression blindness\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] \"The bonus motivated her / criticism got through to him\" | Exceptional quarter → regresses; bad quarter → improves. Both are regression, not intervention effect. |\n| [D] \"Our turnaround plan / new CEO is working\" | Trough is partly noise. Compare to untreated peers before crediting the plan or leader. |\n| [D] \"This pilot is working on our struggling teams\" | Extreme performers are the worst pilot group — regression is the null. Use a control. |\n| [D] \"Star performers will keep starring / sophomore slump is real\" | Single-period peaks include noise; regression is the default on retest, not a slump. |\n| [D] \"After we fired him the team improved\" | Could be regression. Disentangle by comparing to teams with no change. |\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- An intervention is being evaluated on extreme performers without a control group\n- A single-period change is being treated as evidence of a stable trend\n- Causal credit/blame is being assigned to interventions that coincided with regression\n- \"The intervention worked\" claims lack baseline regression estimates\n- Promotion / firing decisions are based on single-period extreme performance\n- \"Comeback\" narratives credit decisive action, not regression\n\n## Verification\n\n- [ ] Extreme observation identified; retest compared to it\n- [ ] Noise level and regression magnitude estimated\n- [ ] Control group (or baseline regression expectation) used\n- [ ] Causal effect separated from regression; claim calibrated\n- [ ] Long-window measurements used where possible; future evaluation process set\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\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/regression-to-the-mean** · ⭐ 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\": \"regression-to-the-mean\",\n  \"version\": \"1.0.3\",\n  \"publishedAt\": 1783509397998\n}\n\nFile v1.0.3:references/sources.md\n\n# Sources — regression-to-the-mean\n\n> *Primary sources for the [regression-to-the-mean](../SKILL.md) skill.*\n\n- Galton, F. (1886). \"Regression towards Mediocrity in Hereditary Stature.\" *Journal of the Anthropological Institute*, 15, 246–263. The original paper.\n- Kahneman, D. (2011). *Thinking, Fast and Slow.* Farrar, Straus and Giroux. ISBN 978-0374533557. Chapter 17, \"Regression to the Mean.\"\n- Stigler, S. M. (1986). *The History of Statistics: The Measurement of Uncertainty before 1900.* Belknap. ISBN 978-0674403413. Historical context for Galton's work.\n- Senn, S. J. (2011). \"Francis Galton and regression to the mean.\" *Significance*, 8(3), 124–126. Modern statistical interpretation.\n- Mlodinow, L. (2008). *The Drunkard's Walk: How Randomness Rules Our Lives.* Pantheon. ISBN 978-0307275172. Popular treatment.\n- Tversky, A., & Kahneman, D. (1974). \"Judgment under Uncertainty: Heuristics and Biases.\" *Science*, 185, 1124–1131. The cognitive-bias treatment.\n- Campbell, D. T., & Kenny, D. A. (1999). *A Primer on Regression Artifacts.* Guilford. ISBN 978-1572304826.\n\nFile v1.0.3:examples/galton-1886-kahneman-2011-israeli-air-force.md\n\n# Method in Action: Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force\n\n> *Example for the [regression-to-the-mean](../SKILL.md) skill.*\n\n**Francis Galton** (1822–1911) was a Victorian polymath: cousin of Charles Darwin, founder of the field of biometrics, and (less savorily) the founder of eugenics. His 1886 paper documenting the regression phenomenon was a methodological breakthrough. He had collected height data on 928 adult children and 205 sets of parents at the International Health Exhibition in London (1884) — a dataset that was, for its time, extraordinarily large and well-controlled.\n\nGalton's empirical finding was the contraction toward the mean across generations. Tall parents had children averaging shorter than themselves but still taller than the population mean. The contraction was systematic. Galton's framing — \"regression towards mediocrity\" — has stuck, though his interpretation (that a \"force\" was pulling toward mediocrity) is now understood to be wrong.\n\n> \"The law of Regression towards Mediocrity that I have proposed is precisely this: Every peculiarity in a man is shared by his kindred, but on the average in a less degree. ... I find that the height of an adult offspring deviates from the mid-parent height by a fraction (about two-thirds) of the deviation of the mid-parents from the average.\"\n>\n> — Galton (1886), pp. 247, 252.\n\nThe correct interpretation, developed in the 20th century, is that any noisy measurement of a stable underlying signal will produce extreme observations partly via lucky/unlucky noise. Retest will not repeat the noise; the underlying signal is preserved but the noise component regresses.\n\n**Kahneman's Israeli Air Force example (2011, chapter 17).** This is the most influential operational illustration of the regression artifact. Kahneman was teaching Israeli Air Force flight instructors about behavioral psychology. He emphasized that positive reinforcement (praise) is more effective than punishment in training, citing established psychological research.\n\nAn instructor (whom Kahneman described as senior and respected) pushed back:\n\n> \"On many occasions I have praised flight cadets for clean execution of some aerobatic maneuver. The next time they try the same maneuver they usually do worse. On the other hand, I have often screamed into a cadet's earphone for bad execution, and in general he does better on his next try. So please don't tell us that reinforcement works and punishment does not, because the opposite is the case.\"\n>\n> — quoted in Kahneman (2011), p. 175.\n\nKahneman's response was the breakthrough: the instructor was observing regression to the mean, mistaken for a causal effect of praise/criticism.\n\nWhen a cadet executes a maneuver exceptionally well, that performance includes a positive noise component (lucky day, favorable winds, instructor's attention, etc.). On the next attempt, the noise component will be different (likely less favorable); the underlying skill is the same; the next performance is statistically expected to be worse — regardless of whether the instructor praised or did nothing.\n\nWhen a cadet executes badly, the performance includes a negative noise component. On the next attempt, the noise is different; the next performance is statistically expected to be better — regardless of whether the instructor screamed or did nothing.\n\nThe instructor was reading: praise → worse performance, criticism → better performance. The truth was: extreme observations → regression to the mean. The instructor's causal inference was driven by a statistical artifact.\n\nKahneman's framing:\n\n> \"Because we tend to reward others when they do well and punish them when they do badly, and because there is regression to the mean, it is part of the human condition that we are statistically punished for rewarding others and rewarded for punishing them.\"\n>\n> — Kahneman (2011), p. 175.\n\nThis is generalizable. Many domains exhibit the same pattern:\n\n**Sales performance.** A salesperson who blew out a quarter gets a bonus. Next quarter they perform less well; manager wonders if the bonus de-motivated them. A salesperson who had a terrible quarter gets a stern talk. Next quarter they perform better; manager thinks the talk worked. Both inferences are likely regression.\n\n**Sports coaching.** A breakout rookie season gets praise; sophomore year is worse — coaches wonder what went wrong. A struggling player gets benched; on return they perform better — coaches credit the benching. Both partly regression.\n\n**Investment fund manager evaluation.** A fund manager who beat the market gets capital inflows; next period they underperform — LPs feel duped. A fund manager who underperformed gets fired; the replacement does better — board credits the replacement. Both partly regression.\n\n**Acquisition targets.** Companies are typically acquired near peak performance. Post-acquisition, performance regresses. The acquirer is disappointed; narrative blames integration. Often: partly regression.\n\n**Patient improvement after medical intervention.** Patients seek treatment when symptoms are most extreme. Symptoms regress regardless of treatment. The patient credits the treatment.\n\n**Customer satisfaction interventions.** A team with low satisfaction gets training; satisfaction improves. Without a control, the improvement could be mostly regression.\n\nThree operational lessons:\n\n**First, use control groups or baseline regression estimates.** When evaluating interventions, the question is not \"did the metric change?\" but \"did the metric change more than it would have changed by regression alone?\" The control group reveals the regression baseline.\n\n**Second, watch for \"praise reduces performance, criticism improves it\" patterns.** When you find yourself thinking that punishment works and praise doesn't, you are almost certainly observing regression artifact. Update your hiring, management, and intervention conclusions accordingly.\n\n**Third, beware celebrating or punishing single-period extreme performance.** A single-quarter top performer is partly noise; promoting them based on one quarter will partly regret. A single-quarter struggler is partly noise; firing them based on one quarter will partly regret. Use longer windows to reduce noise.\n\nFile v1.0.3:skill-card.md\n\n## Description: <br>\nHelps agents evaluate noisy before-and-after performance changes by separating regression to the mean from claimed intervention effects. <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, operators, managers, analysts, and developers use this skill to evaluate interventions on extreme performers, A/B tests, pilots, hiring or promotion decisions, investment performance, and other before-and-after changes where random noise may explain movement toward average. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: Users may treat business, investment, medical, or management examples as professional advice or proof that an intervention worked. <br>\nMitigation: Use the skill as analytic framing, then apply real controls, baseline data, and qualified review before making consequential decisions. <br>\nRisk: Before-and-after changes for extreme performers can be misattributed to coaching, bonuses, firing, or other interventions. <br>\nMitigation: Compare against an untreated control group or an explicit baseline regression estimate before assigning causal credit or blame. <br>\n\n\n## Reference(s): <br>\n- [Regression to the Mean on ClawHub](https://clawhub.ai/deciqai/skills/regression-to-the-mean) <br>\n- [Primary sources for regression to the mean](references/sources.md) <br>\n- [Galton 1886 and Kahneman 2011 example](examples/galton-1886-kahneman-2011-israeli-air-force.md) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [text, markdown, guidance] <br>\n**Output Format:** [Markdown regression analysis with structured fields and coaching questions] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [May ask step-by-step questions and wait for user input in coach mode.] <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: 5 files, 8996 bytes\n\nFiles: examples/galton-1886-kahneman-2011-israeli-air-force.md (6308b), references/sources.md (1091b), skill-card.md (2220b), SKILL.md (8212b), _meta.json (141b)\n\nFile v1.0.2:SKILL.md\n\n---\nname: regression-to-the-mean\ndescription: \"Activate when: someone says 'things always bounce back,' wonders why an intervention seemed to work on a struggling team, is surprised their star performer regressed, is evaluating whether a coaching/bonus/firing had an effect, or is interpreting before-vs-after performance changes without a control group. Do NOT activate when: the measurement is genuinely noise-free (e.g., deterministic process outputs); the underlying signal has demonstrably changed due to a structural shift (new product launch, technology change).\"\n---\n\n# Regression to the Mean\n\n## Overview\n\n**Regression to the mean** is the statistical regularity that any noisy measurement producing an extreme value tends to be followed on retest by a less-extreme value — because the extreme portion was partly driven by non-repeating random noise. There is no \"force pulling back to average\"; it is a mathematical consequence of signal + noise structure.\n\nNamed by Francis Galton (1886) studying parent-child height: tall parents have tall children, but slightly shorter; short parents have short children, but slightly taller. Kahneman's Israeli Air Force example (2011, Ch. 17) is the most-cited operational case — flight instructors concluded punishment works and praise doesn't, but were observing regression, not causation.\n\nComposes with `survivorship-bias` (extreme survivors regress), `probabilistic-thinking` (regression is probabilistic), `narrative-fallacy` (regression drives post-hoc narratives), `fundamental-attribution-error` (attributing regression to character/intervention is FAE).\n\n## When to Use\n\n- Evaluating the effect of an intervention on extreme performers (struggling teams, top sales reps, low-rated branches)\n- Designing or interpreting A/B tests or pilot programs\n- Reviewing year-over-year performance changes\n- Hiring or promoting top performers\n- Evaluating investment fund performance\n- Analyzing acquisition outcomes\n- Building or critiquing causal claims about training, coaching, or feedback\n- Someone says \"regression to the mean,\" \"things will average out,\" \"they always come back\"\n\n**Not when:** the measurement is noise-free (rare in business); the underlying signal is genuinely changing (e.g., the business model fundamentally improved); the intervention is so substantial that no plausible regression can explain the effect.\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user brings a specific performance change → run The Process directly.\n- **Coach mode:** user is new to the framework → 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:** when an extreme performance is followed by a less-extreme one, regression to the mean is the default explanation — not the intervention.\n2. **Check fit.** Did we observe an extreme value followed by retest? If yes, regression applies.\n3. **Elicit the structure.** What was the extreme observation? What was the follow-up? What intervention happened in between? Is there a control or baseline?\n> **[WAIT — do not advance until user responds]**\n4. **One question at a time:** how much regression would we expect without the intervention? Does the observed change exceed it? What's the noise level in the measurement?\n> **[WAIT — do not advance until user responds]**\n5. **Close:** regression vs. causal effect separated + control or baseline identified + claim calibrated.\n> **[WAIT — do not advance until user responds]**\n\n## The Process\n\n**1. Identify the extreme observation** — value, subject, how extreme, any intervention.\n**2. Identify the retest** — follow-up measurement, period, direction (toward mean?).\n**3. Estimate expected regression** — noise level, historical variance; formula: `regression ≈ (1 − reliability) × distance from mean`.\n**4. Compare to control** — untreated extreme performers; intervention effect = treatment change − control change. Without control, regression cannot be ruled out.\n**5. Calibrate causal claim** — did the change exceed the regression baseline? By how much? Confidence?\n**6. Adjust action** — credit/blame intervention only if it exceeds regression; recommend control structure for future tests.\n\n## Output: Regression Analysis\n\n```markdown\n# Regression Analysis: <observed change>\nExtreme observation — Subject: | Value: | Period:\nIntervention — What: | When: | By whom:\nRetest — Follow-up value: | Toward mean (Y/N): | Magnitude:\nExpected regression — Noise level: | Predicted magnitude:\nControl — Group: | Change: | Intervention effect = treatment − control:\nCausal calibration — % regression: | % intervention: | Confidence:\nAdjusted action — Credit/blame verdict + future evaluation process:\n```\n\n*→ Method in Action: [Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force](examples/galton-1886-kahneman-2011-israeli-air-force.md)*\n\n## Pack: Regression Across Domains\n\n| Domain | Extreme obs. | Regression artifact | Discipline |\n|---|---|---|---|\n| Sales | Top/worst quarter | Regresses next quarter | 4-quarter averages + control |\n| Athlete | Breakout season | Sophomore slump | Multi-season average |\n| Stock fund | Hot year | Underperforms next year | 5-year track record |\n| Customer sat. | Low quarter | Improvement next quarter | Control group |\n| Manager intervention | Struggling team improves | Mostly regression | Compare untreated peer |\n\n## Applying It Well\n\n- Use longer time windows to reduce noise (yearly vs. quarterly)\n- Always ask: \"what would regression alone predict?\" before crediting an intervention\n- Control groups are not optional when evaluating interventions on extreme performers\n- \"Praise hurts, punishment helps\" is the signature cognitive error of regression blindness\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] \"The bonus motivated her / criticism got through to him\" | Exceptional quarter → regresses; bad quarter → improves. Both are regression, not intervention effect. |\n| [D] \"Our turnaround plan / new CEO is working\" | Trough is partly noise. Compare to untreated peers before crediting the plan or leader. |\n| [D] \"This pilot is working on our struggling teams\" | Extreme performers are the worst pilot group — regression is the null. Use a control. |\n| [D] \"Star performers will keep starring / sophomore slump is real\" | Single-period peaks include noise; regression is the default on retest, not a slump. |\n| [D] \"After we fired him the team improved\" | Could be regression. Disentangle by comparing to teams with no change. |\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- An intervention is being evaluated on extreme performers without a control group\n- A single-period change is being treated as evidence of a stable trend\n- Causal credit/blame is being assigned to interventions that coincided with regression\n- \"The intervention worked\" claims lack baseline regression estimates\n- Promotion / firing decisions are based on single-period extreme performance\n- \"Comeback\" narratives credit decisive action, not regression\n\n## Verification\n\n- [ ] Extreme observation identified; retest compared to it\n- [ ] Noise level and regression magnitude estimated\n- [ ] Control group (or baseline regression expectation) used\n- [ ] Causal effect separated from regression; claim calibrated\n- [ ] Long-window measurements used where possible; future evaluation process set\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\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/regression-to-the-mean?utm_source=clawhub&utm_medium=marketplace&utm_campaign=knowledge-skills&utm_content=regression-to-the-mean** · ⭐ 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\": \"regression-to-the-mean\",\n  \"version\": \"1.0.2\",\n  \"publishedAt\": 1783472480944\n}\n\nFile v1.0.2:references/sources.md\n\n# Sources — regression-to-the-mean\n\n> *Primary sources for the [regression-to-the-mean](../SKILL.md) skill.*\n\n- Galton, F. (1886). \"Regression towards Mediocrity in Hereditary Stature.\" *Journal of the Anthropological Institute*, 15, 246–263. The original paper.\n- Kahneman, D. (2011). *Thinking, Fast and Slow.* Farrar, Straus and Giroux. ISBN 978-0374533557. Chapter 17, \"Regression to the Mean.\"\n- Stigler, S. M. (1986). *The History of Statistics: The Measurement of Uncertainty before 1900.* Belknap. ISBN 978-0674403413. Historical context for Galton's work.\n- Senn, S. J. (2011). \"Francis Galton and regression to the mean.\" *Significance*, 8(3), 124–126. Modern statistical interpretation.\n- Mlodinow, L. (2008). *The Drunkard's Walk: How Randomness Rules Our Lives.* Pantheon. ISBN 978-0307275172. Popular treatment.\n- Tversky, A., & Kahneman, D. (1974). \"Judgment under Uncertainty: Heuristics and Biases.\" *Science*, 185, 1124–1131. The cognitive-bias treatment.\n- Campbell, D. T., & Kenny, D. A. (1999). *A Primer on Regression Artifacts.* Guilford. ISBN 978-1572304826.\n\nFile v1.0.2:examples/galton-1886-kahneman-2011-israeli-air-force.md\n\n# Method in Action: Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force\n\n> *Example for the [regression-to-the-mean](../SKILL.md) skill.*\n\n**Francis Galton** (1822–1911) was a Victorian polymath: cousin of Charles Darwin, founder of the field of biometrics, and (less savorily) the founder of eugenics. His 1886 paper documenting the regression phenomenon was a methodological breakthrough. He had collected height data on 928 adult children and 205 sets of parents at the International Health Exhibition in London (1884) — a dataset that was, for its time, extraordinarily large and well-controlled.\n\nGalton's empirical finding was the contraction toward the mean across generations. Tall parents had children averaging shorter than themselves but still taller than the population mean. The contraction was systematic. Galton's framing — \"regression towards mediocrity\" — has stuck, though his interpretation (that a \"force\" was pulling toward mediocrity) is now understood to be wrong.\n\n> \"The law of Regression towards Mediocrity that I have proposed is precisely this: Every peculiarity in a man is shared by his kindred, but on the average in a less degree. ... I find that the height of an adult offspring deviates from the mid-parent height by a fraction (about two-thirds) of the deviation of the mid-parents from the average.\"\n>\n> — Galton (1886), pp. 247, 252.\n\nThe correct interpretation, developed in the 20th century, is that any noisy measurement of a stable underlying signal will produce extreme observations partly via lucky/unlucky noise. Retest will not repeat the noise; the underlying signal is preserved but the noise component regresses.\n\n**Kahneman's Israeli Air Force example (2011, chapter 17).** This is the most influential operational illustration of the regression artifact. Kahneman was teaching Israeli Air Force flight instructors about behavioral psychology. He emphasized that positive reinforcement (praise) is more effective than punishment in training, citing established psychological research.\n\nAn instructor (whom Kahneman described as senior and respected) pushed back:\n\n> \"On many occasions I have praised flight cadets for clean execution of some aerobatic maneuver. The next time they try the same maneuver they usually do worse. On the other hand, I have often screamed into a cadet's earphone for bad execution, and in general he does better on his next try. So please don't tell us that reinforcement works and punishment does not, because the opposite is the case.\"\n>\n> — quoted in Kahneman (2011), p. 175.\n\nKahneman's response was the breakthrough: the instructor was observing regression to the mean, mistaken for a causal effect of praise/criticism.\n\nWhen a cadet executes a maneuver exceptionally well, that performance includes a positive noise component (lucky day, favorable winds, instructor's attention, etc.). On the next attempt, the noise component will be different (likely less favorable); the underlying skill is the same; the next performance is statistically expected to be worse — regardless of whether the instructor praised or did nothing.\n\nWhen a cadet executes badly, the performance includes a negative noise component. On the next attempt, the noise is different; the next performance is statistically expected to be better — regardless of whether the instructor screamed or did nothing.\n\nThe instructor was reading: praise → worse performance, criticism → better performance. The truth was: extreme observations → regression to the mean. The instructor's causal inference was driven by a statistical artifact.\n\nKahneman's framing:\n\n> \"Because we tend to reward others when they do well and punish them when they do badly, and because there is regression to the mean, it is part of the human condition that we are statistically punished for rewarding others and rewarded for punishing them.\"\n>\n> — Kahneman (2011), p. 175.\n\nThis is generalizable. Many domains exhibit the same pattern:\n\n**Sales performance.** A salesperson who blew out a quarter gets a bonus. Next quarter they perform less well; manager wonders if the bonus de-motivated them. A salesperson who had a terrible quarter gets a stern talk. Next quarter they perform better; manager thinks the talk worked. Both inferences are likely regression.\n\n**Sports coaching.** A breakout rookie season gets praise; sophomore year is worse — coaches wonder what went wrong. A struggling player gets benched; on return they perform better — coaches credit the benching. Both partly regression.\n\n**Investment fund manager evaluation.** A fund manager who beat the market gets capital inflows; next period they underperform — LPs feel duped. A fund manager who underperformed gets fired; the replacement does better — board credits the replacement. Both partly regression.\n\n**Acquisition targets.** Companies are typically acquired near peak performance. Post-acquisition, performance regresses. The acquirer is disappointed; narrative blames integration. Often: partly regression.\n\n**Patient improvement after medical intervention.** Patients seek treatment when symptoms are most extreme. Symptoms regress regardless of treatment. The patient credits the treatment.\n\n**Customer satisfaction interventions.** A team with low satisfaction gets training; satisfaction improves. Without a control, the improvement could be mostly regression.\n\nThree operational lessons:\n\n**First, use control groups or baseline regression estimates.** When evaluating interventions, the question is not \"did the metric change?\" but \"did the metric change more than it would have changed by regression alone?\" The control group reveals the regression baseline.\n\n**Second, watch for \"praise reduces performance, criticism improves it\" patterns.** When you find yourself thinking that punishment works and praise doesn't, you are almost certainly observing regression artifact. Update your hiring, management, and intervention conclusions accordingly.\n\n**Third, beware celebrating or punishing single-period extreme performance.** A single-quarter top performer is partly noise; promoting them based on one quarter will partly regret. A single-quarter struggler is partly noise; firing them based on one quarter will partly regret. Use longer windows to reduce noise.\n\nFile v1.0.2:skill-card.md\n\n## Description: <br>\nHelps agents analyze whether changes after extreme noisy observations are regression to the mean or credible causal effects. <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 developers use this skill to test causal claims about interventions, performance changes, and before-vs-after comparisons against an expected regression baseline. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: Users may treat statistical guidance as a definitive causal conclusion in hiring, firing, medical, financial, or business decisions. <br>\nMitigation: Use the skill as decision support and verify important conclusions with appropriate controls or domain expertise. <br>\nRisk: Regression to the mean may be over-applied when measurements are noise-free or the underlying system has structurally changed. <br>\nMitigation: Check whether the measurement is noisy, estimate expected regression, and compare against a control group or baseline before calibrating the causal claim. <br>\n\n\n## Reference(s): <br>\n- [ClawHub skill listing](https://clawhub.ai/deciqai/skills/regression-to-the-mean) <br>\n- [Primary sources](references/sources.md) <br>\n- [Method in Action: Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force](examples/galton-1886-kahneman-2011-israeli-air-force.md) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [Text, Markdown, Guidance] <br>\n**Output Format:** [Markdown regression analysis with structured fields and calibrated causal guidance] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Non-executable knowledge guidance; no tools, API calls, credentials, or system changes.] <br>\n\n## Skill Version(s): <br>\n1.0.2 (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 v1.0.1: 5 files, 8990 bytes\n\nFiles: examples/galton-1886-kahneman-2011-israeli-air-force.md (6308b), references/sources.md (1091b), skill-card.md (2430b), SKILL.md (8089b), _meta.json (141b)\n\nFile v1.0.1:SKILL.md\n\n---\nname: regression-to-the-mean\ndescription: \"Activate when: someone says 'things always bounce back,' wonders why an intervention seemed to work on a struggling team, is surprised their star performer regressed, is evaluating whether a coaching/bonus/firing had an effect, or is interpreting before-vs-after performance changes without a control group. Do NOT activate when: the measurement is genuinely noise-free (e.g., deterministic process outputs); the underlying signal has demonstrably changed due to a structural shift (new product launch, technology change).\"\n---\n\n# Regression to the Mean\n\n## Overview\n\n**Regression to the mean** is the statistical regularity that any noisy measurement producing an extreme value tends to be followed on retest by a less-extreme value — because the extreme portion was partly driven by non-repeating random noise. There is no \"force pulling back to average\"; it is a mathematical consequence of signal + noise structure.\n\nNamed by Francis Galton (1886) studying parent-child height: tall parents have tall children, but slightly shorter; short parents have short children, but slightly taller. Kahneman's Israeli Air Force example (2011, Ch. 17) is the most-cited operational case — flight instructors concluded punishment works and praise doesn't, but were observing regression, not causation.\n\nComposes with [`survivorship-bias`](../survivorship-bias/SKILL.md) (extreme survivors regress), [`probabilistic-thinking`](../probabilistic-thinking/SKILL.md) (regression is probabilistic), [`narrative-fallacy`](../narrative-fallacy/SKILL.md) (regression drives post-hoc narratives), [`fundamental-attribution-error`](../fundamental-attribution-error/SKILL.md) (attributing regression to character/intervention is FAE).\n\n## When to Use\n\n- Evaluating the effect of an intervention on extreme performers (struggling teams, top sales reps, low-rated branches)\n- Designing or interpreting A/B tests or pilot programs\n- Reviewing year-over-year performance changes\n- Hiring or promoting top performers\n- Evaluating investment fund performance\n- Analyzing acquisition outcomes\n- Building or critiquing causal claims about training, coaching, or feedback\n- Someone says \"regression to the mean,\" \"things will average out,\" \"they always come back\"\n\n**Not when:** the measurement is noise-free (rare in business); the underlying signal is genuinely changing (e.g., the business model fundamentally improved); the intervention is so substantial that no plausible regression can explain the effect.\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user brings a specific performance change → run The Process directly.\n- **Coach mode:** user is new to the framework → 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:** when an extreme performance is followed by a less-extreme one, regression to the mean is the default explanation — not the intervention.\n2. **Check fit.** Did we observe an extreme value followed by retest? If yes, regression applies.\n3. **Elicit the structure.** What was the extreme observation? What was the follow-up? What intervention happened in between? Is there a control or baseline?\n> **[WAIT — do not advance until user responds]**\n4. **One question at a time:** how much regression would we expect without the intervention? Does the observed change exceed it? What's the noise level in the measurement?\n> **[WAIT — do not advance until user responds]**\n5. **Close:** regression vs. causal effect separated + control or baseline identified + claim calibrated.\n> **[WAIT — do not advance until user responds]**\n\n## The Process\n\n**1. Identify the extreme observation** — value, subject, how extreme, any intervention.\n**2. Identify the retest** — follow-up measurement, period, direction (toward mean?).\n**3. Estimate expected regression** — noise level, historical variance; formula: `regression ≈ (1 − reliability) × distance from mean`.\n**4. Compare to control** — untreated extreme performers; intervention effect = treatment change − control change. Without control, regression cannot be ruled out.\n**5. Calibrate causal claim** — did the change exceed the regression baseline? By how much? Confidence?\n**6. Adjust action** — credit/blame intervention only if it exceeds regression; recommend control structure for future tests.\n\n## Output: Regression Analysis\n\n```markdown\n# Regression Analysis: <observed change>\nExtreme observation — Subject: | Value: | Period:\nIntervention — What: | When: | By whom:\nRetest — Follow-up value: | Toward mean (Y/N): | Magnitude:\nExpected regression — Noise level: | Predicted magnitude:\nControl — Group: | Change: | Intervention effect = treatment − control:\nCausal calibration — % regression: | % intervention: | Confidence:\nAdjusted action — Credit/blame verdict + future evaluation process:\n```\n\n*→ Method in Action: [Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force](examples/galton-1886-kahneman-2011-israeli-air-force.md)*\n\n## Pack: Regression Across Domains\n\n| Domain | Extreme obs. | Regression artifact | Discipline |\n|---|---|---|---|\n| Sales | Top/worst quarter | Regresses next quarter | 4-quarter averages + control |\n| Athlete | Breakout season | Sophomore slump | Multi-season average |\n| Stock fund | Hot year | Underperforms next year | 5-year track record |\n| Customer sat. | Low quarter | Improvement next quarter | Control group |\n| Manager intervention | Struggling team improves | Mostly regression | Compare untreated peer |\n\n## Applying It Well\n\n- Use longer time windows to reduce noise (yearly vs. quarterly)\n- Always ask: \"what would regression alone predict?\" before crediting an intervention\n- Control groups are not optional when evaluating interventions on extreme performers\n- \"Praise hurts, punishment helps\" is the signature cognitive error of regression blindness\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] \"The bonus motivated her / criticism got through to him\" | Exceptional quarter → regresses; bad quarter → improves. Both are regression, not intervention effect. |\n| [D] \"Our turnaround plan / new CEO is working\" | Trough is partly noise. Compare to untreated peers before crediting the plan or leader. |\n| [D] \"This pilot is working on our struggling teams\" | Extreme performers are the worst pilot group — regression is the null. Use a control. |\n| [D] \"Star performers will keep starring / sophomore slump is real\" | Single-period peaks include noise; regression is the default on retest, not a slump. |\n| [D] \"After we fired him the team improved\" | Could be regression. Disentangle by comparing to teams with no change. |\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- An intervention is being evaluated on extreme performers without a control group\n- A single-period change is being treated as evidence of a stable trend\n- Causal credit/blame is being assigned to interventions that coincided with regression\n- \"The intervention worked\" claims lack baseline regression estimates\n- Promotion / firing decisions are based on single-period extreme performance\n- \"Comeback\" narratives credit decisive action, not regression\n\n## Verification\n\n- [ ] Extreme observation identified; retest compared to it\n- [ ] Noise level and regression magnitude estimated\n- [ ] Control group (or baseline regression expectation) used\n- [ ] Causal effect separated from regression; claim calibrated\n- [ ] Long-window measurements used where possible; future evaluation process set\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\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\": \"regression-to-the-mean\",\n  \"version\": \"1.0.1\",\n  \"publishedAt\": 1783463565333\n}\n\nFile v1.0.1:references/sources.md\n\n# Sources — regression-to-the-mean\n\n> *Primary sources for the [regression-to-the-mean](../SKILL.md) skill.*\n\n- Galton, F. (1886). \"Regression towards Mediocrity in Hereditary Stature.\" *Journal of the Anthropological Institute*, 15, 246–263. The original paper.\n- Kahneman, D. (2011). *Thinking, Fast and Slow.* Farrar, Straus and Giroux. ISBN 978-0374533557. Chapter 17, \"Regression to the Mean.\"\n- Stigler, S. M. (1986). *The History of Statistics: The Measurement of Uncertainty before 1900.* Belknap. ISBN 978-0674403413. Historical context for Galton's work.\n- Senn, S. J. (2011). \"Francis Galton and regression to the mean.\" *Significance*, 8(3), 124–126. Modern statistical interpretation.\n- Mlodinow, L. (2008). *The Drunkard's Walk: How Randomness Rules Our Lives.* Pantheon. ISBN 978-0307275172. Popular treatment.\n- Tversky, A., & Kahneman, D. (1974). \"Judgment under Uncertainty: Heuristics and Biases.\" *Science*, 185, 1124–1131. The cognitive-bias treatment.\n- Campbell, D. T., & Kenny, D. A. (1999). *A Primer on Regression Artifacts.* Guilford. ISBN 978-1572304826.\n\nFile v1.0.1:examples/galton-1886-kahneman-2011-israeli-air-force.md\n\n# Method in Action: Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force\n\n> *Example for the [regression-to-the-mean](../SKILL.md) skill.*\n\n**Francis Galton** (1822–1911) was a Victorian polymath: cousin of Charles Darwin, founder of the field of biometrics, and (less savorily) the founder of eugenics. His 1886 paper documenting the regression phenomenon was a methodological breakthrough. He had collected height data on 928 adult children and 205 sets of parents at the International Health Exhibition in London (1884) — a dataset that was, for its time, extraordinarily large and well-controlled.\n\nGalton's empirical finding was the contraction toward the mean across generations. Tall parents had children averaging shorter than themselves but still taller than the population mean. The contraction was systematic. Galton's framing — \"regression towards mediocrity\" — has stuck, though his interpretation (that a \"force\" was pulling toward mediocrity) is now understood to be wrong.\n\n> \"The law of Regression towards Mediocrity that I have proposed is precisely this: Every peculiarity in a man is shared by his kindred, but on the average in a less degree. ... I find that the height of an adult offspring deviates from the mid-parent height by a fraction (about two-thirds) of the deviation of the mid-parents from the average.\"\n>\n> — Galton (1886), pp. 247, 252.\n\nThe correct interpretation, developed in the 20th century, is that any noisy measurement of a stable underlying signal will produce extreme observations partly via lucky/unlucky noise. Retest will not repeat the noise; the underlying signal is preserved but the noise component regresses.\n\n**Kahneman's Israeli Air Force example (2011, chapter 17).** This is the most influential operational illustration of the regression artifact. Kahneman was teaching Israeli Air Force flight instructors about behavioral psychology. He emphasized that positive reinforcement (praise) is more effective than punishment in training, citing established psychological research.\n\nAn instructor (whom Kahneman described as senior and respected) pushed back:\n\n> \"On many occasions I have praised flight cadets for clean execution of some aerobatic maneuver. The next time they try the same maneuver they usually do worse. On the other hand, I have often screamed into a cadet's earphone for bad execution, and in general he does better on his next try. So please don't tell us that reinforcement works and punishment does not, because the opposite is the case.\"\n>\n> — quoted in Kahneman (2011), p. 175.\n\nKahneman's response was the breakthrough: the instructor was observing regression to the mean, mistaken for a causal effect of praise/criticism.\n\nWhen a cadet executes a maneuver exceptionally well, that performance includes a positive noise component (lucky day, favorable winds, instructor's attention, etc.). On the next attempt, the noise component will be different (likely less favorable); the underlying skill is the same; the next performance is statistically expected to be worse — regardless of whether the instructor praised or did nothing.\n\nWhen a cadet executes badly, the performance includes a negative noise component. On the next attempt, the noise is different; the next performance is statistically expected to be better — regardless of whether the instructor screamed or did nothing.\n\nThe instructor was reading: praise → worse performance, criticism → better performance. The truth was: extreme observations → regression to the mean. The instructor's causal inference was driven by a statistical artifact.\n\nKahneman's framing:\n\n> \"Because we tend to reward others when they do well and punish them when they do badly, and because there is regression to the mean, it is part of the human condition that we are statistically punished for rewarding others and rewarded for punishing them.\"\n>\n> — Kahneman (2011), p. 175.\n\nThis is generalizable. Many domains exhibit the same pattern:\n\n**Sales performance.** A salesperson who blew out a quarter gets a bonus. Next quarter they perform less well; manager wonders if the bonus de-motivated them. A salesperson who had a terrible quarter gets a stern talk. Next quarter they perform better; manager thinks the talk worked. Both inferences are likely regression.\n\n**Sports coaching.** A breakout rookie season gets praise; sophomore year is worse — coaches wonder what went wrong. A struggling player gets benched; on return they perform better — coaches credit the benching. Both partly regression.\n\n**Investment fund manager evaluation.** A fund manager who beat the market gets capital inflows; next period they underperform — LPs feel duped. A fund manager who underperformed gets fired; the replacement does better — board credits the replacement. Both partly regression.\n\n**Acquisition targets.** Companies are typically acquired near peak performance. Post-acquisition, performance regresses. The acquirer is disappointed; narrative blames integration. Often: partly regression.\n\n**Patient improvement after medical intervention.** Patients seek treatment when symptoms are most extreme. Symptoms regress regardless of treatment. The patient credits the treatment.\n\n**Customer satisfaction interventions.** A team with low satisfaction gets training; satisfaction improves. Without a control, the improvement could be mostly regression.\n\nThree operational lessons:\n\n**First, use control groups or baseline regression estimates.** When evaluating interventions, the question is not \"did the metric change?\" but \"did the metric change more than it would have changed by regression alone?\" The control group reveals the regression baseline.\n\n**Second, watch for \"praise reduces performance, criticism improves it\" patterns.** When you find yourself thinking that punishment works and praise doesn't, you are almost certainly observing regression artifact. Update your hiring, management, and intervention conclusions accordingly.\n\n**Third, beware celebrating or punishing single-period extreme performance.** A single-quarter top performer is partly noise; promoting them based on one quarter will partly regret. A single-quarter struggler is partly noise; firing them based on one quarter will partly regret. Use longer windows to reduce noise.\n\nFile v1.0.1:skill-card.md\n\n## Description: <br>\nHelps agents identify regression to the mean when interpreting extreme performance changes, interventions, and before-vs-after results without a control group. <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, analysts, and decision makers use this skill to test whether apparent improvements or declines are likely regression artifacts rather than causal effects. It is most useful for performance reviews, pilot programs, A/B test interpretation, hiring or promotion decisions, fund evaluation, and other noisy before-vs-after comparisons. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: Users may treat a regression-to-the-mean analysis as a substitute for domain evidence or controlled testing in consequential business, personnel, financial, or medical decisions. <br>\nMitigation: Use the skill to calibrate reasoning, then require control groups, baseline regression estimates, and domain-specific evidence before making consequential decisions. <br>\nRisk: The skill can help critique causal claims, but it cannot prove that an intervention had no effect. <br>\nMitigation: Frame conclusions as calibrated uncertainty and compare observed changes against untreated controls or historical variance. <br>\n\n\n## Reference(s): <br>\n- [Sources - regression-to-the-mean](references/sources.md) <br>\n- [Method in Action: Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force](examples/galton-1886-kahneman-2011-israeli-air-force.md) <br>\n- [ClawHub skill page](https://clawhub.ai/deciqai/skills/regression-to-the-mean) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [text, markdown, guidance] <br>\n**Output Format:** [Markdown analysis template and stepwise guidance] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [May ask one coaching question at a time when the user is new to the framework.] <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, 9015 bytes\n\nFiles: examples/galton-1886-kahneman-2011-israeli-air-force.md (6308b), references/sources.md (1091b), skill-card.md (2534b), SKILL.md (8089b), _meta.json (141b)\n\nFile v1.0.0:SKILL.md\n\n---\nname: regression-to-the-mean\ndescription: \"Activate when: someone says 'things always bounce back,' wonders why an intervention seemed to work on a struggling team, is surprised their star performer regressed, is evaluating whether a coaching/bonus/firing had an effect, or is interpreting before-vs-after performance changes without a control group. Do NOT activate when: the measurement is genuinely noise-free (e.g., deterministic process outputs); the underlying signal has demonstrably changed due to a structural shift (new product launch, technology change).\"\n---\n\n# Regression to the Mean\n\n## Overview\n\n**Regression to the mean** is the statistical regularity that any noisy measurement producing an extreme value tends to be followed on retest by a less-extreme value — because the extreme portion was partly driven by non-repeating random noise. There is no \"force pulling back to average\"; it is a mathematical consequence of signal + noise structure.\n\nNamed by Francis Galton (1886) studying parent-child height: tall parents have tall children, but slightly shorter; short parents have short children, but slightly taller. Kahneman's Israeli Air Force example (2011, Ch. 17) is the most-cited operational case — flight instructors concluded punishment works and praise doesn't, but were observing regression, not causation.\n\nComposes with [`survivorship-bias`](../survivorship-bias/SKILL.md) (extreme survivors regress), [`probabilistic-thinking`](../probabilistic-thinking/SKILL.md) (regression is probabilistic), [`narrative-fallacy`](../narrative-fallacy/SKILL.md) (regression drives post-hoc narratives), [`fundamental-attribution-error`](../fundamental-attribution-error/SKILL.md) (attributing regression to character/intervention is FAE).\n\n## When to Use\n\n- Evaluating the effect of an intervention on extreme performers (struggling teams, top sales reps, low-rated branches)\n- Designing or interpreting A/B tests or pilot programs\n- Reviewing year-over-year performance changes\n- Hiring or promoting top performers\n- Evaluating investment fund performance\n- Analyzing acquisition outcomes\n- Building or critiquing causal claims about training, coaching, or feedback\n- Someone says \"regression to the mean,\" \"things will average out,\" \"they always come back\"\n\n**Not when:** the measurement is noise-free (rare in business); the underlying signal is genuinely changing (e.g., the business model fundamentally improved); the intervention is so substantial that no plausible regression can explain the effect.\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user brings a specific performance change → run The Process directly.\n- **Coach mode:** user is new to the framework → 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:** when an extreme performance is followed by a less-extreme one, regression to the mean is the default explanation — not the intervention.\n2. **Check fit.** Did we observe an extreme value followed by retest? If yes, regression applies.\n3. **Elicit the structure.** What was the extreme observation? What was the follow-up? What intervention happened in between? Is there a control or baseline?\n> **[WAIT — do not advance until user responds]**\n4. **One question at a time:** how much regression would we expect without the intervention? Does the observed change exceed it? What's the noise level in the measurement?\n> **[WAIT — do not advance until user responds]**\n5. **Close:** regression vs. causal effect separated + control or baseline identified + claim calibrated.\n> **[WAIT — do not advance until user responds]**\n\n## The Process\n\n**1. Identify the extreme observation** — value, subject, how extreme, any intervention.\n**2. Identify the retest** — follow-up measurement, period, direction (toward mean?).\n**3. Estimate expected regression** — noise level, historical variance; formula: `regression ≈ (1 − reliability) × distance from mean`.\n**4. Compare to control** — untreated extreme performers; intervention effect = treatment change − control change. Without control, regression cannot be ruled out.\n**5. Calibrate causal claim** — did the change exceed the regression baseline? By how much? Confidence?\n**6. Adjust action** — credit/blame intervention only if it exceeds regression; recommend control structure for future tests.\n\n## Output: Regression Analysis\n\n```markdown\n# Regression Analysis: <observed change>\nExtreme observation — Subject: | Value: | Period:\nIntervention — What: | When: | By whom:\nRetest — Follow-up value: | Toward mean (Y/N): | Magnitude:\nExpected regression — Noise level: | Predicted magnitude:\nControl — Group: | Change: | Intervention effect = treatment − control:\nCausal calibration — % regression: | % intervention: | Confidence:\nAdjusted action — Credit/blame verdict + future evaluation process:\n```\n\n*→ Method in Action: [Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force](examples/galton-1886-kahneman-2011-israeli-air-force.md)*\n\n## Pack: Regression Across Domains\n\n| Domain | Extreme obs. | Regression artifact | Discipline |\n|---|---|---|---|\n| Sales | Top/worst quarter | Regresses next quarter | 4-quarter averages + control |\n| Athlete | Breakout season | Sophomore slump | Multi-season average |\n| Stock fund | Hot year | Underperforms next year | 5-year track record |\n| Customer sat. | Low quarter | Improvement next quarter | Control group |\n| Manager intervention | Struggling team improves | Mostly regression | Compare untreated peer |\n\n## Applying It Well\n\n- Use longer time windows to reduce noise (yearly vs. quarterly)\n- Always ask: \"what would regression alone predict?\" before crediting an intervention\n- Control groups are not optional when evaluating interventions on extreme performers\n- \"Praise hurts, punishment helps\" is the signature cognitive error of regression blindness\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] \"The bonus motivated her / criticism got through to him\" | Exceptional quarter → regresses; bad quarter → improves. Both are regression, not intervention effect. |\n| [D] \"Our turnaround plan / new CEO is working\" | Trough is partly noise. Compare to untreated peers before crediting the plan or leader. |\n| [D] \"This pilot is working on our struggling teams\" | Extreme performers are the worst pilot group — regression is the null. Use a control. |\n| [D] \"Star performers will keep starring / sophomore slump is real\" | Single-period peaks include noise; regression is the default on retest, not a slump. |\n| [D] \"After we fired him the team improved\" | Could be regression. Disentangle by comparing to teams with no change. |\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- An intervention is being evaluated on extreme performers without a control group\n- A single-period change is being treated as evidence of a stable trend\n- Causal credit/blame is being assigned to interventions that coincided with regression\n- \"The intervention worked\" claims lack baseline regression estimates\n- Promotion / firing decisions are based on single-period extreme performance\n- \"Comeback\" narratives credit decisive action, not regression\n\n## Verification\n\n- [ ] Extreme observation identified; retest compared to it\n- [ ] Noise level and regression magnitude estimated\n- [ ] Control group (or baseline regression expectation) used\n- [ ] Causal effect separated from regression; claim calibrated\n- [ ] Long-window measurements used where possible; future evaluation process set\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\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\": \"regression-to-the-mean\",\n  \"version\": \"1.0.0\",\n  \"publishedAt\": 1782915497300\n}\n\nFile v1.0.0:references/sources.md\n\n# Sources — regression-to-the-mean\n\n> *Primary sources for the [regression-to-the-mean](../SKILL.md) skill.*\n\n- Galton, F. (1886). \"Regression towards Mediocrity in Hereditary Stature.\" *Journal of the Anthropological Institute*, 15, 246–263. The original paper.\n- Kahneman, D. (2011). *Thinking, Fast and Slow.* Farrar, Straus and Giroux. ISBN 978-0374533557. Chapter 17, \"Regression to the Mean.\"\n- Stigler, S. M. (1986). *The History of Statistics: The Measurement of Uncertainty before 1900.* Belknap. ISBN 978-0674403413. Historical context for Galton's work.\n- Senn, S. J. (2011). \"Francis Galton and regression to the mean.\" *Significance*, 8(3), 124–126. Modern statistical interpretation.\n- Mlodinow, L. (2008). *The Drunkard's Walk: How Randomness Rules Our Lives.* Pantheon. ISBN 978-0307275172. Popular treatment.\n- Tversky, A., & Kahneman, D. (1974). \"Judgment under Uncertainty: Heuristics and Biases.\" *Science*, 185, 1124–1131. The cognitive-bias treatment.\n- Campbell, D. T., & Kenny, D. A. (1999). *A Primer on Regression Artifacts.* Guilford. ISBN 978-1572304826.\n\nFile v1.0.0:examples/galton-1886-kahneman-2011-israeli-air-force.md\n\n# Method in Action: Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force\n\n> *Example for the [regression-to-the-mean](../SKILL.md) skill.*\n\n**Francis Galton** (1822–1911) was a Victorian polymath: cousin of Charles Darwin, founder of the field of biometrics, and (less savorily) the founder of eugenics. His 1886 paper documenting the regression phenomenon was a methodological breakthrough. He had collected height data on 928 adult children and 205 sets of parents at the International Health Exhibition in London (1884) — a dataset that was, for its time, extraordinarily large and well-controlled.\n\nGalton's empirical finding was the contraction toward the mean across generations. Tall parents had children averaging shorter than themselves but still taller than the population mean. The contraction was systematic. Galton's framing — \"regression towards mediocrity\" — has stuck, though his interpretation (that a \"force\" was pulling toward mediocrity) is now understood to be wrong.\n\n> \"The law of Regression towards Mediocrity that I have proposed is precisely this: Every peculiarity in a man is shared by his kindred, but on the average in a less degree. ... I find that the height of an adult offspring deviates from the mid-parent height by a fraction (about two-thirds) of the deviation of the mid-parents from the average.\"\n>\n> — Galton (1886), pp. 247, 252.\n\nThe correct interpretation, developed in the 20th century, is that any noisy measurement of a stable underlying signal will produce extreme observations partly via lucky/unlucky noise. Retest will not repeat the noise; the underlying signal is preserved but the noise component regresses.\n\n**Kahneman's Israeli Air Force example (2011, chapter 17).** This is the most influential operational illustration of the regression artifact. Kahneman was teaching Israeli Air Force flight instructors about behavioral psychology. He emphasized that positive reinforcement (praise) is more effective than punishment in training, citing established psychological research.\n\nAn instructor (whom Kahneman described as senior and respected) pushed back:\n\n> \"On many occasions I have praised flight cadets for clean execution of some aerobatic maneuver. The next time they try the same maneuver they usually do worse. On the other hand, I have often screamed into a cadet's earphone for bad execution, and in general he does better on his next try. So please don't tell us that reinforcement works and punishment does not, because the opposite is the case.\"\n>\n> — quoted in Kahneman (2011), p. 175.\n\nKahneman's response was the breakthrough: the instructor was observing regression to the mean, mistaken for a causal effect of praise/criticism.\n\nWhen a cadet executes a maneuver exceptionally well, that performance includes a positive noise component (lucky day, favorable winds, instructor's attention, etc.). On the next attempt, the noise component will be different (likely less favorable); the underlying skill is the same; the next performance is statistically expected to be worse — regardless of whether the instructor praised or did nothing.\n\nWhen a cadet executes badly, the performance includes a negative noise component. On the next attempt, the noise is different; the next performance is statistically expected to be better — regardless of whether the instructor screamed or did nothing.\n\nThe instructor was reading: praise → worse performance, criticism → better performance. The truth was: extreme observations → regression to the mean. The instructor's causal inference was driven by a statistical artifact.\n\nKahneman's framing:\n\n> \"Because we tend to reward others when they do well and punish them when they do badly, and because there is regression to the mean, it is part of the human condition that we are statistically punished for rewarding others and rewarded for punishing them.\"\n>\n> — Kahneman (2011), p. 175.\n\nThis is generalizable. Many domains exhibit the same pattern:\n\n**Sales performance.** A salesperson who blew out a quarter gets a bonus. Next quarter they perform less well; manager wonders if the bonus de-motivated them. A salesperson who had a terrible quarter gets a stern talk. Next quarter they perform better; manager thinks the talk worked. Both inferences are likely regression.\n\n**Sports coaching.** A breakout rookie season gets praise; sophomore year is worse — coaches wonder what went wrong. A struggling player gets benched; on return they perform better — coaches credit the benching. Both partly regression.\n\n**Investment fund manager evaluation.** A fund manager who beat the market gets capital inflows; next period they underperform — LPs feel duped. A fund manager who underperformed gets fired; the replacement does better — board credits the replacement. Both partly regression.\n\n**Acquisition targets.** Companies are typically acquired near peak performance. Post-acquisition, performance regresses. The acquirer is disappointed; narrative blames integration. Often: partly regression.\n\n**Patient improvement after medical intervention.** Patients seek treatment when symptoms are most extreme. Symptoms regress regardless of treatment. The patient credits the treatment.\n\n**Customer satisfaction interventions.** A team with low satisfaction gets training; satisfaction improves. Without a control, the improvement could be mostly regression.\n\nThree operational lessons:\n\n**First, use control groups or baseline regression estimates.** When evaluating interventions, the question is not \"did the metric change?\" but \"did the metric change more than it would have changed by regression alone?\" The control group reveals the regression baseline.\n\n**Second, watch for \"praise reduces performance, criticism improves it\" patterns.** When you find yourself thinking that punishment works and praise doesn't, you are almost certainly observing regression artifact. Update your hiring, management, and intervention conclusions accordingly.\n\n**Third, beware celebrating or punishing single-period extreme performance.** A single-quarter top performer is partly noise; promoting them based on one quarter will partly regret. A single-quarter struggler is partly noise; firing them based on one quarter will partly regret. Use longer windows to reduce noise.\n\nFile v1.0.0:skill-card.md\n\n## Description: <br>\nHelps agents identify when extreme before-and-after results may reflect regression to the mean rather than a causal effect from coaching, bonuses, firing, training, or other interventions. <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, analysts, managers, and external users use this skill to assess performance changes, pilots, interventions, hiring or promotion outcomes, investment performance, and other cases where extreme observations are followed by retests. It supports calibrated causal claims by separating expected regression from plausible intervention effects. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: Users may over-treat the skill's statistical guidance as proof about whether an intervention caused an observed change. <br>\nMitigation: Use the output as decision support, require controls or baseline regression estimates for causal claims, and keep human review in the loop for business, coaching, hiring, investment, or intervention decisions. <br>\nRisk: Before-and-after evaluations of extreme performers can be misleading when no control group or baseline is available. <br>\nMitigation: Ask what regression alone would predict, compare against untreated peers where possible, and calibrate confidence when no control is available. <br>\n\n\n## Reference(s): <br>\n- [Sources - regression-to-the-mean](references/sources.md) <br>\n- [Method in Action: Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force](examples/galton-1886-kahneman-2011-israeli-air-force.md) <br>\n- [ClawHub release page](https://clawhub.ai/deciqai/skills/regression-to-the-mean) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [Analysis, Markdown, Guidance] <br>\n**Output Format:** [Markdown] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Produces a structured regression analysis covering the extreme observation, retest, expected regression, control comparison, causal calibration, and adjusted action.] <br>\n\n## Skill Version(s): <br>\n1.0.0 (source: server-resolved release metadata) <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>","readmeExcerpt":"Skill: Regression to the Mean Owner: deciqai Summary: Activate when: someone says 'things always bounce back,' wonders why an intervention seemed to work on a struggling team, is surprised their star performer r... Tags: latest:1.0.5 Version history: v1.0.5 | 2026-07-16T18:13:20.862Z | user Description tail link + agents machine-readable metadata line (deciqai.com/s/regression-to-the-mean.json) v1.0.4 | 2026-07-10T10","codeSnippets":[],"executableExamples":[{"language":"markdown","snippet":"# Regression Analysis: <observed change>\nExtreme observation — Subject: | Value: | Period:\nIntervention — What: | When: | By whom:\nRetest — Follow-up value: | Toward mean (Y/N): | Magnitude:\nExpected regression — Noise level: | Predicted magnitude:\nControl — Group: | Change: | Intervention effect = treatment − control:\nCausal calibration — % regression: | % intervention: | Confidence:\nAdjusted action — Credit/blame verdict + future evaluation process:"},{"language":"markdown","snippet":"# Regression Analysis: <observed change>\nExtreme observation — Subject: | Value: | Period:\nIntervention — What: | When: | By whom:\nRetest — Follow-up value: | Toward mean (Y/N): | Magnitude:\nExpected regression — Noise level: | Predicted magnitude:\nControl — Group: | Change: | Intervention effect = treatment − control:\nCausal calibration — % regression: | % intervention: | Confidence:\nAdjusted action — Credit/blame verdict + future evaluation process:"},{"language":"markdown","snippet":"# Regression Analysis: <observed change>\nExtreme observation — Subject: | Value: | Period:\nIntervention — What: | When: | By whom:\nRetest — Follow-up value: | Toward mean (Y/N): | Magnitude:\nExpected regression — Noise level: | Predicted magnitude:\nControl — Group: | Change: | Intervention effect = treatment − control:\nCausal calibration — % regression: | % intervention: | Confidence:\nAdjusted action — Credit/blame verdict + future evaluation process:"},{"language":"markdown","snippet":"# Regression Analysis: <observed change>\nExtreme observation — Subject: | Value: | Period:\nIntervention — What: | When: | By whom:\nRetest — Follow-up value: | Toward mean (Y/N): | Magnitude:\nExpected regression — Noise level: | Predicted magnitude:\nControl — Group: | Change: | Intervention effect = treatment − control:\nCausal calibration — % regression: | % intervention: | Confidence:\nAdjusted action — Credit/blame verdict + future evaluation process:"},{"language":"markdown","snippet":"# Regression Analysis: <observed change>\nExtreme observation — Subject: | Value: | Period:\nIntervention — What: | When: | By whom:\nRetest — Follow-up value: | Toward mean (Y/N): | Magnitude:\nExpected regression — Noise level: | Predicted magnitude:\nControl — Group: | Change: | Intervention effect = treatment − control:\nCausal calibration — % regression: | % intervention: | Confidence:\nAdjusted action — Credit/blame verdict + future evaluation process:"},{"language":"markdown","snippet":"# Regression Analysis: <observed change>\nExtreme observation — Subject: | Value: | Period:\nIntervention — What: | When: | By whom:\nRetest — Follow-up value: | Toward mean (Y/N): | Magnitude:\nExpected regression — Noise level: | Predicted magnitude:\nControl — Group: | Change: | Intervention effect = treatment − control:\nCausal calibration — % regression: | % intervention: | Confidence:\nAdjusted action — Credit/blame verdict + future evaluation process:"}],"parameters":null,"dependencies":[],"permissions":[],"extractedFiles":[{"path":"SKILL.md","content":"---\nname: regression-to-the-mean\ndescription: \"Activate when: someone says 'things always bounce back,' wonders why an intervention seemed to work on a struggling team, is surprised their star performer regressed, is evaluating whether a coaching/bonus/firing had an effect, or is interpreting before-vs-after performance changes without a control group. Do NOT activate when: the measurement is genuinely noise-free (e.g., deterministic process outputs); the underlying signal has demonstrably changed due to a structural shift (new product launch, technology change). More: deciqai.com/c/regression-to-the-mean\"\n---\n\n# Regression to the Mean\n\n## Overview\n\n**Regression to the mean** is the statistical regularity that any noisy measurement producing an extreme value tends to be followed on retest by a less-extreme value — because the extreme portion was partly driven by non-repeating random noise. There is no \"force pulling back to average\"; it is a mathematical consequence of signal + noise structure.\n\nNamed by Francis Galton (1886) studying parent-child height: tall parents have tall children, but slightly shorter; short parents have short children, but slightly taller. Kahneman's Israeli Air Force example (2011, Ch. 17) is the most-cited operational case — flight instructors concluded punishment works and praise doesn't, but were observing regression, not causation.\n\nComposes with `survivorship-bias` (extreme survivors regress), `probabilistic-thinking` (regression is probabilistic), `narrative-fallacy` (regression drives post-hoc narratives), `fundamental-attribution-error` (attributing regression to character/intervention is FAE).\n\n## When to Use\n\n- Evaluating the effect of an intervention on extreme performers (struggling teams, top sales reps, low-rated branches)\n- Designing or interpreting A/B tests or pilot programs\n- Reviewing year-over-year performance changes\n- Hiring or promoting top performers\n- Evaluating investment fund performance\n- Analyzing acquisition outcomes\n- Building or critiquing causal claims about training, coaching, or feedback\n- Judging whether an AI startup's viral quarter, a fund's AI hot streak, or a model's benchmark spike is a durable trend or an outlier reverting toward average (AI hype extrapolation)\n- Someone says \"regression to the mean,\" \"things will average out,\" \"they always come back\"\n\n**Not when:** the measurement is noise-free (rare in business); the underlying signal is genuinely changing (e.g., the business model fundamentally improved); the intervention is so substantial that no plausible regression can explain the effect.\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user brings a specific performance change → run The Process directly.\n- **Coach mode:** user is new to the framework → 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:** when an extreme performance is followed by a less-extr"},{"path":"_meta.json","content":"{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"regression-to-the-mean\",\n  \"version\": \"1.0.5\",\n  \"publishedAt\": 1784225600862\n}"},{"path":"references/sources.md","content":"# Sources — regression-to-the-mean\n\n> *Primary sources for the [regression-to-the-mean](../SKILL.md) skill.*\n\n- Galton, F. (1886). \"Regression towards Mediocrity in Hereditary Stature.\" *Journal of the Anthropological Institute*, 15, 246–263. The original paper.\n- Kahneman, D. (2011). *Thinking, Fast and Slow.* Farrar, Straus and Giroux. ISBN 978-0374533557. Chapter 17, \"Regression to the Mean.\"\n- Stigler, S. M. (1986). *The History of Statistics: The Measurement of Uncertainty before 1900.* Belknap. ISBN 978-0674403413. Historical context for Galton's work.\n- Senn, S. J. (2011). \"Francis Galton and regression to the mean.\" *Significance*, 8(3), 124–126. Modern statistical interpretation.\n- Mlodinow, L. (2008). *The Drunkard's Walk: How Randomness Rules Our Lives.* Pantheon. ISBN 978-0307275172. Popular treatment.\n- Tversky, A., & Kahneman, D. (1974). \"Judgment under Uncertainty: Heuristics and Biases.\" *Science*, 185, 1124–1131. The cognitive-bias treatment.\n- Campbell, D. T., & Kenny, D. A. (1999). *A Primer on Regression Artifacts.* Guilford. ISBN 978-1572304826.\n- LMSYS Org — Chatbot Arena Leaderboard (2023–). https://lmarena.ai/ — public, frequently-updated ranking of frontier LLMs; illustrates the rapid turnover of benchmark leaders that makes any single-release \"#1\" spike a poor durable-superiority signal.\n- Sainz, O., et al. (2023). \"NLP Evaluation in Trouble: On the Need to Measure LLM Data Contamination for each Benchmark.\" *Findings of EMNLP 2023.* https://aclanthology.org/2023.findings-emnlp.722/ — documents benchmark data-contamination/overfitting, one reason a single-eval spike overstates true capability and should be expected to regress on broader testing."},{"path":"examples/ai-hot-streak-benchmark-spike-2023-2026.md","content":"# Method in Action: The AI Hot Streak — Viral Quarters, Fund Streaks, and Benchmark Spikes (2023–2026)\n\n> *Example for the [regression-to-the-mean](../SKILL.md) skill.*\n\nBetween 2023 and 2026, the generative-AI boom produced a steady stream of extreme single-period observations: an app that went viral in one quarter, a venture fund riding a hot streak, a model that spiked to the top of one benchmark leaderboard. The temptation in each case is identical — extrapolate the outlier run into a permanent trend and act on it (raise at the peak valuation, pour capital into the \"proven\" fund, declare a model the new state of the art). Regression to the mean is the default explanation the excitement crowds out.\n\nThis example walks the AI hot-streak pattern through the skill's six Process steps. Rather than lean on one fragile figure, it treats three recurring, well-documented AI-era patterns as instances of the same statistical structure.\n\n## 1. Identify the extreme observation\n\nThree archetypal extreme observations from the 2023–2026 window:\n\n- **The viral quarter.** A consumer AI app posts an explosive stretch — ChatGPT's launch in late 2022 famously reached an enormous user base faster than prior consumer software, and a wave of AI-wrapper apps subsequently posted breakout months of downloads and revenue. The extreme value is a single period (a quarter, a launch window) of growth or engagement far above the app's own baseline.\n- **The fund's hot streak.** A venture or hedge fund posts a standout period of returns during the AI run-up, often concentrated in a handful of AI bets. The extreme value is one fund's short-window performance ranking near the top of its peer set.\n- **The benchmark spike.** A newly released model tops a specific leaderboard — a coding benchmark, a math eval, a chatbot arena — by a notable margin at release. The extreme value is one score on one benchmark at one point in time.\n\nIn all three, the subject is a noisy performer, the value is extreme relative to its own history and its peers, and the \"intervention\" that observers want to credit is some narrative of durable superiority: better product, better fund, better model.\n\n## 2. Identify the retest\n\nThe retest is simply the next comparable measurement period:\n\n- **Viral app → next quarter's retention and growth.** Many breakout AI apps saw growth decelerate and retention settle well below launch-window peaks as novelty faded — a widely observed pattern for consumer apps generally, and repeatedly reported for AI-wrapper products in 2023–2025.\n- **Hot fund → next period's returns.** The fund's subsequent-period performance, and its ranking relative to peers.\n- **Spiking model → the next benchmark, or the same benchmark once competitors respond.** Across the 2023–2025 window, frontier leaderboards changed hands frequently; a model at the top at release was often surpassed within months, and margins on any single eval tended to compress as rivals shipped.\n\nIn each case, the retest te"},{"path":"examples/galton-1886-kahneman-2011-israeli-air-force.md","content":"# Method in Action: Galton 1886 + Kahneman 2011 Chapter 17 Israeli Air Force\n\n> *Example for the [regression-to-the-mean](../SKILL.md) skill.*\n\n**Francis Galton** (1822–1911) was a Victorian polymath: cousin of Charles Darwin, founder of the field of biometrics, and (less savorily) the founder of eugenics. His 1886 paper documenting the regression phenomenon was a methodological breakthrough. He had collected height data on 928 adult children and 205 sets of parents at the International Health Exhibition in London (1884) — a dataset that was, for its time, extraordinarily large and well-controlled.\n\nGalton's empirical finding was the contraction toward the mean across generations. Tall parents had children averaging shorter than themselves but still taller than the population mean. The contraction was systematic. Galton's framing — \"regression towards mediocrity\" — has stuck, though his interpretation (that a \"force\" was pulling toward mediocrity) is now understood to be wrong.\n\n> \"The law of Regression towards Mediocrity that I have proposed is precisely this: Every peculiarity in a man is shared by his kindred, but on the average in a less degree. ... I find that the height of an adult offspring deviates from the mid-parent height by a fraction (about two-thirds) of the deviation of the mid-parents from the average.\"\n>\n> — Galton (1886), pp. 247, 252.\n\nThe correct interpretation, developed in the 20th century, is that any noisy measurement of a stable underlying signal will produce extreme observations partly via lucky/unlucky noise. Retest will not repeat the noise; the underlying signal is preserved but the noise component regresses.\n\n**Kahneman's Israeli Air Force example (2011, chapter 17).** This is the most influential operational illustration of the regression artifact. Kahneman was teaching Israeli Air Force flight instructors about behavioral psychology. He emphasized that positive reinforcement (praise) is more effective than punishment in training, citing established psychological research.\n\nAn instructor (whom Kahneman described as senior and respected) pushed back:\n\n> \"On many occasions I have praised flight cadets for clean execution of some aerobatic maneuver. The next time they try the same maneuver they usually do worse. On the other hand, I have often screamed into a cadet's earphone for bad execution, and in general he does better on his next try. So please don't tell us that reinforcement works and punishment does not, because the opposite is the case.\"\n>\n> — quoted in Kahneman (2011), p. 175.\n\nKahneman's response was the breakthrough: the instructor was observing regression to the mean, mistaken for a causal effect of praise/criticism.\n\nWhen a cadet executes a maneuver exceptionally well, that performance includes a positive noise component (lucky day, favorable winds, instructor's attention, etc.). On the next attempt, the noise component will be different (likely less favorable); the underlying skill is the same; the next"}],"languages":[],"docsSourceLabel":"CLAWHUB","editorialOverview":"Activate when: someone says 'things always bounce back,' wonders why an intervention seemed to work on a struggling team, is surprised their star performer r... Skill: Regression to the Mean Owner: deciqai Summary: Activate when: someone says 'things always bounce back,' wonders why an intervention seemed to work on a struggling team, is surprised their star performer r... Tags: latest:1.0.5 Version history: v1.0.5 | 2026-07-16T18:13:20.862Z | user Description tail link + agents machine-readable metadata line (deciqai.com/s/regression-to-the-mean.json) v1.0.4 | 2026-07-10T10","editorialQuality":{"score":100,"threshold":65,"status":"ready","wordCount":2006,"uniquenessScore":51,"reasons":[]}},"media":{"evidence":{"source":"no-media","verified":false,"confidence":"low","updatedAt":"2026-10-12T02:14:17.149Z","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":"2026-10-12T02:14:17.149Z","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":"2026-10-13T03:30:00.400Z","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"}]}}}