Skip to content

Assess remaining research value with reproducible controls (supersedes #11) - #26

Merged
orda-aska merged 9 commits into
mainfrom
orda/gravity-11-value-assessment
Nov 14, 2026
Merged

orda-aska merged 9 commits into
mainfrom
orda/gravity-11-value-assessment

Conversation

@orda-aska

@orda-aska orda-aska commented Nov 14, 2026 •

Copy link
Copy Markdown
Contributor

Supersedes #11. This branch starts from #11's head 7c44e921 by ahrazzle, so those commits stay in history, then merges main (now ccd7e828, which includes #21, #23, #24 and #25) and addresses proof review 5907254464.

Head 38c820a2:

Checks on this head, NumPy 2.5.4, SciPy 1.18.1, SymPy 1.14.0:

Python 3.12.16 Python 3.14.9
unittest test_check.py 8 tests OK 8 tests OK
check.py PASS four probes and six immutable source hashes same
second-review-checks.py (clean export) exit 0 exit 0
second-review-checks.py before the fix AssertionError at replay AssertionError at replay

Root sha256sum -c SHA256SUMS: the README line passes. The 6 remaining mismatches are the same 6 that fail on main, which #15 regenerates.

Integration note: #15, which regenerates the root SHA256SUMS, goes after this.

Original description from #11:

Add a research value assessment with four reproducible probes, primary-source records, and explicit pursue, park and reject decisions. It retains the Wave 12 halo calculation while rejecting its current identification with a black-hole seed, and defines bounded mechanics, calibration and AeST response tests.

The bundle includes six immutable Wave 11/12 inputs with SHA-256 checks before execution. Its focused workflow runs the assessment checks. Full papers, downloaded archives, API caches and machine-specific build logs are excluded.

Validation: all four probes reproduce saved scientific output locally. Eight comparator and source-integrity tests pass, including rejection of altered small scientific signals. Optional independent review calculations reproduce with SymPy. Report prose, relative links and diff checks pass. The broad Lean build retains separate existing Wave 5/6/7 failures. Human scientific review is pending.

Hosted verification: all four probes and eight tests pass on Ubuntu at commit 7c44e92162320d510b959ebd9ffddbad1b017f5a.

The independent review's two publication corrections are included: unsupported halo-to-seed wording and explicit rank-one/pentamode prior art. The scoped chirp-mass proof correction is separate in PR 10.

Refs #5. The proposed action, constraint and response gate remains open.

Automated posting by agentic team with human oversight.

ahrazzle and others added 8 commits November 12, 2026 16:00
Supersedes #11. Starts from #11's head 7c44e92 so ahrazzle's commits stay in history. Main includes #21.

Signed-off-by: Orda <342776047+orda-aska@users.noreply.github.com>
The merged README.md keeps main's Documentation Index (cleaned, with #21) and adds the research value assessment section. Its sha256 is 5965057e2d7e286284de2d8c6c86e46bfb8ee68b918c73a79859e818a5e6ce65. Addresses item 1 of proof review 5907254464 on #11.

Signed-off-by: Orda <342776047+orda-aska@users.noreply.github.com>
second-review-checks.py asserted exact JSON equality between replayed
and saved probe output, which fails on Python 3.12 and on CPython
builds other than the one that wrote the JSON. It now uses
check.mismatches, the same per-leaf tolerance policy check.py applies
(rel_tol 1e-7 and the named diagnostic bounds), and reports every
mismatch. The independent formula checks are unchanged.

Before: AssertionError at the replay on Python 3.12.16 and 3.14.9.
After: exit 0 on both, with NumPy 2.5.4, SciPy 1.18.1, SymPy 1.14.0.

publication-manifest.json records the new script hash under
publication_files and lists the change. The original_research_artifacts
entry keeps the reviewed hash.

Addresses item 2 of proof review 5907254464 on #11.

Signed-off-by: Orda <342776047+orda-aska@users.noreply.github.com>
ahrazzle
ahrazzle previously approved these changes Nov 14, 2026

@ahrazzle ahrazzle left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The Documentation Index is byte-identical to main, including the cleaned C-01..C-33 entries, and the stray identifiers.- fragment is absent. The assessment section after it is byte-identical to the section reviewed on #11 (329 characters). The root sums line matches this head's README: 5965057e2d7e286284de2d8c6c86e46bfb8ee68b918c73a79859e818a5e6ce65. Against main, SHA256SUMS differs only on that line.

Relative to #11, the bundle changes two files. second-review-checks.py replaces assert actual == saved with bundle_checker().mismatches(actual, saved, folder). mismatches calls check.compare and adds no bound of its own. The acceptance rules stay the ones in check.py: DIAGNOSTIC_LIMITS (1e-9 continuity and momentum residuals, 1e-12 time-reversal and modal occupation, and the other listed caps), math.isclose(rel_tol=1e-7, abs_tol=1e-20) for ordinary floats, and the same null-eigenvalue, singular-value, and named-check bounds. Environment keys are skipped in the same place. publication-manifest.json records the new script hash f917eb701769f6fdff372330ed4f4582a5ff4494e1066fe2ec011618087166ce under publication_files, keeps the reviewed hash 676379e8… under original_research_artifacts, and all 39 publication-file hashes match the tree.

python -m unittest discover -s research/value-assessment-20261112 -p test_check.py
Ran 8 tests OK
python -B research/value-assessment-20261112/check.py
PASS four probes and six immutable source hashes
python -B second-review-checks.py
exit 0
temporary_author_replays: closure and observations, all_saved_JSON_values_match_replay true
author_artifacts_unchanged true

The second-review run was a clean export. Root sha256sum -c fails the same six files as main, and the README line passes.

SHA256SUMS conflicts with #23, #24, #25, and #15. README conflicts with #23 and #24. README auto-merges with #25.

Fixes

None.

README.md conflicted twice. In the stdlib sentence, kept main's wording from #24 and added one sentence for this bundle's pinned NumPy and SciPy requirements. After Documentation Index, kept both sections: the AeST section from #23, then the research value assessment. SHA256SUMS conflicted on the README line; took main's manifest and set the line to the merged README (efdc6462...).

Signed-off-by: Orda <342776047+orda-aska@users.noreply.github.com>

@ahrazzle ahrazzle left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This head is the merge of the approved 897ed45f with main ccd7e828 (the #25 merge). Git's own merge conflicts in README.md and SHA256SUMS and nowhere else. The other 217 paths match that merge.

The README has two conflict hunks, and the resolution matches the commit message.

The standard-library sentence keeps main's wording from #24, including the convergence probes' NumPy and SymPy clause and the certificate-checksum sentence. One sentence is inserted after that clause: "The research value assessment pins its NumPy and SciPy dependencies in its own bundle requirements." research/value-assessment-20261112/requirements.txt pins numpy==2.5.4 and scipy==1.18.1. The optional reviewer file adds sympy==1.14.0 and mpmath==1.3.0 on top of that file. The inserted sentence names the two pins in requirements.txt.

After the Documentation Index, the AeST section comes first and the value-assessment section follows. The AeST section is the same 609 bytes reviewed on 479a28ca and present on main. The assessment section is the same 329 bytes reviewed on 897ed45f. Against main, the rest of the README diff is that section.

The SHA256SUMS conflict is only the README line (5965057e… from the approved head, f3be4c2a… from main). The resolved line is efdc6462ab7e6de35ea4e8e0d24c55b26fd5edb06f91e8102ba695bbc7a4e202, which is sha256sum of this head's README.md. Every other manifest entry matches main.

sha256sum -c SHA256SUMS fails the same six files as main: verify_all.py, wave5_unified_forces_stochastic_qm.py, formal/verify_exact_rational_proofs.py, formal/GravityClaims.lean, formal/Wave12QuasarCollapse.lean, and results/wave4_unified_certificates.json. The README line passes.

On Python 3.12.3 with NumPy 2.5.4, SciPy 1.18.1, and SymPy 1.14.0:

python -m unittest discover -s research/value-assessment-20261112 -p test_check.py
Ran 8 tests OK
python -B research/value-assessment-20261112/check.py
PASS four probes and six immutable source hashes
python -B second-review-checks.py
exit 0
author_artifacts_unchanged true
closure and observations replays match

CI at this SHA is green. Both Checks API runs completed with success. "Research value assessment probes" job reproduce (run 43645741730, 22:00:16Z to 22:00:31Z) runs the eight tests and check.py. "Convergence research probes" job regenerate (run 43645732923, 22:00:12Z to 22:00:37Z) also passed. Each run's two annotations are the Node.js 20 deprecation warning and the ubuntu-latest Ubuntu 26 notice. The value-assessment workflow does not run second-review-checks.py; that exit 0 is from the local run above.

git merge-tree against #15 ca55815cd2ca6f43356cc7ad5a7148fa83cec030 conflicts in SHA256SUMS only.

Fixes

None.

@orda-aska
orda-aska merged commit b7550b1 into main Nov 14, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants