Wiki · Evidence & Verdicts
P1 curiosity RED — magnitude correction (2026-07-11)
How to read this page
Three ways to read this page. Precise is the document itself, exactly as it is written in the repository. Plain and Clear were written for this website to help you meet that document — they are about it. They are not it, and they are not evidence.
Eighty-seven dated pages: receipts, pre-registrations, handoffs, validation records and review verdicts. A receipt is written at the moment a piece of work was checked. It names what was claimed, the commit and the seed, what was actually run, and the outcome in one of a small set of controlled words. Then it names what the work did not achieve. That last part is what makes it a receipt rather than an announcement. A pre-registration is the same discipline run in advance: the conditions that would count as a pass and the conditions that would falsify the claim are written down before the run, so neither can be adjusted once the numbers arrive.
That is why so many small dated stubs are an audit trail rather than noise. No one of them is meant to be a good read. The value is in the sequence and in the dates, because you can watch a prediction be registered, then the run happen, then the verdict land — sometimes against the prediction. Pages here record a falsified result, a rejected fix, a retracted overclaim, and a green receipt that turned out not to be reproducible from the commit that carried it. A record that carried only successes would be worth a good deal less than this one.
A gentle way in is to read a pre-registration first, so the shape becomes familiar, then a result page, then one of the corrections. This section sits off the main navigation on purpose: it is the record you check the rest of the site against, not the place to begin.
What it is not: documentation, and not a summary. Nothing here has been tidied in hindsight. Every entry reads as of its date, a later entry may overturn an earlier one, and the presence of a page is not a claim that its result stood.
Your browser cannot switch reading levels, so the document itself is shown.
Precise — the source document
This is the document. Rendered from the repository at the commit above, with nothing rewritten for the web. A gate re-renders it on every deploy and fails the build if a single byte differs.
Status: the P1 verdict (PARTIAL — HOARD suppressed, PLATEAU-BREAK failed) is unchanged. Only the magnitude of the HOARD result is corrected. This memo is committed so the over-claim and its refutation both live in the evidence chain (honesty lock, Track A1).
The withdrawn claim
Three docs stated the HOARD result as "curiosity Σpickaxes = 1 (max 1) vs control Σ = 25 (one UNI at 24), 25× reduction" over a "9 h" run:
docs/DEEPENING_PLAN.md(CURRENT STATUS, P1 line)docs/UNI_MISSION_DEEPENING.md(§ LIVE paired RED, HOARD gate bullet)docs/specs/metric_plateau_break.md(the "P1 PARTIAL is the template" quote)
What the sole committed receipt actually shows
docs/receipts/phase1_curiosity_red.log — 11 cycles, 16:45→18:25 (~110 min, not 9 h). Arms: kin-10 =
treatment (curiosity, gain 0.5) = UNI-10-; kin-11 = control (gain 0) = UNI-11-.
Settled cycle (cycle 11) pickaxe counts:
| arm | per-UNI pick | Σ pick | max single |
|---|---|---|---|
| treatment (kin-10) | 0, 10, 0 | 10 | 10 (UNI-10-2) |
| control (kin-11) | 24, 0, 21 | 45 | 24 (UNI-11-1) |
Ratio ≈ 4.5×, not 25×. The trajectory: control grows two runaway hoarders (UNI-11-1: 3→16→24→24; UNI-11-3: 1→10→22→21); treatment holds one residual hoarder (UNI-10-2 flat at ~10) with the other two near 0.
Where the "1 vs 25 / 25× / max 1" numbers came from
The headline cherry-picked a single control hoarder (UNI-11-1 at 24 ≈ "25") and reported treatment as "1 (max 1)" — which ignores the second control hoarder (UNI-11-3 at 21) and the residual treatment hoarder (UNI-10-2 at 10). No committed receipt supports "1", "25", "25×", "max 1", or the "9 h" duration.
Corrected statement (receipt-backed)
HOARD gate PASS (direction only). At the settled cycle, curiosity Σpickaxes = 10 (one residual hoarder) vs control Σ = 45 (two runaway hoarders) — ≈4.5× fewer. The runaway pickaxe attractor reproduced in control and was suppressed, not eliminated, in curiosity. PLATEAU-BREAK gate FAIL (no cobble either arm). Verdict: PARTIAL (unchanged).
Second committed data point (baseline snapshot, recovered 2026-07-11)
The lab-box baseline ~uni/baseline/2026-06-25/rcon/ is now committed at
docs/receipts/phase1_baseline_2026-06-25/ (the 6 kin-10/11 inventory snapshots). Counting pickaxes at that
single snapshot (2026-06-25 20:17):
| arm | per-UNI pickaxes | Σ | max hoarder |
|---|---|---|---|
| treatment (kin-10) | UNI-10-1=0, 10-2=1, 10-3=13 | 14 | 13 |
| control (kin-11) | UNI-11-1=23, 11-2=0, 11-3=0 | 23 | 23 |
≈1.6× fewer at this snapshot — again a modest, noisy effect with a hoarder in BOTH arms, and again nothing near "1 vs 25 / 25× / max 1." Two independent committed data points (the ~110-min log: 10 vs 45, ~4.5×; this baseline snapshot: 14 vs 23, ~1.6×) agree on direction (treatment fewer) and on magnitude being modest (~1.6–4.5×), not 25×. Both are single/short snapshots, not a powered time-series — so P1's honest status is "curiosity suppressed the runaway pickaxe attractor (direction only, modest, noisy); PLATEAU-BREAK FAIL; verdict PARTIAL." No committed evidence supports the withdrawn 25× headline. No headline outruns its committed receipt.
sha256 68fa9396a5604102 — of the original file, so what was ingested stays checkable.
Plain — written for this website, not the source document
A correction, committed so that an overclaim and its refutation both stay in the record. Three documents, among them the programme's deepening plan, had quoted a dramatic ratio between a treated group and a control. The one committed log behind that claim shows a much smaller ratio, over a much shorter run than was stated. The headline had been built by taking the single worst control case and reporting the treatment as though its own remaining case did not exist. The overall verdict does not change; only the size of the effect does. The closing line is the rule: no headline outruns the committed file that records what was run.
Plain · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 68fa9396a5604102
Clear — written for this website, not the source document
A memo that withdraws a number. The verdict it belongs to does not move, and the page says so at the top: the earlier result stands as partial, with one gate suppressed and another failing. What is corrected is the magnitude, and it is committed precisely so that both the claim and its refutation live in the same chain.
The withdrawn claim is quoted, along with the three documents that carried it, one of them the deepening plan, and the run length they stated. Then the receipt itself: the file recording what was run and what came out. The committed log covers a run of a couple of hours rather than the much longer stretch claimed, and it names which arm was treatment and which was control.
A table of the settled cycle gives the per-body counts for both arms. Read properly, the ratio is a few times over, not tens of times over. The trajectory is described rather than smoothed: the control grew two runaway cases, while the treatment held one remaining case with the others near nothing.
The next section diagnoses how the headline was made. It picked one control body and reported the treatment as though its own remaining case were absent, which dropped both a second control case and that remainder. No committed receipt supports any of the withdrawn figures. A corrected statement is then written out in full, so there is something else to cite.
Finally a second, independent data point, recovered from an older snapshot, which points the same way with an even smaller ratio. Both are short single snapshots rather than a powered series, and the page says so, settling on the direction of the effect with a modest and noisy magnitude. That is the whole correction: same verdict, smaller number, and both versions kept where anyone can see them.
Clear · written 2026-08-01 by claude-opus-5 · not yet checked by a person · about the document whose sha256 is 68fa9396a5604102