Hygiene: finish public terminology cleanup - #266
Merged
Aaron Aspinwall (AaronAspinwall123) merged 2 commits intoJul 24, 2026
Merged
Hygiene: finish public terminology cleanup#266Aaron Aspinwall (AaronAspinwall123) merged 2 commits into
Aaron Aspinwall (AaronAspinwall123) merged 2 commits into
Conversation
Replace remaining user-facing seed terminology with test set/test case wording and capitalize the Inference stage consistently, while preserving intentional overrefusal and internal legacy keys. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 53766e86-ef96-4fd8-b3d9-ed60ac272fcf
Aaron Aspinwall (AaronAspinwall123)
requested review from
Minsoo Thigpen (minthigpen) and
sooyeonni (sooyeonni)
as code owners
July 16, 2026 18:55
Use test set for collections, test case for individual rows, scenario for multi-turn scores, and Inference only when naming the pipeline stage, following PR #23. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 53766e86-ef96-4fd8-b3d9-ed60ac272fcf
Chang Liu (changliu2)
approved these changes
Jul 23, 2026
Chang Liu (changliu2)
left a comment
Chang Liu (changliu2)
left a comment
Collaborator
There was a problem hiding this comment.
Approving — clean, tightly-scoped public-terminology pass.
Verified:
- Canonical #23 vocab applied consistently:
Seed Generation→Test Set Generation,seed id→test case ID, and "policy + seeds" → "a behavior and its test set" on the new-run screen. overrefusalcorrectly preserved — the diff never touches it, and flagging #152 as partially stale is right: it's an intentional built-in judge dimension now, not a rename target.- Legacy internal keys untouched — only the
seeds:label changes; the object key stays, so no route/stage behavior changes. Inferencecapitalized only as the pipeline-stage noun, consistent with #135. Thecsv/[suite_id]404 "No scenario scores found" reads correctly for the audit/scenario export path.- CI green (viewer-build + CodeQL); label-only diff, no logic changes.
Chang Liu (changliu2)
added a commit
that referenced
this pull request
Jul 24, 2026
Merging main into this branch surfaced a pre-existing mismatch: #266 capitalized the viewer label to 'Refused before Inference' (pipeline-stage noun) but test_viewer_server_artifacts.py still asserted the lowercase form in three places (previewTesterLabel / previewDrawerLabel / scoredTesterLabel). #266 was viewer-only so Tier 1 unit tests never ran on it. Align the assertions with the intentional capitalized label. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: d429caee-f2da-4fb0-8b2c-d0eff7cf40a3
Aaron Aspinwall (AaronAspinwall123)
added a commit
to ricfog/ASSERT
that referenced
this pull request
Jul 28, 2026
The viewer stop-reason label is 'Refused before Inference' (capital I, result-view.ts:20), but three assertions in test_load_run_page_data_exposes_refusal_stop_reason_display still expected lowercase 'inference'. This mismatch is inherited from main (present at base 7a6e95b, from responsibleai#266's terminology cleanup) and fails Tier 1 unit tests. Align the test to the code label; mirrors the fix in a403385. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: ef089fbc-c563-427f-b8e1-f87370e98c76
Aaron Aspinwall (AaronAspinwall123)
added a commit
that referenced
this pull request
Jul 29, 2026
* Fix permissibility-split policy violation metrics * test(viewer): align inference stop-reason label expectation The viewer stop-reason label is 'Refused before Inference' (capital I, result-view.ts:20), but three assertions in test_load_run_page_data_exposes_refusal_stop_reason_display still expected lowercase 'inference'. This mismatch is inherited from main (present at base 7a6e95b, from #266's terminology cleanup) and fails Tier 1 unit tests. Align the test to the code label; mirrors the fix in a403385. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: ef089fbc-c563-427f-b8e1-f87370e98c76 * fix(export): derive not_permissible_policy_violation_rate from node split export_suite_results.py still computed not_permissible_policy_violation_rate by row-filtering on the originating permissibility label and reading the row-level policy_violation verdict - the overrefusal-contaminated approach issue #272 calls out. Reuse compute_policy_violation_by_permissibility so the export matches the CLI/results and viewer definition: one vote per conversation over non-permissible nodes only. Keeps permissible_overrefusal_rate on the ok-filtered permissible rows. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: ef089fbc-c563-427f-b8e1-f87370e98c76 --------- Co-authored-by: Aaron Aspinwall <94419331+AaronAspinwall123@users.noreply.github.com> Co-authored-by: AaronAspinwall123 <aaron.aspinwall@gmail.com> Copilot-Session: ef089fbc-c563-427f-b8e1-f87370e98c76
Chang Liu (changliu2)
added a commit
that referenced
this pull request
Aug 5, 2026
…r-YAML (ACS demo superseded by #262) (#271) * fix(examples): incident_triage_agent - baseline-only, one-behavior-per-YAML (remove preview ACS/GEPA surface) This example previously shipped a full ACS + AgentShield + GEPA demo into the public repo. That preview integration is not meant to be public, and public examples must not depend on unreleased/preview integrations. This strips the example back to a self-contained baseline ASSERT eval that runs with just assert-ai + an LLM key. Removed (preview surface): agent_guarded.py, eval_config_guarded.yaml, eval_config_guarded_gepa.yaml, incident-triage.guardrails.yaml, optimize_with_gepa.ipynb, prompts/ (optimized prompt + cross/pii_detection + xpia_paraphrase), and the whole artifacts/ tree (guarded/gepa/naive result snapshots + trade_off.png). Also removed eval_config_naive_prompt.yaml (folded into a chat_naive contrast callable). Scrubbed agent.py and SOP.md of every guarded/GEPA reference. Restructured to one-behavior-per-YAML under behaviors/: sop_tool_ordering (ordering_violation), pager_severity_discipline (pager_violation), channel_severity_routing (channel_violation), pii_redaction (pii_leak; folds pii_leak_under_pressure), xpia_resistance (xpia_relay), escalation_correctness (escalation_violation), alert_id_integrity (alert_id_drift), severity_classification (wrong_severity; folds severity_drift_under_pressure), no_fabrication (fabrication). Each disables the built-in policy_violation/overrefusal headline and scores one custom rubric dimension copied verbatim from the old bundled config. eval_config_baseline.yaml stays as a scrubbed bundled overview (baseline target, built-in headline + one dim per mode). Validation: all 10 configs load via assert_ai.runner._load_context (0 errors); agent.py passes py_compile; preview-surface grep across the example is clean. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: d429caee-f2da-4fb0-8b2c-d0eff7cf40a3 * test(incident-triage): update smoke test for baseline-only structure The smoke test still referenced the removed guarded/GEPA surface (agent_guarded.py, incident-triage.guardrails.yaml, eval_config_guarded.yaml), so Tier 1 unit tests failed after those files were deleted. Drop the GuardrailsYamlShapeTest and guarded-agent import test, and rewrite EvalConfigShapeTest to assert the baseline config + all nine one-behavior-per-YAML configs target the baseline callable, reference no guarded target, and each carry a distinct suite. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: d429caee-f2da-4fb0-8b2c-d0eff7cf40a3 * test(viewer): fix stale 'Refused before inference' assertions Merging main into this branch surfaced a pre-existing mismatch: #266 capitalized the viewer label to 'Refused before Inference' (pipeline-stage noun) but test_viewer_server_artifacts.py still asserted the lowercase form in three places (previewTesterLabel / previewDrawerLabel / scoredTesterLabel). #266 was viewer-only so Tier 1 unit tests never ran on it. Align the assertions with the intentional capitalized label. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: d429caee-f2da-4fb0-8b2c-d0eff7cf40a3 --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Co-authored-by: Aaron Aspinwall <94419331+AaronAspinwall123@users.noreply.github.com> Copilot-Session: d429caee-f2da-4fb0-8b2c-d0eff7cf40a3
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
2 participants
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
Test Set Generationand test case terminology defined in refactor(p2m): terminology migration to maximize broad-developer recognition #23Inferenceonly when referring to the pipeline stage, consistent with fix(viewer): export evaluation set links to canonical test_set.jsonl and Inference capitalise #135overrefusalterminology and legacy internal route/stage keysIssue #152 is partially stale:
overrefusalis now an intentional built-in judge dimension. This PR applies only the remaining valid public terminology cleanup and follows the canonical vocabulary established in #23.Closes #152