Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2,602 changes: 2,602 additions & 0 deletions research/r439-block1-schedule/CANDIDATE.json

Large diffs are not rendered by default.

728 changes: 728 additions & 0 deletions research/r439-block1-schedule/CERTIFICATE.json

Large diffs are not rendered by default.

73 changes: 73 additions & 0 deletions research/r439-block1-schedule/CLAIM.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,73 @@
# Local sprint claim

- Track: bit supplier. OS assignment seed **8**, drawn with Python
`secrets.randbelow(100)` on 2026-10-13 at approximately 08:37 UTC.
- Base: `f567884aeb8851b6e794af1184119a3bfbe1e631`.
- Scope: selected PR589 R439 phase-B block **1**, `B[128:163]`, frame 507,
ordered registers `[0,3,11,61,62,63,68,69,72,465]`. This punctured ten-wire
block includes arbitrary dirty helper 465, unlike the twelve-source blocks.
- Hypothesis: exact positive-integer rescheduling of the two existing factors
(first 14 gates, last 21 gates) reduces the scalar additions while preserving
the physical output assignment and not increasing positive reverse bounds.
- Method: complete subset-state dynamic programming in each factor's
destination-once positive-addition class. No helper, output permutation,
repeated destination batch, frame change, or additional copied total.
- Deliverable: literal circuit, finite search certificate, full local arbitrary
input replay, exact complete cost, native checks if available, source pins,
negative controls and a scoped research note. No kappa claim.
- Timebox: 40 minutes from 08:37 to 09:17 UTC; orientation ends by 08:45;
reserve 09:07 onward for verification and handoff.
- Claim recorded: 2026-10-13 approximately 08:43 UTC. No pinned issue exists
(GitHub GraphQL `pinnedIssues(first:10)` returned an empty list immediately
before this claim); no separate issue opened.

## Candidate selection and overlap

Three candidates: (1) a fixed-frame rewrite of a punctured phase-B block;
(2) an A-stage single-summand transport at frame 509; (3) smaller cleanup frames
for the fixed decoder. Candidate 2 overlaps PR589's fixed-point transport and
PR843's new source sum; candidate 3 is excluded in the stated fixed-prefix model
by PR827 and has the restored-sharing obstruction in PR848. Choose candidate 1.

Diversification: apply `random.Random(8).sample([0,1,3,5],4)` to the four
punctured blocks, obtaining `[1,5,3,0]`; take block 1. This selects a different
matrix and dirty-helper interface, not merely a new seed for an old search.
The completed greedy transport search in PR589 does not search this class.
PR844 changes block 4 and PR857 changes block 8. PR728/792 change A mini-nodes.
Historical R420 block work (PR789/790/810) supplies related scheduling ideas,
but no pending code or savings are imported. PR661 fixes the gate word and
PR827 excludes smaller cleanup frames; neither excludes a scalar rewrite.

Read the contribution rules, selected certificate, current review and status,
central review ledger and contribution-review index. Current conditional
kappa is 0.001661634653669567 (PR799/589), not a historical PR title.
Initial live overlap check: recent 100 open PRs through #867, all R420/R439
search results; exact R439 search returned 11 open/closed PRs and no block-1
claim. Source PR589 head is `a5afc8c6169ea61ba7c51dfd3cade02cc7b13b46`.

Contributor: @Deftsalt-debug, with substantial OpenAI Codex assistance.

## Result and handoff — 2026-10-13, approximately 09:02 UTC

Completed: [construction, proof and replay](README.md). Block 1 falls 35 -> 30
ADDs and complete native ADDs fall 24,305 -> 24,295. Both fixed-factor optima
(14 + 16) are checked over all 2,048 states by two different implementations.
All 661 dirty-input columns, thirteen controls, identical frame paths/profile
and the reduced positive prefix pass. Both words also pass unchanged native
legality and 5/9/11-stage checks with 96 bridge controls; minimal relocated
optimized replay passes. No new kappa, unrestricted optimum or hosted invoice.

Final overlap refresh through #874 found navinor's #873 at
`117c48527dd4f066dad103121761154cf1bb39cd`, a distinct whole-block rescheduling of
frame 508. It is credited in the README; no priority or composition is claimed.
PR844 and PR857 still concern frames 520 and 523. Main remains the exact base.

Initial GitHub fork creation returned HTTP 403, `Resource not accessible by
integration`, so the first handoff was a local commit and verified patch.
After the contributor created the fork, the finished branch was successfully
[pushed to Deftsalt-debug/integer-mult-bounds](https://github.com/Deftsalt-debug/integer-mult-bounds/tree/research/r439-block1-schedule-seed8).
A publication-time overlap refresh through PR876 found no matching block-1
contribution; main remained the exact base. Both CLI and connected-app attempts
to create the upstream PR returned HTTP 403. No PR was created. The contributor
can [open the prepared comparison](https://github.com/CrocSwap/integer-mult-bounds/compare/main...Deftsalt-debug:research/r439-block1-schedule-seed8?expand=1).
Validation details are in [VALIDATION.md](VALIDATION.md).
Loading