fix(ablation): harness import path, cell fault-isolation, harness stats#3
Merged
Merged
Conversation
Three bugs in the forge-hybrid ablation tooling, all surfaced by an end-to-end pre-flight smoke before the multi-day M-host run: 1. ablation_harness.py only put src/ on sys.path, so the top-level `benchmarks` package failed to import when run by path (sys.path[0] = scripts/, not repo root) -> ModuleNotFoundError: benchmarks. Add REPO_ROOT alongside src/. 2. ablation_run_cell.sh ran each benchmark under `set -e -o pipefail`. maintain_suite returns rc=1 whenever the v1_release_gate (>=8/10) is unmet (run.py:1628) -- an EXPECTED outcome for an ablation cell, which measures rather than gates -- so every cell aborted before SWE-bench. Fault-isolate all three benchmark steps (set +e + PIPESTATUS + log-and-continue), matching run_all.sh's stated "partial data over no data" philosophy. 3. ablation_harness.py stats print iterated run_harness's dict[str, HarnessResult] as a list of dicts (r.get(...)) -> AttributeError after scoring completed. Use .values() + dataclass attrs, mirroring write_harness_summary. Also: ABLATION_OUT_ROOT override (defaults to the canonical path) so smokes can run the real orchestration in isolation. Verified: BFCL 10/10 both cells, SWE-bench Docker scoring end-to-end (astropy__astropy-12907 resolved), clean harness exit, resume cache live. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Three bugs in the forge-hybrid agentic ablation tooling (shipped in
f609edb), all surfaced by an end-to-end pre-flight smoke run before committing the M-host to the ~80-100h matrix. Each would have silently crippled the SWE-bench dimension.ablation_harness.pyput onlysrc/onsys.path; run-by-path setssys.path[0]=scripts/, not repo rootModuleNotFoundError: benchmarks— harness never runs, no scoring in any cellREPO_ROOTtosys.pathset -e;maintain_suitereturnsrc=1whenever the v1_release_gate (≥8/10) is unmet (run.py:1628) — expected for an ablation cellset +e+PIPESTATUS+ log-and-continue)run_harness'sdict[str,HarnessResult]as a list of dictsAttributeErrorafter scoring; nonzero harness exit.values()+ dataclass attrs (mirrorswrite_harness_summary)Also adds an
ABLATION_OUT_ROOToverride (defaults to the canonical path) so a smoke can exercise the real orchestration in isolation.Verification
baselineandtrajectorysmoke cellsastropy__astropy-12907resolved,harness_summary.jsoncorrect, clean exitablation_run_all.sh's stated "partial data over no data" intentFixes are running in the live matrix now (working tree). This PR makes them durable + resumable on any host.
🤖 Generated with Claude Code