Repository navigation
Freeze the selected record as provenance in 138 more replay checks (no new κ) - #5417
Merged
Mister-Meeseeks merged 2 commits intoOct 24, 2026
Merged
Conversation
rank7-v45-margin and selected-record-parse-cache were missed by CrocSwap#5410 because their TASK.json author field is "jwr". Both pinned the selected record (and rank7 the family-status page) as live inputs, so every frontier change broke them. The pins become provenance, checked at base_commit by scripts/research_package.py. The parse-cache payload is still the live record; its one-digit mutation no longer hard-codes the old decimal. Results are unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
These verifiers pinned certificates/selected-result.json as a live input, so every frontier change broke them. Each SOURCES.json row becomes provenance (checked at base_commit by scripts/research_package.py), the stale live hash checks are removed, and each live read is replaced by the pinned record's value, so every downstream check is unchanged. Results are unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Mister-Meeseeks
pushed a commit
that referenced
this pull request
Oct 24, 2026
…rd (no scientific change) The frozen PR2809 loader inside inputs/pr2809.tar authenticates certificates/selected-result.json against the record pinned at this package's base, so every frontier promotion broke the replay (working source drift, flagged in #5417). Vendor those exact bytes (blob b1fea0c, sha256 bb2eaa58...) as inputs/selected-result-pinned.json and let source.py serve them to the loader from a private repository view; every other declared input is copied unchanged and still authenticated. SOURCES.json marks the selected record as provenance and VERIFY.json drops it from the live inputs. Archives, checkers and RESULT.json are unchanged; all eight registered commands reproduce the stored receipt digests. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What changes
This follows up #5410. It fixes 138 more of my research packages whose replays fail on main because a frontier file changed under them. #5410's search missed them: their
TASK.jsonauthor field isjwr, and their verifiers use different code patterns.136 packages pinned
certificates/selected-result.jsonas a liveinput. In each:SOURCES.json: that one row changes from"usage": "input"to"usage": "provenance". The blob and SHA256 stay the same.verify.py:json.loads(...selected-result.json...)is replaced by the pinned record's value,{"kappa": "836330098538459/500000000000000000"}. All 136 pin the same blob,b1fea0c9. Some packages use κ in their arithmetic (e.g.saving = kappa / (1 - kappa), "above the selected kappa"), so freezing it rather than deleting the read leaves every downstream check unchanged.VERIFY.jsonstill lists the record ininputs, so a frontier change still triggers these replays, and they now pass.2 packages needed small manual changes:
rank7-v45-margin: both the record anddocs/research/research-family-status.mdwere live inputs. Both become provenance, and their live checks are dropped.selected-record-parse-cache: the record stays the experiment's payload. Its one-digit mutation no longer hard-codes the old decimal. The README sentence is updated.RESULT.jsonfiles, results and all other pins are unchanged.Why
The research package guide says: "The original card and selected record are historical provenance in Git, so future priority changes do not invalidate a package."
scripts/research_package.pychecks provenance rows atbase_committhrough Git.Validation
All 138 packages were run on this branch (main d277c97 + 2 commits), with CPython 3.12:
python3 -B scripts/check_research_commit.py --package P: 138/138PASS_RESEARCH_COMMIT_BYTES.python3.12 -B scripts/research_package.py --package P(full registered replay, includingtest_verify.py): 138/138 pass. Before this change, all 138 failed withworking source drift.Not included
selected-odd-prime-unit-root,selected-unit-root-tape-boundaryandselected-unit-weight-reversal. They run the vendored loader ofresearch/physical-ordinary-interchange, which is another author's package. That loader validates the live selected record, and that package's own replay fails on main for the same reason. They need that package fixed first, so I left them unchanged here.pr-ledger,pr-intake-decisions,current-status,research-family-status,CONTRIBUTORS.md). Their checks quote those files' contents, so I will review each one separately in a follow-up.This is replay maintenance only. No κ, frontier or mathematical claim changes.
🤖 Generated with Claude Code