Repository navigation
Adopt GH-146 Stern finder as default (both engines); tier5 stems get verified upper bounds - #149
Merged
Merged
Conversation
…shared finder Stern collision step (two random halves of the information set, subsets of size <= P per side bucketed by L random redundancy rows; Dumer variant with the window pivot columns as elements), Canteaut-Chabaud pivot swaps for re-randomisation, detector-parity precheck per colliding pair. Applies to the CC -ub-search finder and the MaxSAT -ub-pre pre-pass; default off (GH-140 enumeration unchanged). Defaults from S0 tuning: P 2, L automatic, W 32, B 1. Tests: make test-cc (six collision configurations on 400 halves, driver and pre-pass runs, mutants 11/12 and 1/2 in collision mode), make test-cc-flags. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…n-d, pre-pass, open codes 3600 s), S1 runner Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…uns, verified witnesses, report Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…arch (CC) (PI 2026-10-23) GH-146: the Stern collision step of the shared UbFinder becomes the default (UbParams::isd = 1, S0 parameters P 2, L auto, W 32, B 1, R 0) for both the MaxSAT pre-pass and the CC engine's finder. -ub-isd=off restores the previous finder (the GH-140 enumeration) byte for byte. README / MODIFICATIONS flag rows and the -h text updated; tests that relied on the old default now pass -ub-isd=off / -ub-pre=off / -ub-search=off explicitly, and two -v checks cover the new default on both engines. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…/ -ub-search) GH-146's 'needs the finder' check predates PR #147 (both finders on by default), so it never fired: -engine=maxsat counted the CC -ub-search default and -engine=cc counted the default -ub-pre. Now CC needs -ub-search, MaxSAT needs -ub-pre, auto either. The -seed=4 ub-pre -v check expects the stern suffix of the new default. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…nds (PI 2026-10-23) 17 tier5 codes get an upper bound in the stem; lower bounds unchanged: BB_1080_4_30to48, BB_756_16_30to34, BB_864_4_30to40, TN_288_2_9to24, TN_324_4_9to25, TN_360_4_17to24, TN_396_4_17to26, TN_432_8_9to29, TN_468_4_17to29, TN_496_2_17to30, TN_504_2_17to34, TN_540_4_9to34, TN_576_2_9to37, TN_612_2_17to40, TN_648_10_9to32 (was 9to52), TN_648_14_9to28, TN_684_2_9to29. Each upper bound is a nontrivial logical found by the GH-146 finder and re-checked before the rename with the independent pure-Python verify_witnesses.py (CSS, H e = 0, e not in the stabilizer row space, odd overlap with the logical basis): 17/17 VALID on the old and on the renamed matrices. Witness copies under the new stems: optimization/experiments/GH-146/witnesses/renamed/ (referenced from data/matrices/README.md). Same matrices (git mv; Gx/Gz header comment updated). Rename-history rows in data/matrices/README.md and NOTICE; live references updated (tier list, SERVER.md and main_server_reference.json name columns, README example). Old experiment reports, history rows and frozen raw results are unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This was referenced Oct 28, 2026
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.
Two PI-approved changes (2026-10-23). Do not merge before the PI's final confirmation.
1. Adopt GH-146's Stern finder as the default (both engines)
experiment/gh-146-isd-collisiononto main 12b3e05. Conflicts in the README / MODIFICATIONS flag rows were resolved by keeping main's PR Defaults: -ub-pre=on (MaxSAT) and -ub-search=on (CC); re-tier LP_442/BB_288/LP_714 #147 rows and adding the [Mac agent] Stern/Dumer collision step in the verified upper-bound finder (large open codes) #146 rows.UbParamsdefaultisd = 1:-ub-isd=stern(S0 parameters P 2, L auto, W 32, B 1, R 0) is now on for the MaxSAT-ub-prepre-pass and the CC-ub-searchfinder.-ub-isd=offrestores the previous finder ([Mac agent] In-house randomized upper-bound finder feeding the CC stop rule (TN_252, BB_756+ upper bounds) #140 enumeration). README, MODIFICATIONS.md and-htext updated: default stern; off = previous finder ([Mac agent] In-house randomized upper-bound finder feeding the CC stop rule (TN_252, BB_756+ upper bounds) #140).-ub-isdneeds the finder" check was written before Defaults: -ub-pre=on (MaxSAT) and -ub-search=on (CC); re-tier LP_442/BB_288/LP_714 #147 turned both finders on by default, so it never fired. It now checks the finder of the engine in use: CC needs-ub-search, MaxSAT needs-ub-pre, auto accepts either.-ub-isd=off,-ub-pre=offor-ub-search=offexplicitly. The [Mac agent] In-house randomized upper-bound finder feeding the CC stop rule (TN_252, BB_756+ upper bounds) #140 Part E finder tests intest_cc_engine.ccsetisd = 0. Two new-vchecks cover the stern default on both engines.Results (yfclab2, cores 100-123, final tree d699180):
make test-cc: PASSmake test-cc-flags: 323 checks, 0 failedscripts/smoke_test.sh: passed-ub-isd=off}: 52/52 match the known distancesoptimization/experiments/GH-146/REPORT.md.2. Verified upper bounds in the tier5 stems
ge{lb}(and TN_648_10_9to52) ->{lb}to{ub}. Each lower bound is exactly as in the current name. Each upper bound comes from a GH-146 witness that was re-checked before the rename with the independent pure-Pythonverify_witnesses.py(CSS condition, H e = 0, e not in the stabilizer row space, odd overlap with the logical basis): 17/17 VALID on both the old and the renamed matrices.git mvof Hx/Hz/Gx/Gz; only the Gx/Gz header comment changes), following data: BB_756/864/1080 -> ge30 (GH-138) #141.optimization/experiments/GH-146/witnesses/renamed/(withVERIFY.txt).data/matrices/README.mdreferences them and gives the re-check command.data/matrices/README.mdand NOTICE. They note that the upper bounds come from [Mac agent] Stern/Dumer collision step in the verified upper-bound finder (large open codes) #146 verified witnesses.optimization/baselines/SERVER.mdandmain_server_reference.json, the README naming example, and the matrices-README open item. Existing history rows, old experiment reports and the frozen raw results (optimization/baselines/raw/) are unchanged. No tests or scripts referenced the old names.QDistSAT mirror: guluchen/QDistSAT#13 (only TN_360_4 ships there).
🤖 Generated with Claude Code