Repository navigation
feat(igla-coder): record the pilot's measured t27 result (Closes #1041) - #4796
Merged
Merged
Conversation
…ine spec Closes #1041 The igla-coder-gpu pilot produced trained 100m fp/tern checkpoints and a t27 benchmark judged by t27c. This records that measured result into specs/igla/coder/pipeline.t27 (the spec the WARS arena's IGLA CODER lane reads), moving the pipeline from "described" to "measured": - strict 391: fp 8, tern 5 solved within 10 tries - all 466: fp pass@10 5.8% (27), tern 5.4% (25); compile@1 12-14% is the wall - source: gHashTag/igla-coder-gpu research/T27_BENCH.md Honesty limit: the checkpoint is not yet published with a verifiable SHA, so the live WARS lane stays unregistered until it is; a software result judged by t27c, not hardware or a deployed network.
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
Contributor
This was referenced Sep 24, 2026
Merged
dmitrii-f-t27
added a commit
that referenced
this pull request
Oct 2, 2026
dmitrii-f-t27
added a commit
that referenced
this pull request
Oct 10, 2026
Corpus Ratchet failed its first step on master: three files gained a test that cannot fail. specs/igla/coder/pipeline.t27 added one whose body is 'assert true' (#4796); ternary_mac_synth.t27 (on_clock_reset) and gf16_matmul_top.t27 (led_r23_toggles) added ones whose bodies are only comments, the second ending "we will skip and hope". None of the three can be given an assertion that runs: t27c test-report reports all three specs BLOCKED in Zig, so no test in them executes. The three bodies are removed, as the gate asks; the pilot measurement in pipeline.t27 stays as written, with its invariant line. The ledger drops the one row that was already slack: specs/igla/race/cordic.t27 144 -> 0 (total 4049 -> 3905). The pipeline seal, stale on master since #4796, is resealed; --verify reports all MATCH. All four backends generate the three specs before and after, and the corpus suite summary is identical to master's. Refs #5497
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
1 participant
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.
Records the
igla-coder-gpupilot's measured result intospecs/igla/coder/pipeline.t27— the spec the WARS arena's IGLA CODER lane (lane 03) reads. This is the "one form": the corpus is.t27specs, 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):.t27fine-tune)gHashTag/igla-coder-gpuresearch/T27_BENCH.mdHonesty limits (BINDING)
t27c, not a hardware or deployed-network one.🤖 Generated with Claude Code