- Phase 1/2 inventory and diagnosis of the restored 108-run grid
- six candidate aggregators screened on validation (seed 0)
- top-2 confirmed on new seeds 3/4; v_shrink met the pre-registered improvement rule (4/4 pairs, mean +0.0180 validation), v_cap4 did not
- report, tables, frozen plan/ranking hashes, driver scripts and tests
Server-side, read-only summary of the completed 30/30 seed-0 sweep:
queue outcome, per-run wall time and peak RSS, OOM evidence, the
reference check against the pre-fixed 2-sigma tolerance, deviations,
four analysis-script outputs, and the reproducibility runs. Does not
claim significance, recommend a beta, or commit data.
Copy analyze_sweep.py, check_noise.py, compare_with_grid.py and
compare_repro.py out of the git-ignored results/ tree into scripts/sweep/.
Replace the implicit script-location repository root with a --repo-root
argument (relative default) without changing any computation. Outputs were
verified byte-identical to the originals on a scratch copy of the data.
- pyproject.toml: add a figures extra with matplotlib (used by the two figure
scripts) so the environment is reproducible from pyproject alone.
- partition.py: correct the partition_eval_split docstring (simulated per-client
eval sets from the global split, not client-owned held-out data).
- protean_cnn.py: extract_embeddings now moves batches to the model device.
- posthoc_statistical_audit.py: refuse to write an empty table.
- README: add deep-comparison, post-hoc audit, and defense-preparation phases;
correct the not-computed list (exploratory post-hoc tests now exist, with
explicit small-n limits); add kisikisithesis/ and POSTHOC_STATISTICAL_AUDIT.md
to the structure, artifact index, key documents, and reproduction commands.
- SESSION_HANDOFF.md: reflect HEAD 34bab45 and the post-16 sessions, document
the prototype-snapshot tracking rule, refresh the next-session prompt.
willtanoe