The Register.
The thresholds FP1 fixed before tracking opened, the observations used to grade them, and the permanent record of what happened next. Grade dates do not move. Publication lateness remains visible.
Thresholds, observations, and publication are separate.
Pre-registration
Each dimension names what is tracked, the threshold that counts as movement, the direction the evidence can resolve, and the grade date. No threshold changed inside this window.
Late-grade rule
A late grade remains valid only when the threshold, observation date, instrument composition, calculation rule, source series, and required lookback were fixed in advance or can be independently recovered from the registered method. If reconstruction requires a new discretionary choice, the window is marked compromised or ungraded.
Reconstruction
The original v0.1 production code and archived July values were recovered. The weekly implementation reproduced the 20 July FNC-1 and SOX targets within the predeclared tolerance. The same implementation was then run through 31 August. The complete Belief Index path was replayed at three observation times; its threshold verdict did not change.
Both registered dimensions held.
The Belief Index crossed 55 once on 31 August, but its rule required two consecutive weekly readings above 55. That single reading is preserved as a near-miss, not converted into a trip.
The grade date was 31 August. The historical replay finished on 5 September after the original code and inputs were recovered. The lateness does not alter either verdict, but it is an operational failure and remains visible here. Future windows use the prospective v0.2 reproducibility specification with pinned observation times, identifiers, source series, and code hashes.
The six Belief Index readings used.
Primary observation time: Monday 21:00 UTC. The original method selected the maximum Yes price among the historically open submarkets in each registered Polymarket event. That code-as-run rule is retained for this reconstruction and replaced prospectively by pinned contract identifiers.
| Date | Belief Index | Rule status |
|---|---|---|
| 27 Jul | 48.745 | Inside hold range |
| 3 Aug | 48.380 | Inside hold range |
| 10 Aug | 49.300 | Inside hold range |
| 17 Aug | 49.320 | Inside hold range |
| 24 Aug | 45.315 | Inside hold range |
| 31 Aug | 57.490 | Near-miss · not a second consecutive reading |
Versions and hashes retained with the grade.
Capital acceptance
Archived target: FNC-1 128.45 and SOX 209.30 on 20 July. Replayed weekly values: 128.452 and 209.337. The weekly-explicit construction passed; the daily alternatives did not and were not used for the grade.
Replay code
SHA-256 f3f85a537b7bd8b6aee2a74eeca31d327c2194e26698088003383f92f1bc28d5
Environment: Python 3.14.4, pandas 3.0.3, requests 2.34.2, yfinance 1.4.1. Full local replay output hash: 6ba505fbd27e1762f5df1a41a1f99a533cddaf27aec6277730cd91cf12eb6c56.
Public files
Measurements excluded from this window.
The v0.1 basket rested on a single listed company carrying a fifteen-percent weight. One company could not defensibly stand for a whole substrate.
The rolling rebase made the level only approximately comparable across issues. The signed FNC-1 minus SOX spread was registered instead.
This page preserves the original window, the reconstruction, and the late-publication scar. It does not create a new Radar sweep or R-013.