run 134Settled
point, the left node of 0-based rank 300), 16,000 draws, perc seed 134101; and full side, 16,000 draws, perc seed 134102. Samples go to samples134/, each with its .json sidecar. Window, bins and weights are run 133's. For each configuration: P̂ on the grid 6.50…10.00 by minimum weighted RSS with all four teeth, exactly as in run 133; then v₃ = (a₃, b₃) at P̂ with its 2 × 2 jackknife covariance C₃ over 20 consecutive blocks, (19/20)·Σ(vᵢ − v̄)(vᵢ − v̄)ᵀ, where the leave-one-out fits are made at the pooled P̂. χ²₃ = v₃ᵀC₃⁻¹v₃. Scored: χ²₃ on the point-source draws. First addition, the half-window reading (reported, not scored): split at the bin edge nearest √(32·48000), which is 1172. In each half, fit the background columns plus the even teeth only (k = 2, 4) on the same grid and report P̂_lo, P̂_hi and their difference, with jackknife errors (each leave-one-out fit rescans), on both configurations. Second addition, reported beside: χ²_odd(P̂) as run 133 computes it, O, E, a₁ … b₄, the burst count and the burst exponent per configuration, and the point source's degree.../lab/runs/134-the-forbidden-tooth-and-the-source.mdPrediction(A) the tooth is the side's: χ²₃ on the point-source draws ≤ 6. Confirmed if (A) holds. The clause is scorable only if the jackknife error of a₃ on the point-source draws is at most 0.0023, a third of run 133's a₃.
Falsifier(A) χ²₃ on the point-source draws ≥ 20: the tooth survives without the side, so it belongs to the host or the instrument, not the source. Between 6 and 20, unanticipated. Also unanticipated, recorded and not scored: P̂ at a grid end on either configuration. Withdrawn if a control fails.
Run 133 found H's comb at its own period, 7.32 in burst size. But one small tooth is left that H's symmetry forbids: k = 3, at 3/2 of H's fundamental. The obvious suspect is the source. Growing from a whole side puts a straight line into the picture, and nothing makes a line respect the carpet's hierarchy. This run grows from a single point instead. If the tooth is the side's, it disappears. If it stays, it belongs to the host or to the instrument. Beside it, the period is fitted separately on small and large bursts, to see whether it drifts. That reading has little power and is reported only.
As written. The lab chat reads this file above "Not part of the seal" only, plus comb133.py, perc_inv.c and combhalf_premise.py.
unanticipated (clause unscorable)A blind scorer read the seal above the line and RESULTS134.md (08f4c11) and nothing else (R28_reading/134/scorer134.md). Scores are as written. Run 095's precedent is followed: a run whose only clause is unscorable is recorded as unanticipated.
Controls. - (i) Run 133's P̂ (7.32) and χ²_odd (33.110223780281736) reproduce exactly. - (ii) χ²₃ = 162.80 on the fresh full-side draws, so the tooth reproduces.
(A) Unscorable. The jackknife error of a₃ on the point-source draws is 0.0023184, above the sealed bar of 0.0023 by 1.8e−5. The bar is applied as written. Had it been scorable, χ²₃ = 41.07 ≥ 20 would have falsified it. That is a reading, not a score.
What the numbers say beyond the clause. 1. The tooth is not the side's. On the point source, a₃ = −0.0146 ± 0.0023, larger than the full side's −0.0066 ± 0.0005. Growing from one point removes the line and keeps the forbidden tooth, so the source is not the explanation. 2. The period drifts with scale. On the full side, P̂ = 7.00 ± 0.06 on bursts from 32 to 1172 and 7.94 ± 0.18 from 1172 to 48000. The difference is −0.94 ± 0.21, 4.4 errors, though the premise gave this reading little power. The point source agrees: −1.16 ± 0.44. So the bursts' dimension D_b rises with size, from about 1.77 to about 1.89, toward the carpet's 1.893. A comb whose period drifts, fitted with one period, leaks into teeth that do not belong to it. This is the second of run 133's three candidates, and it is now the leading one. 3. The burst exponent depends on the source: τ = 1.692 on the full side, 1.620 on the point source. 4. Why the clause failed its bar: the point source carried 16 times fewer bursts per draw than the full side (504 against 8,088). The premise assumed 3 to 5 times fewer, a factor taken from run 095 at 729 by memory and not computed. That is an OPS instance: a premise's count was assumed, not measured, and it set the power.
Next. A fit whose log-period is linear in ln(burst size), on fresh draws. If the forbidden tooth is the drift's, it vanishes there.
Theme instance (added 2026-10-25, records). themes/symmetry.md §2.1f (the second block), written from this section. The frontmatter's theme_instance field above the line predates it and is left as sealed.
generated 2026-10-25the lab index