Repository navigation
Expose adopted H25 full callers and original criteria as readable text - #6744
Clear1nsight wants to merge 1 commit into
Conversation
Preserve five existing PR6715 archive subjects byte-for-byte with their public origin mapping and licenses. Distinguish selected baseline/candidate from the inherited epoch driver and preserve all original activity criteria. Source access only; no original qualification, execution, performance or goal credit. AI assistance: OpenAI Codex and ChatGPT Pro; distinct GPT-6.1-sol Max export review.
|
Delivery verified for the readable-source export at exact head Hosted run 42107811590, attempt 1, finished successfully: plan job succeeded; replay was skipped. These are archive planning/onboarding checks, with no candidate or scientific execution. Canonical handoff: #6321 comment 6669340766. Short hub pointer: discussion #1166 comment 19426986. Both were read back. The subsequent Pro consultation could read these public callers and separated full-caller payload assembly from later control/runtime qualification. This PR supplies readable source access only; it earns zero additional original activity credit. No performance, full-control, runtime admission or goal-completion claim is made. Known preparation/review/consultation/publication/final metadata and historical physical costs remain OPEN_NOT_ZERO. No merge or rerun was performed. Original H25 benchmark: PR #6029. Original source investigation: PR #6321. |
Change and mathematical scope
Expose five exact subjects already published inside PR #6715's compressed source archive as ordinary readable files. A source-only Pro review could inspect the public map but could not retrieve the compressed members through its UTF-8 GitHub interface. The new
research/h25-full-caller-readable-inputs/directory resolves that transport gap with 12 files / 431,107 bytes.The selected full baseline (102,614 bytes, SHA256
9bbafe3f53febec7a984751b9e06741e06df34e2e6fa7c328c805b9eea700bd7) and full candidate (104,276 bytes,fa9f473320e9df3a2328bd8388f12eca4e7fc3eeed9289f7fdbf2eeb32a7452f) remain distinct from the inherited 89,606-byte1e8202b7…driver. The actual complete CLI and original activity document are also byte-exact. All 49 original completion-test strings in historical v16 match v19; v16's progress/status notes remain historical.Source transport only. The program logic is unchanged. Full-workload integration, complete control execution, current runtime and native qualification, resource admission, output equality, and measured performance remain open. This export establishes none of those results and makes no kappa claim.
Task and overlap
Originating H25 benchmark: PR #6029. Original source investigation: PR #6321. Immediate source predecessor: PR #6715.
Explicit user-assigned Pro consultation, implementation and upstream continuation. Base
7c49fe32c55de02c2c873655c804b16e4c3fef0a. Scope is refined to the finite readable-input dependency exposed by the consultation; the requested full-caller implementation and qualification remain incomplete.Predecessor #6715, exact head
dc34598c22e2ef43ae7e11fc1d8c14de3915cf89, supplies the public archive, SOURCE-MAP, adopted full images and original criteria.PROVENANCE.jsonrecords exact map pointers, archive members, bytes and hashes. #6736, head0f2a86fa9ef0daf760b77c662a5b762e17848598, completes the distinct retention source contract; its selected DATA driver must not replace these full scientific images. #6588, headff7df8391c3bb3f5518ce43682ba4c1e41380e3e, supplies earlier caller/preparation machinery that must be reconciled against the adopted successors. Fixed scientific baseline remains #6029e05aad954689d0ee21baff19b4674f1fcb2e838f.A fresh all-author/all-state PR delta and task-specific H25 financial-meter, full-control, batching and
run_controls_v1queries were inspected. They retain other-contributor cleanup and historical batching precedents, without a global novelty or arbitrary-diff completeness claim. No task picker, selected-frontier change or predecessor rewrite.Validation
A separate GPT-6.1-sol Max source/export reviewer accepted this exact package after 44 metadata/text checks with no findings: the exact 12-file allowlist, raw Git blobs and sizes; five subjects equal retained originals and public SOURCE-MAP rows; selected role identities; licenses/NOTICE; all 49 original completion tests; inert suffixes; and honest scope. The review is AI source/byte review, not human or proof-kernel acceptance.
From the new directory,
sha256sum -c SHA256SUMSis an ordinary byte-integrity check only. The checksum list self-excludes; the publication tree/allowlist anchors it. No candidate module or integrity helper was executed by preparation/review. No compressed archive was opened, decoded or decompressed. No compiler/native/scientific work, target inspection or previous control rerun occurred. There is no VERIFY hook or archived candidate replay registration.Original source mode 0444 is recorded separately from archival Git mode 100644. Absolute source locators are retained DATA, not executable portability or runtime authority. Dependencies are not duplicated or silently repinned. Complete scientific source history is not inferred from the local object cache. Hosted checks will be read at the actual submitted head; planning success/replay skipped supplies no scientific execution credit.
Recorded guarded policy usage remains 670.890533479 / 2,400 seconds, with no new guarded attempt or budget. Full physical/source/Pro/review/network/publication/final metadata costs and historical gaps remain OPEN_NOT_ZERO. The source/readiness window froze; this is only delivery of the exact frozen input export, not a research extension, retry, merge, automation or goal-service mutation.
Credit
Retain Chris R/Clear1nsight, Douglas R. Colkitt and original H25/source/review contributors, unpaidmoon/Cursor-Grok predecessor work, all notices from PRs #6029, #6321 and #6715, and substantial OpenAI Codex/ChatGPT Pro assistance. The exact predecessor NOTICE, Apache-2.0 and CPython licence files are copied. New wrapper metadata/documentation is Apache2; source-specific rights remain intact.
Open next questions
Read these adopted full bodies and exact original criteria before modifying the family dispatcher, independent receiver or actual full-caller selection. The inspected older orchestrator admits unit-only observations but requires complete-family observations before full execution; preserve that refusal until genuine full-family evidence exists. Reconcile any newer adopted implementation before replacing it. Full scientific comparison is assertion-on; applicable controls retain normal and optimized obligations. The selected registry/opening owner route must not acquire an unrelated manual financial-meter genesis prerequisite from a different manual route. Exact current owner/native/profile/resource/funding inputs, complete normal/O campaigns, full output equality and measured performance remain separate later gates. Canonical coordination remains PR #6321.