Wrote 31 new candidate crypto features from research cards, then swept them across the entire price history to see which carry information the live columns do not. 28 got built, 26 survived the sweep, and 18 remain once near-duplicates are removed.
What it measured
Quantity
Value
Meaning
Kernels authored
28 of 31 cards (3 deferred: cards 53, 56, 84)
three were left unbuilt rather than guessed
Sweep grid
441 cells (21 symbol-threshold x 21 slices), 0 errors, 0 skips; 416 of 441 clear the 1000-bar floor
Authored 6 distributional-shape kernels clean-room from cards
02
6 serial-dependence / trend / asymmetry kernels
03
5 more; OHLC substrate dispatch validated; first card deferred
05
Kernel phase closed at 28 built, 3 deferred
07
Full single-pass orthogonality sweep launched across 441 cells
11
Sweep verdict plus adversarial review worked into verdict.md; loop closed
What it produced
PR #540 (terrylica/opendeviationbar-py)
FINAL_FEATURE_SET.md (the 18, with FOSS reference, licence, validation level per feature), CANDIDATE_REGISTRY.md, DEDUP_REDUNDANCY.md, verdict.md
Promotion tracker as of 2026-06-30: 8 promoted, 4 next, 4 xi-held, 2 held
Carries forward
0 of 18 are FOSS-oracle-validated โ each still needs its <= 1e-9 oracle. 4 features scheduled next, 4 xi-watch features need an explicit operator ruling, 2 held. Cards 53/56/84 re-enter once their conventions are pinned. Also: logret re-sweep of vonneumann_ratio and cid_ce_arclength, a recompute==materialized assertion, removal of redundant materialized Hjorth columns.
Lifecycle status is process state, not a judgement of the findings. Results are stated as numbers with their uncertainty. Generated between CLOSING-RECONCILIATION markers โ regenerate rather than edit.