Claim status audit — three contested figures, and what the archive actually holds

Date: 2026-09-17 (Loom) Trigger: the /lab principal-investigator page sends scientific reviewers to /about and the research library. Two stale figures were found on the way; John asked for an archive search before anything was changed, so that no genuine finding is removed. Rule applied: a figure stays if the archive holds a completed test for it. A figure with no test on file is a hypothesis and is labelled as one. Nothing is deleted from the archive.

---

1. Verdicts

Figure	What the archive holds	Verdict
Solar → VIX, r = 0.68	Tested twice. 7-day pilot, Oct 2025: r = 0.12, p = 0.43 (original-research.ts:1602). Full test, Dec 2025: 419 months, NOAA SSN × FRED VIXCLS, r = 0.075, p = 0.118, CI crosses zero, Granger null at lags 1, 2, 7 (COSMIC_CLAIMS_VALIDATION_LOG_2025-12-15.md, Test 1; endpoint /api/cosmic/solar).	REFUTED. The refutation is the genuine work and is kept and shown.
Housing → Confidence, r = −0.87 / −0.788 / −0.894	All three are level correlations from different runs. On monthly changes: −0.041, p = 0.66 (FRED_LEVELS_VS_CHANGES_2026-09-11.md); validated-correlations.ts re-measure gives +0.007.	WITHDRAWN. Two opposing trends. Shown as a visible withdrawal, not erased.
Schumann → Events, r = 0.74	Never computed. The library paper states it in its own text: "Any correlation statistics for the full historical period are ILLUSTRATIVE of the proposed framework, not from completed empirical analysis" and "The regression coefficients and R² values in other sections of this paper are ILLUSTRATIVE" (original-research.ts:558, 562). No script, no dataset, no event table exists in the repo. Direct measurement began ~1960, so the "1800–2024" and "1954" series are proxy reconstructions that were proposed and never built. No public numeric Schumann feed exists (Sep 9 finding).	HYPOTHESIS, untested. Kept as a proposed framework. The Schumann Station project is the route to a real test.
GCP → Events, r = 0.67	No test on file. The Global Consciousness Project's own published result (Nelson et al. 2002) is a cumulative deviation statistic, not a Pearson r of 0.67; the 0.67, p = 0.02, n = 500 figure appears only in research-validation.ts and documents that cite it.	HYPOTHESIS, untested by us. Cite GCP's published work as theirs.

The tell that separates computed results from placeholder figures

Identical test statistics recur across unrelated claims:

• Granger F = 12.3, p = 0.002 appears for Solar → VIX (research-validation.ts:137) and for Schumann → Events (original-research.ts:574, 619).
• Granger F = 18.7 appears for M2 → Gold (research-validation.ts:53) and for the energy↔food claim that was found to have no test record on Aug 2. Two independent analyses do not return the same F to three figures. Any entry carrying one of these values should be treated as unverified until a script reproduces it.

---

2. What is genuinely ours and must be kept

These have a completed test, a named data source and, where marked, a replication script.

1. The level-versus-changes audit itself — FRED_LEVELS_VS_CHANGES_2026-09-11.md, scripts in scripts/research/. Finding a systematic error in our own published results and correcting it at the source is the strongest credential the program has.
2. Fed Funds → 2-Year Treasury: r = +0.476 on changes (0.413 ex-COVID, n = 248). Strongest result held.
3. Oil → Industrial Production: r = +0.381 on changes (n = 246), 0.463 with oil leading one month. The only pair that improved when measured properly (level figure was 0.316, published 0.841).
4. Consumer Credit → Employment: r = +0.250 on changes — real, modest.
5. M2 → CPI lag structure: near zero contemporaneously, roughly 0.4–0.5 at a two-to-three-year lag on year-over-year growth, 1960–present. The shape survives every window; the exact lag does not.
6. Solar → DXY: r = 0.234 on levels (p = 0.001), 0.082 on changes (n.s.). Reported as weak.
7. Published refutations and one retraction — Solar → VIX, Tectonic → Gold (17,443 earthquakes, r = 0.131), Solar → DXY as claimed, and the Hormuz/Bitcoin retraction (CLAIM_RETRACTION_HORMUZ_BITCOIN_TOLL_2026-08-17.md). Lunar is NOT on this list: the Sep 11 audit established that no lunar test was ever run.
8. The preregistration — PREREG_SOLAR_PHASE_CRISIS_TEST_2026-08-16.md, registered before analysis, not yet run.
9. The Record — 66 calls, Merkle root anchored on Base, block 50553167.
10. The solar-forced overshoot thesis — an independently derived framework; a framework, labelled as one.

GDP ↔ Consumption (+0.931 on changes) survives and is not a finding: it is close to an accounting identity.

---

3. Changed today

• src/app/about/page.tsx — Housing tile now shows −0.04 as a visible withdrawal with the history stated. Also corrected the tile beside it: "0.97 Fed → Treasury" was the level figure; it now shows 0.48 on changes with the prior figure stated, matching how the Credit tile was already handled.
• src/app/library/research/static-state-of-change/page.tsx — "The Empirical Anchor" listed Solar 0.68, GCP 0.67 and Schumann 0.74 under the sentence "These are not illustrations of the concept. They are measurements of it." That sentence is now true: the three entries are Fed → 2Y (0.48), Oil → Industrial Production (0.38) and Solar → VIX as measured (0.075, with the proposed 0.68 stated). The essay's argument is untouched. A short note keeps Schumann and GCP on the page as open hypotheses.
• .../static-state-of-change/layout.tsx — metadata description no longer advertises 0.68 / 0.74.
• src/data/lab-profile.ts — audit figures aligned to the audit document (−0.89 levels, −0.04 changes).

Typecheck: 7-error baseline, none in touched files. Nothing committed.

---

4. The sweep — done the same day on John's go

Scale, stated once (John, Sep 17). The thesis is civilizational: societies respond to environmental forcing the way herds follow the rains, through slower and more complex channels (harvests, prices, migration, conflict) over years to centuries. A monthly Pearson r against a market index is the wrong instrument for that claim. The refutations above rule out a fast, direct market link. They do not test the thesis; the preregistered solar-phase test does. This is now written into the registry header, the Solar → VIX entry, the shared API block and the /lab page, with three verified citations for the climate–society literature (Zhang 2007 PNAS, Büntgen 2011 Science, Hsiang/Burke/Miguel 2013 Science; DOIs checked against Crossref).

New measurements (scripts committed alongside):

• Yield curve → USD/EUR: 0.809 reproduces exactly on levels, 0.097 on changes (n.s.). Withdrawn.
• Yield curve → recession: -0.353 at an 18-month lead (n = 585), not the published -0.771. All 6 recessions since 1976 followed an inversion; 4 of 7 inversion episodes were followed by a recession within 24 months (5 of 7 within 30). That 4/7 is now the only probability the Intel scenario panel shows.
• M2 → CPI lag is window-dependent — absent in 1990-2019. The Sep 11 doc overstated it; corrected there.

Files changed (all uncommitted):

• src/lib/library/research-validation.ts — rebuilt: 3 empirical, 6 in-research (2 measured, 4 awaiting a first test and carrying NO statistics), 9 failed/withdrawn with measured values, 4 framework. The tested-count in the summary now counts only entries that carry a measurement.
• src/lib/financial/measured-correlations.ts — NEW. One shared block + one disclaimer. Seven API routes import it instead of hand-typing coefficients: stock-market, commodities, commodities-enhanced, fred/economic-indicators, predictive, stocks-enhanced, trending-stocks. Never re-type a coefficient in a route again.
• src/app/api/v1/correlations/route.ts + src/app/api-docs/page.tsx — public API returns change-based figures; entry ids unchanged; methodology block states the transform and the correction date.
• src/app/api/intel/market-scenarios/route.ts + src/components/intel/financial/MarketImpactForecast.tsx — rebuilt. No projected magnitudes, no "Granger confidence" (no Granger test was ever on file). Probability appears only as the 4/7 historical base rate, labelled as one.
• src/app/about/page.tsx (tier tiles now read the registry), src/app/pitch/page.tsx, src/components/pitch/MethodologyShowcase.tsx, src/lib/library/research-sources.ts, floating-books.tsx, vocabulary-narrowing-data.ts, original-research.ts (Schumann key-findings marked illustrative), MatrixConfluencePanel.tsx, MacroLauncherModal.tsx, noaa-solar-fetcher.ts.
• /lab: src/app/lab/page.tsx, src/data/lab-profile.ts (now carries the thesis), src/app/sitemap.ts.

Typecheck: no errors in any touched file. Not yet walked in a browser.

4b. The preregistered solar-phase test was RUN (same day)

SOLAR_PHASE_CRISIS_TEST_RESULT_2026-09-17.md. H1 not supported as registered: banking crises no clustering; sovereign defaults a 12–16 % declining-phase tilt, significant on the country count (p = 0.02), not on the year count (p = 0.05–0.07); NBER US no clustering on the primary window. Sources disagree → list-dependence per §6. Registry entry moved to not-validated with the numbers. The clustering-aware follow-up was registered and run the same day (SOLAR_PHASE_SOVEREIGN_DEFAULT_TEST_RESULT_2026-09-17.md): ratio 1.14, primary p = 0.053, permutation p = 0.12. Not supported. Independent sample needed.

5. Left open

• src/data/whaleTooltipContent.ts — claims "empirical market data 1970-2024, n = 1,247 whale movements", r = 0.68, χ² = 15.3, "34% of moves on Fibonacci days". Nothing on file. John's decision, untouched.
• commodities-enhanced economic_indicators block: gold 0.78, dollar -0.85, fed_policy_impact 0.92 — unsourced.
• The four literature entries in FAILED_CLAIMS have no replication script — DONE 17 Sep: scripts/research/literature_claims_replication.mjs (output beside it). Lunar phase → S&P: r = −0.012, p = 0.53, n = 2,512. Mercury retrograde → VIX level: r = 0.025, p = 0.07; absolute daily VIX moves are larger on retrograde days (p < 0.001, survives ex-2008/2020) but the per-year sign test is 14/21 (p = 0.095), so it fails the robustness rule declared before the run — recorded as an anomaly for a fresh-window registration, not a finding. Fibonacci intervals: 0 of 12 near a Fibonacci number (small n). Planetary alignments → M6+ quakes: r = −0.035, p = 0.40, 599 months. Registry entries now carry these numbers.
• Schumann paper in original-research.ts: coreThesis.professional (factor betas) and methodology ("tested on out-of-sample data") still read as completed work.
• "372 tests" has never been re-derived from MASTER_CORRELATION_RESEARCH_DATABASE_2025-10-15.md.
• "Published on a DeSci platform, peer-review in progress" was removed from three API disclaimers as unverifiable. If true, it belongs in PUBLICATIONS on /lab.