Repository navigation
Fix calibration scaling and complete the bounded release-preparation pass - #33
Conversation
Fix the valid-small-scale calibration exception using exact scale-aware dyadic square-root enclosures and an explicit precision-limit result. Add 3082 regression controls, a current verification entry point, and pinned reproduction instructions. Preserve prior saved reports while recording revised implementation fingerprints separately. Clarify current, historical and pending material; retain the Gaberdiel-first physics tutorial and Hopf-style math presentation. Add citation metadata, a source map, archive index, changelog and release scope. Desired About-sidebar settings are documented, not claimed applied. The canonical proof mathematics, comparison thresholds, license, numbered historical notes, provenance and original replay policy are unchanged. The full declared 33-path delta and candidate tree were verified against ecd0623; the exact current checks pass in normal/-O/-OO on a GitHub runner. Complete PR checks remain required before merge. No helper transport files, manuscript, tag, release or unrelated branch is included.
… edit Remove only the proposed STATUS.md banner and keep that original file fully byte-identical. Its historical role is identified in the archive index instead. Protect STATUS.md explicitly in the new integrity guard; keep the original provenance verifier and all historical report checks unchanged. The corrected local tree c7c7ae1 matches the connector-created tree exactly. Current exact/default checks and original provenance verification pass; full PR checks will be repeated. Final delta is 32 paths, with 96 baseline files unchanged. No science, numerical tolerance, or historical evidence was relaxed.
Final corrected candidate: runner evidence inspectedHead The current job Exact current reports matched all four registered fingerprints: extraction63 checks ( Package checks:21 active pages,473 expressions (85 display and388 inline),177 relative links,22 malformed-input controls,8 valid-input controls and82 retained teaching checks. The actual runner parsed all473 expressions with MathJax3.2.1. The separate preserved-history job also completed its unchanged frozen-evidence workflow successfully. Historical strict byte replay still FAILED on the current runner, with the recognized prior-average-toy report mismatch. The unchanged bounded-portability comparison PASSED with98 differing floating-point leaves and no byte-identity claim. The exact final status is Final delta is32 paths:14 edits and18 additions, with96 other baseline files preserved. Canonical proof mathematics, original LICENSE, all numbered research notes, all provenance archives, and all pre-existing result files are intact. No manuscript, tagged release, website build, pending scientific branch, or helper transport file is included. About-sidebar values are prepared in metadata but not claimed applied. These implementation/layout checks are not live GitHub rendering, an independent human review, or verification of imported VOA classifications. |
Post-merge main verification inspectedConfirmed main at The actual main checkout passed all 20 current checker entry points in normal/-O/-OO (60 executions), with no current failures and matching output across modes. All four registered current exact-report fingerprints matched, including the new 3082-check scale-regression suite. Integrity verified 128 files against the pre-release baseline. The reader check reports 473 expressions (85 displays and 388 inline), 21 pages, 177 relative links, 22 negative controls, eight positive controls and 82 retained teaching checks; all 473 exported expressions parsed with MathJax 3.2.1. Unlike the earlier PR runner documented above, this post-merge current job reports strict historical replay PASS and bounded portability PASS, with zero differing floating-point leaves and byte-identical reports. This is a runner-specific result, not a repair or erasure of earlier strict failures and not a guarantee of cross-platform byte identity. No report or tolerance was changed during this continuation. The runner retained stdout/stderr as artifact Repository-content release preparation is integrated. The live repository GET still reports an empty About description and topic list; |
1 participant
Bounded release preparation
Base
ecd0623e9de41b49754b6ca16dc23752443143af; current head80e1ec3bc1362e44babd72dfaf5fb7f102c5c974; exact current treec7c7ae165a8754c370120644f6b233198432914a.Implements the owner-authorized R01–R05 repairs. No release, tag, manuscript, DOI, outreach, or pending scientific PR is included.
Calibration reliability
The fixed absolute root grid raised an exception for valid fields scaled by 2^(-39). Scale-aware exact integer/Fraction enclosures now keep positive radicands' lower bounds positive. A typed grid-work limit returns an explicit inconclusive result, not a false certificate or swallowed programming exception. All seven-scalar identities, physical premises and comparison thresholds are unchanged.
The original 74 calibration and 23 adversarial controls run against the repaired implementation. A new 3082-check suite covers tiny/large and independently scaled fields, finite-width intervals, stress shifts, zero norms, signs/duplicates/cap failures, and precision limits. New report fingerprints are separate; all pre-existing saved reports remain untouched.
Reproduction and reader access
python verify_current.pyruns the current exact/package checks with the standard library.--alladds the complete checker inventory and reports strict historical replay separately from bounded portability.--require-strict-replaymakes a strict mismatch fatal. A pinned reproduction requirements file and guide describe the tested environment.The gate-testing and physical-selection ledgers gain historical banners without changing their bodies. The original seed-protected
STATUS.mdstays wholly unchanged and is explained by the archive index. Missing inline paths now link to pinned PR22/23/27 sources as provenance or pending work, not omitted core premises. All fourteen numbered notes and archived evidence remain unchanged; legacy rendering limitations are explicitly indexed. Gaberdiel stays primary, Yamauchi secondary, and the five physics lessons/ten self-checks stay unchanged.Added citation metadata,
llms.txt, changelog and release scope..github/repository_metadata.jsonrecords desired About values; it does not itself apply sidebar settings, and no settings-write success is claimed.Integrity and validation
Final delta: 32 paths, 14 intentional edits and 18 additions. All 96 other baseline files remain byte-identical; the current inventory is 128 files. Canonical proof mathematics is identical expression by expression. The workflow retains all historical evidence on frozen trees and separately checks the real release candidate. No tolerance or historical report has been replaced to manufacture a pass.
The first head
8055bf9failed full CI because its proposed STATUS.md banner violated original seed preservation. The correction restores the exact original, explicitly protects it in the new checker, and records the regression inaudits/release_preparation.md. The old provenance checker and replay policy were not altered. Corrected local provenance/default checks pass; exact local and connector trees match. Full CI is rerunning on the current head and will be inspected before merge.Local rendering: 473 expressions parsed with MathJax 3.2.1; 21 active pages were checked at 1060 and 430 pixels. This is not live GitHub rendering or a mathematical proof audit. A full local --all attempt hit the execution time limit; actual full-run results are taken from GitHub Actions, not presumed.
Transfer provenance
A read-only Actions artifact supplied the exact baseline source, whose Git tree was verified locally. The initial candidate was transferred through a SHA256-pinned textual patch. Its helper runner passed current checks and validated all33 initial blobs but could not create the workflow-containing tree (HTTP403); the authorized connector created that exact tree instead. No token scope was changed. The final correction used direct connector file/tree writes. No helper/transport files are in this PR.