Korthals protocol-v2 first-scientific-execution gate

Scope

This tranche prepares, but does not run, the first scientific AOI execution for the Korthals target-tracking case study. It is built only after the public source was successfully frozen and independently verified and after the resulting source identity was merged as an exact-main-certified repository lock.

At qualification time no Korthals hard AOI effect, Monte Carlo distribution, sign probability, or scientific classification has been executed or inspected.

Certified prerequisites

The execution workflow is anchored to:

protocol-v2 fingerprint 1cd4150c9341db145b8f6aee544f9634c67c46116e61e7e675ae3c50282d2d55;
source-lock fingerprint 8bcd9e1dfc8d6ca07c6222a20f563bf8b3bd67aed0de0d3b0e97628799de3a08;
source-manifest fingerprint 04dd531fb5e0eaa8aa77cb3743d7dadfd4c218895f7edf8186cc3ff9362b7541;
companion commit 1d7ebec23e3fe20f6db952eefda1a0dd58ae09da;
successful source-freeze run 34719412734, artifact 10305414874;
source-freeze critical environment Python 3.12.14, NumPy 2.5.3, pandas 2.3.3 and SciPy 1.18.1.

Fail-closed execution order

The manual-only korthals-scientific-execution workflow runs on main only and uses this order:

checkout and verify the exact GITHUB_SHA;
checkout the exact frozen companion commit;
recreate and assert the source-freeze-matched critical Python stack;
download the complete public OSF source using the already-certified resumable transfer path;
checksum the current source and require exact equality with the packaged source lock before companion preprocessing;
run memory-bounded companion preprocessing and protocol-v2 preparation;
require exact reproduction of the locked participant/split identity, 448,336 prepared rows, 914 retained trials, 12 validation groups, missingness policy and the exact zero-finite/matched-cell records before the endpoint call;
evaluate the unchanged frozen AOI endpoint and 2,000-draw measurement-error Monte Carlo audit without printing scientific results;
write a checksummed execution archive bound to the source lock, exact execution commit, run ID and dependency snapshot;
verify both archive integrity and the locked endpoint-blind intake facts;
upload the scientific artifact with compression disabled; and
only after successful upload, verify again and reveal the archived classification, summary and scientific fingerprints.

Thus source identity is checked twice before scientific evaluation: once against the fresh byte manifest and once against the fully prepared endpoint-blind representation.

Scientific invariants

The protocol-v2 execution reuses the frozen endpoint semantics without modification: target-centered circular AOI radius 1 dva; jumping-minus-moving matched-cell contrast; unweighted averaging over complete cells within participant and participants within study; validation-block error_avg converted to isotropic Gaussian sigma by the frozen model; zero bias; 2,000 Monte Carlo draws in batches of eight; seed 20260316; 95% measurement-model interval; reference zero; and the predeclared robust-positive, robust-negative, or measurement-sensitive interpretation rule.

No interpolation, sample recovery, phase shift, cohort change, matched-cell change, validation filtering, seed change, error-model change, AOI change, endpoint change or post-outcome adaptation is introduced.

Qualification boundary

CI may execute the protocol-v2 scientific functions only on synthetic fixtures whose known construction exercises the existing endpoint semantics. The real OSF workflow has workflow_dispatch as its only trigger; opening or merging this tranche cannot start the real scientific execution. A separate manual action on a subsequently exact-main-certified commit is therefore required for the first public-data outcome.