Repository navigation
Preserve startup evidence in H25 optimized-refusal handling - #6818
Open
Clear1nsight wants to merge 1 commit into
Open
Clear1nsight wants to merge 1 commit into
Clear1nsight wants to merge 1 commit into
Conversation
Contributor
Author
|
Delivery verified for exact head Hosted run 42300241137, attempt 1 completed successfully: plan passed; replay skipped. This is an archival planning result, not a complete-caller or scientific replay. The separately retained local and package-interface pure checks passed 22/22 in each mode. Canonical handoff and remaining blocker; short contribution-hub pointer. Complete native/environment compatibility, actual scientific comparison and performance qualification remain open. The PR remains draft and unmerged. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Change and mathematical scope
Preserve the original startup evidence when the H25 optimized invocation reaches its intended assertion-enabled-entry refusal. The candidate runner previously required the output directory to be absent, although the original startup creates it before that refusal. The correction retains those artifacts and requires the original exit-1 traceback, no positive scientific output, complete capture and the existing terminal ownership checks.
The normal and optimized command profiles now match their separate full source selections. This is a scoped correctness repair and inert source handoff. Complete caller/environment qualification, actual scientific comparison and performance gains remain unclaimed.
Task and overlap
Explicit user-assigned repair against canonical base
7c49fe32c55de02c2c873655c804b16e4c3fef0a; original scope is preserved. Follow-up to #6321 and #6029, using the unchanged caller/source contracts in #6744, #6736 and #6715. Exact dependency heads and byte mappings are in the package README and SNAPSHOTS.json.The #6780 compatibility question remains open. This patch supplies the concrete original-refusal correction; it does not treat a related archive or passing CI as complete integration.
Validation
Independent AI review accepted the actual refusal/profile connection. The authenticated pure predicate passed 22 fixed controls in normal mode and 22 under optimization, including the preserved startup case and meaningful wrong-result, incomplete-evidence and limit-boundary cases. Raw reports and exact subject pins are retained.
The complete runner, scientific program, compiler and native entry were not executed. Source snapshots are inert. Full scientific payloads and admission inputs remain explicitly external; complete cold replay availability is unverified. No generic complete-science replay hook is added.
See README.md for package-local integrity and pure-control commands, adapter verification and the exact evidence boundary.
Credit
Preserves CrocSwap/H25 source, predecessor and parallel contributor attribution. Prepared with substantial OpenAI Codex assistance, GPT-6.1-sol agents and independent AI review; Apache-2.0. AI review does not constitute human or proof-kernel acceptance.
Open next questions
Establish the finite native/profile compatibility consequences, independently accept the complete caller, then obtain genuine execution admission and perform the actual baseline/candidate comparison with complete costs. The smallest next check and blockers are recorded in the package README.