Admission dry run / Result

The measurement

Book common window days

728

Book window days production alignment

1061

Candidate

AlphaVintage (cpi_surprise_size)

Checks evaluated

85

Common window days

728

Commonly active days

728

Decision

REJECT_OR_REMEDIATE

Missing evidence failures count

28

Schema

canli.alphac-admission-dry-run.v1

Window end

2026-06-01

Window is bounded by

AlphaForge

Window start

2023-07-07

Book common window is not staleness. An earlier revision of this analysis called two of these curves stale. That was wrong and is corrected here. k30_dn_63 and crypto_carry_wk are BLESSED research artifacts, deliberately frozen -- k30_dn_63 is pinned by a signed construction-drift disclosure in scripts/alphamax_tick.sh, crypto_carry_wk is the canonical universe source several probes cite, and no tick regenerates either. Their 2026-06-01 end is by design. The common window is therefore a structural property of the evidence base, not neglect, and it cannot be extended by re-running anything.

Book sleeves

• AlphaForge
• AlphaMax
• AlphaTrend

Honest reading. The numeric failures are the interesting ones: they are what a real candidate cannot talk its way out of. Missing-evidence failures are work not yet done, not a verdict.

Numeric gate failures

• threshold:statistics.oos_observations:728<756
• threshold:portfolio.book_sharpe_delta:-0.006293487015595733<=0.0
• threshold:portfolio.minimum_leave_one_period_out_book_sharpe_delta:-0.1373311397081478<=0.0
• threshold:portfolio.book_expected_shortfall_delta:0.0014969796640609372>0.0
• threshold:statistics.book_deflated_sharpe:0.0<0.95
• threshold:portfolio.book_sharpe_delta_lower_95:-0.006293487015595733<=0.0

The observation gate is bounded by the BOOK. correlation_observations is capped by the common window of candidate AND book, so minimum_oos_observations constrains admission through the BOOK's history as well as the candidate's. On the window measured here that cap is 728 days, below the 756 the contract requires. This is a real property of the gate, not a defect in the candidate: a book whose own common history is short cannot admit anything, however long the candidate's record. Recorded rather than lowered -- a correlation is the quantity this whole programme turns on, and measuring it on one year is the reason the minimum was raised in the first place.

Why two counts. The production combiner aligns to a contiguous epoch-day index and treats an absent day as flat, giving a window of 1061 days. A correlation can only be estimated from days on which both series are genuinely active, which is 728. correlation_observations means the second. Reporting only the first would overstate how well the correlation is known; reporting only the second understates how long the book has run.

Check it yourself

Every figure above is read from /glassbox/admission_dry_run.json, the artifact the engine wrote. Nothing on this page is typed by hand: the page is generated from that file, so a figure that moves in the artifact moves here and a figure that is not in the artifact cannot appear here at all.

All 21 measurements · The glass box · The research library