Repository navigation
Bind delayed-cover batches to recursive allocation and return - #5829
Merged
Mister-Meeseeks merged 4 commits intoOct 27, 2026
Merged
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The delayed-cover allocation boundary now has an explicit whole-batch address and return map. A fresh row consumes
W=L*Hstock values, but each child retains all H high matrices and has volume V/L, not V/W. At depth j the retained batch volume is V/L^j while unused stock is P/(L*H)^j. The resulting active-chain space is belowK*V*L/(L-1) <= 2*K*Vunder the stated local machine premise.The source-bound proof applies to the final fixed compiled template. It preserves ancestor high matrices, complete child stock, arbitrary payloads and return addresses. Parent-dependent weights remain independent of the child's fresh cover. Private banks increase local space without receiving productive credit. The cover, fresh-cylinder counting theorem and fixed-PR589 packet implementations retain predecessor credit; they are not counted as new constructions.
Validation: 186,252 complete all-depth addresses against a separate coordinate-product oracle; literal nested arbitrary-label descent/return; all 7,776 two-role GL2(Z/9Z) cover rows; 31,104 route/spectator checks; ancestor-weight and exact geometric checks; seven rejection controls. Registered Python3.12, committed-byte/mode, source-corruption and read-only Git-free optimized replay pass. Full-history preflight was not run in the shallow checkout. Hosted CI and independent review remain pending.
This closes a conditional geometry implication named by PR5814. It is a reference address model, not a physical dispatcher. Old atom-interchange, routing/fallback local space, generated descriptor evaluation, numerical children and private return remain implementation obligations. All five project assumptions retain their status. No new kappa or selected-frontier change.
Canonical claim and handoff. PR5814 source:
68614094a31bd483083cfaf1f8735ceb6aec5224. Credit Mason Hale, Nick Petalas, Jose Manuel/jmbarrancoml, Cinna Davis, David Arias, zxv and all retained BIT/compiler/cover sources. Ali Romman with substantial OpenAI Codex assistance; Apache-2.0.CI dependency: unchanged PR5761 test-only repair
ba8a9e640e1983f3c29993fefa3f46e8ed03313a, cherry-picked with original authorship and provenance. No research artifact or selected claim is changed by that repair.Clean-checkout source repair: the initial hosted job could not resolve pending predecessor commits. SOURCES.json now binds the exact retained input copies in this PR history; retained_predecessors preserves every original head, path, blob and SHA256 plus the retained file/archive member. Original and retained bytes were checked equal. Metadata, committed bytes/modes and declared-pin ancestry checks pass. Source bytes, mathematical assertions and scientific receipts are unchanged. The full-history preflight remains unrun in this shallow checkout. Hosted replay is queued for the new head.