feat(igla-coder): record the pilot's measured t27 result (Closes #1041)

[gHashTag]

Records the igla-coder-gpu pilot's measured result into specs/igla/coder/pipeline.t27 — the spec the WARS arena's IGLA CODER lane (lane 03) reads. This is the "one form": the corpus is .t27 specs, the pilot is their execution, and the WARS arena reads the result back.

Closes #1041

What landed

The IGLA coder pipeline had no measured checkpoint, so WARS lane 03 shows "no executable checkpoint verified." The pilot now provides one, judged by t27c (the exact compiler judge):

• checkpoints: fp100m + t27 fine-tune, tern100m + t27 fine-tune (100m params, 2.5B pretrain tokens, 3-epoch .t27 fine-tune)
• strict set (391 items, both constant mutants fail): fp 8/391, tern 5/391 solved within 10 tries
• all 466 items: fp pass@10 5.8% (27), tern 5.4% (25); compile@1 12–14% is the wall
• t27 val bits/byte: fp 0.4589, tern 0.4957
• source: gHashTag/igla-coder-gpu research/T27_BENCH.md

Honesty limits (BINDING)

• The checkpoint is not yet published with a verifiable SHA, so the live WARS lane stays unregistered until it is. Flipping the tab additionally needs the checkpoint published and an arena run registered.
• A 100m model solves ~2% of held-out functions in ten tries.
• This is a software result judged by t27c, not a hardware or deployed-network one.

🤖 Generated with Claude Code

[github-actions]

📓 NotebookLM Notebook linked to this PR

• Notebook ID: b83263109fb055dc
• URL: https://notebooklm.google.com/notebook/b83263109fb055dc

This notebook contains session context, decisions, and artifacts for this work.

[github-actions]

PR Dashboard

Generated at: 2026-09-24 17:45:51 UTC

• feat(igla-coder): record the pilot's measured t27 result (Closes #1041) #4796 feat(igla-coder): record the pilot's measured t27 result (Closes #1041) | Created: 2026-09-24 | Updated: 2026-09-24 | 24 failing / 31 total
• Bare for statements: 2 loops recovered, and 235 of the 240 sites I cited were out of scope #4777 Bare for statements: 2 loops recovered, and 235 of the 240 sites I cited were out of scope | Created: 2026-09-24 | Updated: 2026-09-24 | 10 failing / 31 total
• Generic type parameters dropped from the declaration and kept in the body: -23 errors #4745 Generic type parameters dropped from the declaration and kept in the body: -23 errors | Created: 2026-09-24 | Updated: 2026-09-24 | 12 failing / 31 total
• Implement the 8 empty function bodies in specs/vsa/packed_vsa.t27 #4729 Implement the 8 empty function bodies in specs/vsa/packed_vsa.t27 | Created: 2026-09-24 | Updated: 2026-09-24 | 12 failing / 31 total
• Wave 684: occlusion four deep, and the two files no list ever reached #4660 Wave 684: occlusion four deep, and the two files no list ever reached | Created: 2026-09-23 | Updated: 2026-09-23 | 10 failing / 31 total
• Capture given/when/then and render them: 6540 clauses, and 169 specs share one unfilled placeholder #4642 Capture given/when/then and render them: 6540 clauses, and 169 specs share one unfilled placeholder | Created: 2026-09-23 | Updated: 2026-09-23 | 10 failing / 31 total
• Implement part 2 of 2 of the empty bodies in specs/git/diff.t27 (7 functions) #4634 Implement part 2 of 2 of the empty bodies in specs/git/diff.t27 (7 functions) | Created: 2026-09-23 | Updated: 2026-09-23 | 10 failing / 31 total
• feat(specs/skills): trinity/connect-accounts #4600 feat(specs/skills): trinity/connect-accounts | Created: 2026-09-22 | Updated: 2026-09-22 | 7 failing / 32 total
• Port part 1 of 2 of tools/check_withdrawn_live.py to specs/port/tools/check_withdrawn_live.t27 (8 functions) #4581 Port part 1 of 2 of tools/check_withdrawn_live.py to specs/port/tools/check_withdrawn_live.t27 (8 functions) | Created: 2026-09-22 | Updated: 2026-09-22 | 9 failing / 31 total
• Port part 1 of 2 of tools/check_coq_in_build.py to specs/port/tools/check_coq_in_build.t27 (8 functions) #4578 Port part 1 of 2 of tools/check_coq_in_build.py to specs/port/tools/check_coq_in_build.t27 (8 functions) | Created: 2026-09-22 | Updated: 2026-09-22 | 8 failing / 31 total
• Implement the 1 empty function body in specs/test_framework/verilog_bench_harness.t27 #4565 Implement the 1 empty function body in specs/test_framework/verilog_bench_harness.t27 | Created: 2026-09-21 | Updated: 2026-09-21 | 0 failing / 2 total
• Implement part 2 of 2 of the empty bodies in specs/vm/jit_semantics.t27 (4 functions) #4563 Implement part 2 of 2 of the empty bodies in specs/vm/jit_semantics.t27 (4 functions) | Created: 2026-09-21 | Updated: 2026-09-21 | 8 failing / 31 total
• The C tests handed cc a clang-only flag: 23 red on gcc, and the silent half compiled nothing #4551 The C tests handed cc a clang-only flag: 23 red on gcc, and the silent half compiled nothing | Created: 2026-09-21 | Updated: 2026-09-22 | 5 failing / 44 total
• classify: the browser could not tell a Markdown file from a broken spec (refs #2133) #4541 classify: the browser could not tell a Markdown file from a broken spec (refs #2133) | Created: 2026-09-21 | Updated: 2026-09-22 | 7 failing / 41 total
• Implement the 1 empty function body in specs/test_framework/graph_drift_detection.t27 #4536 Implement the 1 empty function body in specs/test_framework/graph_drift_detection.t27 | Created: 2026-09-21 | Updated: 2026-09-21 | 8 failing / 31 total
• Nothing builds bindings/wasm-explorer, which is how the artifact it replaces drifted in the first place #4498 Nothing builds bindings/wasm-explorer, which is how the artifact it replaces drifted in the first place | Created: 2026-09-21 | Updated: 2026-09-21 | 6 failing / 30 total
• specs/functions: render-job-run, morph-images-generate -- a failed delivery is not delivered #4123 specs/functions: render-job-run, morph-images-generate -- a failed delivery is not delivered | Created: 2026-09-17 | Updated: 2026-09-17 | 4 failing / 30 total
• specs/functions: payment-ai-server-process -- claim-invoice before crediting #4116 specs/functions: payment-ai-server-process -- claim-invoice before crediting | Created: 2026-09-17 | Updated: 2026-09-17 | 4 failing / 30 total
• specs/functions: crm-proactive-sweep -- one step per seller after the 502 runs #4113 specs/functions: crm-proactive-sweep -- one step per seller after the 502 runs | Created: 2026-09-17 | Updated: 2026-09-17 | 4 failing / 30 total
• specs/functions: withdraw five functions from the registry (code-only/unregistered) #4038 specs/functions: withdraw five functions from the registry (code-only/unregistered) | Created: 2026-09-17 | Updated: 2026-09-17 | 4 failing / 30 total
• Feed the swarm from CI, with fuel that does not run out #3978 Feed the swarm from CI, with fuel that does not run out | Created: 2026-09-17 | Updated: 2026-09-17 | 6 failing / 28 total
• specs: bind every tool and skill to the Queen, and name a domain lead #3833 specs: bind every tool and skill to the Queen, and name a domain lead | Created: 2026-09-15 | Updated: 2026-09-15 | 4 failing / 30 total
• agent-provider-chain: probe 4 of the reserve, 402 with max_tokens bounded #3683 agent-provider-chain: probe 4 of the reserve, 402 with max_tokens bounded | Created: 2026-09-15 | Updated: 2026-09-15 | 0 failing / 30 total
• specs/automation: inngest-functions-status.t27 (lastRun без проб, lastProbe отдельно) + inngest-probe-suite VERSION 2 #3593 specs/automation: inngest-functions-status.t27 (lastRun без проб, lastProbe отдельно) + inngest-probe-suite VERSION 2 | Created: 2026-09-13 | Updated: 2026-09-13 | 5 failing / 30 total
• fix(ci): make the JSON gate strict, and make the one file it refuses legal #3395 fix(ci): make the JSON gate strict, and make the one file it refuses legal | Created: 2026-09-07 | Updated: 2026-09-07 | 0 failing / 30 total
• coq(CorePhi): replace proofs that were never checked by a compiler #3343 coq(CorePhi): replace proofs that were never checked by a compiler | Created: 2026-09-06 | Updated: 2026-09-06 | 5 failing / 32 total
• chore(ci)(deps): bump actions/attest-build-provenance from 1 to 4 #3001 chore(ci)(deps): bump actions/attest-build-provenance from 1 to 4 | Created: 2026-09-01 | Updated: 2026-09-14 | 1 failing / 31 total
• chore(ci)(deps): bump softprops/action-gh-release from 2 to 3 #3000 chore(ci)(deps): bump softprops/action-gh-release from 2 to 3 | Created: 2026-09-01 | Updated: 2026-09-14 | 1 failing / 29 total
• chore(ci)(deps): bump github/codeql-action from 3 to 4 #2999 chore(ci)(deps): bump github/codeql-action from 3 to 4 | Created: 2026-09-01 | Updated: 2026-09-14 | 1 failing / 31 total
• Land the formal verification harness on master (it has never run) #2727 Land the formal verification harness on master (it has never run) | Created: 2026-08-27 | Updated: 2026-09-05 | 5 failing / 24 total
• chore(deps)(deps): bump thiserror from 2.0.18 to 2.0.21 #2693 chore(deps)(deps): bump thiserror from 2.0.18 to 2.0.20 | Created: 2026-08-24 | Updated: 2026-08-28 | 3 failing / 34 total
• chore(deps)(deps): bump hyper from 1.10.1 to 1.11.1 #2692 chore(deps)(deps): bump hyper from 1.10.1 to 1.11.0 | Created: 2026-08-24 | Updated: 2026-08-28 | 3 failing / 33 total
• chore(deps)(deps): bump base64 from 0.22.1 to 0.23.1 #2691 chore(deps)(deps): bump base64 from 0.22.1 to 0.23.1 | Created: 2026-08-24 | Updated: 2026-08-28 | 3 failing / 34 total
• chore(deps)(deps): bump serde from 1.0.228 to 1.0.229 #2690 chore(deps)(deps): bump serde from 1.0.228 to 1.0.229 | Created: 2026-08-24 | Updated: 2026-08-28 | 3 failing / 34 total
• chore(deps)(deps): bump uuid from 1.23.2 to 1.26.1 #2689 chore(deps)(deps): bump uuid from 1.23.2 to 1.25.0 | Created: 2026-08-24 | Updated: 2026-08-28 | 3 failing / 34 total

Summary

Status	Count
Total Open PRs	35
PRs with Failing Checks	32
PRs with All Checks Green	3
READY	3
FAILING	32
PENDING	0
NO CHECKS YET	0

Seal Status

• ⚠️ STALE -- sha256(compiler.rs)=403499176a5d != manifest seal=87e5cbd3ad94.
  The committed NMSE numbers were certified against an older compiler.rs.
  Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).