Repository navigation
Description
Task
T-074a· Milestone M6 — Analysis + risk · Size S · Priority P1 · Epic #6Implements FR-ANA-005 · PR-14
Branch
feat/t-074a-interactiontester-plan-confirmatory-scenarios-and— one task = one branch = one PR (doc 32 §6)
Summary
Implement analysis/interaction.py per doc 09 §6: singles, pairs (if |C| ≥ 3) and full scenarios from x_min and characteristic set C; repeats; verdict 'interaction failure' if full fails ≥ success_ratio and no single does; order = smallest failing subset size.
Context / motivation
The core research claim (H3): RoboProbe finds failures that require combinations of conditions.
Scope
In scope
plan(mode, space) -> list[Scenario]andconclude(mode, confirm_runs) -> InteractionVerdict(status, order, params, evidence run ids).InteractionVerdict(status, order, params, per-subset ratios, evidence run ids).
Implementation guidance
- Module
roboprobe/analysis/interaction.py:class InteractionTester: def __init__(self, settings: InteractionSettings) -> None; def plan(self, mode: FailureModeSummary, space: ScenarioSpace) -> list[ConfirmScenario]; def conclude(self, mode: FailureModeSummary, confirm_runs: Sequence[RunRecord]) -> InteractionVerdict(doc 05 §7). - Planning (doc 09 §6): start from minimal scenario
x_min; C = characteristic set (≤ 4 params; if more, take the 4 smallest-coverage ones). Generate: singles (only one c ∈ C at itsx_minvalue, others at nominal), pairs if |C| ≥ 3, and full (all of C atx_min). Each ×analysis.interaction.repeats, with seeds derived viaderive_seed(campaign_seed, "confirm", mode_code, subset_key, rep). ConfirmScenariocarriessubset: frozenset[str]andtags=("confirm", mode_code).- Conclusion: failure ratio per subset;
interaction = full_ratio ≥ success_ratio and all(single_ratio < success_ratio);order = min |S| over subsets with ratio ≥ success_ratio; statusCONFIRMED | NOT_INTERACTION | INCONCLUSIVE(inconclusive if any INVALID runs reduce n below repeats). - Pure; no DB, no queue.
- Standards: respect the layering contracts (doc 16 §2) and conventions in doc 34; no hard-coded tunables — thresholds/defaults come from typed settings (doc 27); no magic numbers (ruff
PLR2004); public APIs typed (mypy strict) and documented.
Configuration keys
From the configuration reference (doc 27). Keys marked (new) must be added to the typed settings models and appear in the generated reference (T-003c).
analysis.interaction.repeatsanalysis.interaction.success_ratio
Acceptance criteria
- Synthetic 3-way interaction detected with order 3; single-factor mode (GT-D style) not flagged.
- Plan size = (|C| + C(|C|,2)·[|C|≥3] + 1) × repeats — property test.
- Seeds reproducible and distinct per subset/repeat.
- INCONCLUSIVE when INVALID runs present.
- Tests listed below added and green in CI; coverage of new code meets the gate in doc 18 §3.
- Docs, config reference and CHANGELOG
Unreleasedupdated where behaviour, keys or interfaces changed. - Definition of Done (doc 32 §3) satisfied.
Tests required
tests/unit/analysis/test_interaction.py.- Table-driven conclude() tests covering all three statuses.
Dependencies
Blocked by: #129 (T-071a)
Blocks: #134 (T-074b)
Parallelization notes
Files / paths owned by this task — avoid editing other paths; shared files (pyproject.toml entry points, cli/main.py, api/routers/__init__.py, router.tsx, docker/compose.yaml, ci.yml) are append-only:
backend/src/roboprobe/analysis/interaction.py
Independent.
Definition of Done
Per doc 32 §3 and AGENTS.md §5: lint, format, mypy strict and import-linter green; tests added and passing; docs updated; PR template completed with Closes #<this issue>; Conventional Commit messages referencing T-074a (doc 16 §6); commits authored only by the repository owner identity (AGENTS.md §4); CI green; reviewer checklist (doc 32 §5.1) satisfied.
References
Design docs: doc 09 §6 · doc 19 §1 · doc 21 §3
Requirements: FR-ANA-005 · PR-14 (traceability: doc 28)
External:
Metadata
Metadata
Assignees
Labels
Projects
- StatusShow more project fieldsBacklog
Activity
confirmruns and update mode #134