[mihir-s-05]

The selected #4359 certificate uses productive-helper budget stages. The documented historical baseline authenticates its sources, then crashes on KeyError: 'bit_lower'. This adds an explicit successor command that reads that schema and reports its compiler, ordinary/admitted BIT and nine-row outer budgets separately.

python3 -B scripts/research_selected_baseline.py

The new command authenticates the entire selected closure with the existing checker before printing success. It reports the exact outer minimum 115098849/50000000000, binding row main_transform, and strict slack 98849/50000000000 over selected 23/10000. It retains symbolic template ranks and does not infer the old balanced-transfer cap.

Only three files are added: the command, 13 focused tests and usage/scope documentation. All existing files, scientific inputs and selected records remain unchanged. #3725 documents why the old helper and its pinned documentation should not be edited casually. #4436 already owns the separate current-status/CI navigation repair.

Validation at 108509a, Windows / CPython 3.12.0:

• All 13 focused tests pass. They exercise exact stage labels, ties, unsupported schemas, duplicate/missing templates, invalid budgets, omitted outer rows, inconsistent minimum/slack and nonstrict endpoints. Synthetic complete source closures exercise the real checker and CLI in ordinary and optimized Python, including altered sources/proofs and an authenticated but inconsistent certificate.
• The actual selected source checker passes; real-source ordinary and optimized reports are byte-identical. Selected-record SHA256: 5732707cc1e460b090b53f2463162749ad860de878a18f7408938671d295a9d9.
• Changed-package planning selects zero existing packages. Committed-byte comparison for all three added files and git diff --check pass.
• The broader Windows/sparse onboarding run is not green: 331 tests, 8 failures, 32 errors, 15 skips. Diagnostics include Windows-invalid newline filenames, locale/broken-pipe issues, one Bash timeout, omitted unrelated sparse fixtures and the unchanged status assertions. The four status/navigation failures were separately reproduced on unchanged main before implementation. This PR does not claim a complete onboarding or hosted CI pass.

Focused reproduction:

python3 -B -m unittest discover -s tests -p test_research_selected_baseline.py -v

Scope and coordination: portable-replay, priority revision 7, explicitly changed from an environment mismatch to the selected certificate-format mismatch in the card's startup command. Canonical main started and refreshed at 2a13eb1827784206202db90e5dc8327cf8a250db. Discussion #4458 is the canonical investigation. Submission overlap coverage: all 5,731 tracked local note paths, 314 open PR bodies, 3,313 Discussion records and 3,787 closed PR bodies, complete pagination at 22:52–22:54 UTC. This is bounded text coverage, not a novelty claim or exhaustive diff search.

Limitations / next task: this is source binding and metadata consistency, not native replay, supplier admission or a theorem proof; no new kappa. The pinned catalogue/brief still name the historical command. Their migration needs a source-receipt policy; the new usage note supplies the replacement now. Future certificate schemas need explicit adapters. No selected cold replay was run.

Credit Douglas Colkitt and the retained checker contributors; zxv, Jacob Alan Sussman and the full #4359 contributor map for the scientific data; Mister-Meeseeks and the credited Claude agent in #3725 for the immutable-helper precedent. Prepared by Mihir Srivastava (@mihir-s-05) with substantial OpenAI Codex assistance; no subagents. Apache-2.0.