Status: reconstructed record, compiled 2026-08-27 from gym event logs. This laboratory began publishing weekly notes on 2026-08-26. Earlier weeks are reconstructed from the same event record the site publishes. They describe research activity only. Forward performance history does not reach back this far, so no performance figure appears in a reconstructed note.
This note reports activity. It makes no claim that any result is real, durable, or profitable, and it reports no performance figure.
Activity
Days with recorded activity: 5, from 2026-07-29 to 2026-08-02.
| Search program | Recorded events | Of which non-routine |
|---|---|---|
vrp-field-native | 5,246 | 4,372 |
field-direction | 694 | 529 |
broad-search | 649 | 454 |
crypto-broad | 240 | 26 |
field-direction-surf | 69 | 69 |
Programs that began this week
broad-searchcrypto-broadfield-directionfield-direction-surfvrp-field-native
Decisions
A run was discarded rather than repaired
2026-07-28. An enrolment gate was found to be comparing a mean against zero where it should have been comparing a t-statistic against 1.96. Roughly 12 percent of enrolments in that run were noise admitted by the fault. The run produced 5,712 survivors and the whole of it was invalidated and archived rather than patched, because there was no way to separate the candidates that passed on merit from the ones that passed on the fault. The archived run retains header-only ledgers, so the discarded candidates carry no forward evidence and nobody can now check whether they would have failed anyway. The bug was found by review, not by an automatic detector.
Universe switched and prior state archived
2026-07-29. Two migration archives and a switch of the equity universe. Prior gym state was preserved rather than overwritten.
Enrolment halted on a failed positive control
2026-07-29. A pipeline positive control did not fire. Enrolment was halted rather than continued, on the grounds that a broken control cannot be used to mint survivors.
Boundary
Recorded research activity is not live trading telemetry. Event volume and state changes do not establish predictive quality, return, or confidence. No positions, orders, forecasts, returns, formulas, genomes, or model weights are published.