Avatar for the InauguralSystems user
InauguralSystems
EigenScript
BlogDocsChangelog

Performance History

Latest Results

ci: merge strict and gfx differentials; observer arm in jit_diff (#1275) (#1318) * ci: merge strict and gfx differentials; observer arm in jit_diff (#1275) Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KC99CmwatKssQYggkBCgwF * ci: drop spent pixel waivers; OBS arm uses the full corpus (#1275) Pixel identical-when-off no longer waives zero-variant rows, so a gfx build compared with itself is green and a flag-off canvas change is red. The OBS arm walks every tracked .eigs file (floor 500) and the ledger keeps the two gate-closed rows as comments. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KC99CmwatKssQYggkBCgwF * ci: ledger the one unadjudicable OBS row with its reason (#1275) Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KC99CmwatKssQYggkBCgwF * ci: jit_diff normalises only address-shaped hex (#1275) The observer arm inherited the JIT arms' normaliser, which folded every hex literal to 0xADDR, so a gate fault turning a printed 0x1 into 0x0 compared equal. The deleted observer oracle compared raw bytes and caught it (blind critic, round 2). Addresses are 8+ hex digits; shorter hex is data. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KC99CmwatKssQYggkBCgwF * ci: unique OBS row names for nested fixtures; visible gfx SKIP; doc drift (#1275) From /code-review of #1318: nested tests/ fixtures shared ledger names (peer.eigs x10), so one ledgered row could mask another program; [99s] dropped the tool's 'not a gfx build' SKIP on a pass; BUILTINS.md still described the deleted waiver proof; the retired timing block was half-edited. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KC99CmwatKssQYggkBCgwF * ci: waive the [99s] non-gfx sub-check SKIP line with its reason (#1275) Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KC99CmwatKssQYggkBCgwF --------- Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
main
58 minutes ago
ci: delete variant section plans; shard the ASan suite on the http+model build (#1275) (#1310) * ci: run full variant suites and shard the ASan HTTP build Remove variant section plans and capability markers. Run the complete suite on every code event for each variant, and consolidate the sanitizer workers on three shards of the asan-http build. Keep shard coverage, receipts, leak tally, and one owner for each ASan extra. Retain section_plan's chunk, weight, shard, and skip-audit checks in 769 lines. Gates: section_plan selftest 16/16; shard union 443 chunks across 3/3 disjoint shards; skip audit 24 emitters and 24 routes; release shard 2: 2372/2372, 0 failed, 3 skipped, 118 section headers; GC traversal 1 asan-http variant; ci-tier 21 jobs and 19 required names; workflow YAML 9 files/213 names; pipefail 23 scripts; child-exit 94 sites; portability 125 files and 3/3 runtime gates; docs claims PATHS 222, FLAGS 45, TARGETS 34, NAMES 463, enrolment 12. Union, disjointness, sync, and skip-audit mutants each made the selftest red. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KC99CmwatKssQYggkBCgwF * Restore independent shard section counts and refresh ASan weights Derive each shard's expected section headers from selected runner text and the last-built binary's Makefile flags before execution. Refuse a mismatch with headers printed on the run's stdout. The critic's 148-section hidden-body plant now exits 1 after executing one section; a real release shard printed 124/124 headers and finished 3433/3433 passed, 0 failed, 8 skipped. Change chunk and skip-audit population pins to nonvacuous floors. Keep the bare-SKIP routing rule, add the hidden-body selftest, and regenerate weights from the three green asan-http shard logs (run 36102915685, head 05bda90). Rewrite the Unreleased CI entry for full variant suites, ASan sharding, and skip auditing; correct the runner's witness and [99w] comments. Gates: section-plan selftest 18/18; shards 443/443, union full, disjoint, 1 unmeasured alternate label; skip audit 24 emitters/24 routes, 0 unaccounted; critic control rc 0 and hidden fault rc 1; copied skip-audit gutting rc 1. CI tier 21 jobs, workflow YAML 9 files, pipefail 23 scripts, child exit 94 sites, portability 125 files and 3/3 gates, docs claims 222 paths/45 flags, and release shard 2/3 all passed. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KC99CmwatKssQYggkBCgwF * Witness shard execution by chunk boundaries, not build capabilities Remove the planner's Makefile flag and skip-message mirror. Mark selected chunks with literal section-header echoes, emit a wrapper-owned sentinel before each chunk, and require every sentinel once in order plus a visible header inside every bearing chunk. Report bearing=N in PLAN and require a positive bearing count in sanitizer receipts. Filter sentinels from displayed output without buffering long sections. Update CI docs and Unreleased wording to describe the structural check. section_plan.sh shrinks from 849 to 811 lines. No extension flags or skip message spellings remain in its execution witness. Gates: section-plan selftest 19/19; copied witness gutting rc 1 with two plants surviving; critic control rc 0 and hidden fault rc 1; shard check 443 chunks, full union, disjoint; skip audit 24 emitters/24 routes, zero unaccounted. Release shard 2/3: 3433/3433 passed, 220/220 sentinels and 124 bearing chunks; shard 1/3: 1/1 passed, 1/1 sentinel and bearing chunk. CI tier 21 jobs, workflow YAML 9 files, pipefail 23 scripts, child exit 94 sites, portability 125 files and 3/3 gates, docs claims 222 paths and 45 flags all passed. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KC99CmwatKssQYggkBCgwF * Make shard headers unconditional and preserve witness output Hoist the primary header in every formerly conditional-only chunk before its capability branch. Route missing model and network support through section_skip. Reject future indented-only header chunks statically in shard planning and exercise that rejection with a plant. Keep shard sentinels visible by removing the display filter, and restore the prior exit status after each wrapper sentinel. Remove current-population counts and the deleted marker reference from planner comments. Gates: section-plan selftest 20/20; shape-gutted mutant 1 failure; shard partition 455 chunks, 441 groups, union full, disjoint yes, unmeasured 1; skip audit 24 emitters, 26 routes, unaccounted 0; release shards 1/3, 2/3, 3/3 green; http+model shard 41/64 green; critic2 minimal/real/display/status reproducers green; ci-tier, workflow-yaml, pipefail, child-exit, portability, docs-claims green. Closes #1314 Closes #1315 Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KC99CmwatKssQYggkBCgwF * Unify shard header recognition and print one truthful heading Validate complete single-line source header echoes against one shared text predicate. Derive bearing metadata from that scan, pass the same predicate to the runner witness, and refuse wrapped or conditional-only source headers. Add wrapped and column-zero branch plants; a gutted shard-check caller turns the selftest red. Print one build-neutral heading for each hoisted section. Keep section_skip and assertion accounting, remove duplicate skip headings and their audit waivers, and document visible chunk sentinels. Gates: section-plan selftest 22/22; gutted caller 3 failures; shard union full/disjoint over 455 chunks, unmeasured 0; skip audit 16 emitters/26 routes, unaccounted 0; release shards and HTTP 41/64 shard green; ci-tier, workflow-yaml, pipefail, child-exit, portability, docs-claims, suite-label green. The release shards still sum to 5318 passed and 13 skipped; removing the unmeasured duplicate heading rebalanced 66 checks and one skip between shards 2 and 3. Closes #1316 Closes #1317 Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KC99CmwatKssQYggkBCgwF * ci: section_plan's self-test runs when the runner or ci.yml changes (#1275) The self-test plants faults in tests/run_all_tests.sh's shard witness, cuts a stub runner out of that file, and reads ASAN_SHARDS from ci.yml, but tools/selftests.txt selected it only for tools/section_plan.sh. Found by /code-review on #1310. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KC99CmwatKssQYggkBCgwF --------- Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
main
9 hours ago

Latest Branches

CodSpeed Performance Gauge
0%
ci: merge strict and gfx differentials; observer arm in jit_diff (#1275)#1318
2 hours ago
ae8934c
gates-1275-strict-merge
CodSpeed Performance Gauge
0%
CodSpeed Performance Gauge
0%
© 2026 CodSpeed Technology
Home Terms Privacy Docs