Repository navigation
probes(D-TFLP 6.5): compact softmax data files, drop the run record - #1803
Merged
Merged
Conversation
…input run record Follow-up to #1800. Of its 43,250 added lines, 41,223 were data, mostly pretty-printed JSON with one number per line. - recipall63.json, recipall65_heldout.json and fixtures65/softmax_tables65.json are rewritten compactly (content-equal by json.load); their generators (recipall63.py, heldout65.py, fixtures65.py) now write the same form, so regeneration stays byte-identical. softcert65.py writes its record compactly. - softcert65_run.json (17,880 lines) is removed: it is the gate's output, read by nothing on main. SOFTMAX65.md says how to regenerate it and where the landing copy lives (26666aa). - manifest65.json re-stamped: only the tables file digest changes; the per-table digests (canonical form) do not. Checked: fixtures65.py regenerates every fixture file byte-identical to the committed ones; replay 11 tests pass and --full passes 529+27+7+10; a full softcert65.py --mutations re-run gives identical decisions, witnesses, unreachable pairs and mutation verdicts. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01ArVbbq3DsToBM7r79zGeEf
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars.
 |
AdaWorldAPI
marked this pull request as ready for review
October 26, 2026 20:59
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
AdaWorldAPI
pushed a commit
that referenced
this pull request
Oct 26, 2026
…#1802 - PR_ARC_INVENTORY: compact entries for #1772, #1774–#1787, #1794–#1803, recorded from merge commits, changed files and each PR's own Status line. #1773 and #1793 are board-hygiene-only and get none. - INTEGRATION_PLANS: one index row for thirteen plans merged without one (D-NAM-1..5 + PZL, DT64-1..5 + QB pointer, D-ONNX-INTAKE-0). - LATEST_STATE: "main after #1803" section. - d-onnx-relalg-cap-v1: OP-ONNX-2 status cell names #1802's Burn onnx-ir vs Tract proposal; still OPEN. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VncKoyG6JkpsoqhqTGTv3f
AdaWorldAPI
pushed a commit
that referenced
this pull request
Oct 28, 2026
Brings the research branch up to main eeaaafb (136 commits) without rewriting its history: the audit and ledger cite this branch's hashes. Conflicts were all files that landed on main in hardened form through #1800/#1801/#1803 (table-identity check, compact data, pinned-codec-set claim); main's versions were taken. The research-only run record softcert65_run.json stays here as evidence. The audit entry heading now reads D-TFLP-AUDIT so its index id no longer collides with main's D-TFLP 6.5 landing entry; the entries index is regenerated. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01ArVbbq3DsToBM7r79zGeEf
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.
Why
#1800 added 43,250 lines, and 41,223 of them were data, mostly JSON pretty-printed one number per line. That made the diff look like code and buried the roughly 1,800 lines that actually were code.
Scope
Net change: +25 / −40,595. No change in content.
recipall63.json,recipall65_heldout.jsonandfixtures65/softmax_tables65.jsonnow each fit on one line. They are content-equal to the old files underjson.load. Their generators (recipall63.py,heldout65.py,fixtures65.py) now write the same compact form, so regenerating them stays byte-identical.softcert65.pyalso writes its run record compactly.softcert65_run.json(17,880 lines). It is the gate's output, and nothing onmainreads it.SOFTMAX65.mdnow explains how to regenerate it and that the landing copy is in26666aa0f(probes(D-TFLP 6.5): softmax fixture replay + certified preparation gate #1800).manifest65.json. Only the tables file's own digest changes; the per-table digests are computed over a canonical form and do not change.Tests run
fixtures65.pyregenerates every fixture file byte-identical to the committed ones, including the compacted tables file.replay: 11 tests pass, and--fullpasses 529 + 27 + 7 + 10 rows.softcert65.py --mutationsagainst TFLM5f64e4adgives identical decisions, witnesses, unreachable pairs and mutation verdicts. The identity controls still pass.What this does not change
Results, thresholds, code paths and claims are all unchanged.
Production impact / rollback
None; these are lab files only. Rollback is reverting this one commit.
Links
Follows #1800 (merged) and #1801.
Generated by Claude Code