Repository navigation
Conversation
- scripts/run_phase5.py: 30-job matrix (6 conditions x 5 files), freeze lock/manifest, resume, --jobs, retries, --max-calls, --check integrity - scripts/report_phase5.py: headline, breakdowns (attacker/user tool, split, risk), paired stats, tau x delta Pareto, latency, utility proxies, cost - tests/test_report_phase5.py; roadmap Stage 0 checked; CHANGELOG + AGENTS updated (158 tests)
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What this PR does
Phase 5 Stage 0 (pre-run engineering, everything needed before the frozen full evaluation) plus the reporter/orchestrator finished on top of the merged ablation commit.
Runner/analysis changes
gate_events— benchmark FPR proxy + full latency samples.--ablation semantic-only|rule-only|raw-request) wired throughGateMiddleware(A4 derived from traces); ablation tag in trace metadata.eval/stats.py: exact two-sided binomial McNemar (<25 discordant pairs) + Yates chi-square fallback, seeded bootstrap CI; tests.eval/sweep.py:sweep_grid(τ×δ) with escalate share;scripts/sweep_gated.py --delta-gridtables.LLMClient.usage+eval/cost.pyestimates; every run report includes usage +estimated_cost_usd.Orchestration (new)
scripts/run_phase5.py: 30-job matrix (6 conditions × {4 InjecAgent files, MCPTox}),--freezelock/manifest (git SHA + config hashes + model IDs),--runwith resume/--jobs/retries/--max-calls,--checkintegrity mode,--dry-run(reports 34,560 estimated calls).scripts/report_phase5.py: headline tables, paired stats vs B1, breakdowns (attacker tool, user tool, split, MCPTox risk), τ×δ Pareto, latency p50/p95, utility proxies, cost.Verification
pytest -q→ 158 passed.Notes