Tricoder/backend/app/services/stub_runner_service.py

from dataclasses import dataclass

from app.schemas.conclusion import ConclusionSummary, RankedFinding
from app.schemas.evidence import EvidenceSummary
from app.schemas.experiment import ExperimentSummary, ScoreCard
from app.schemas.observation import ObservationSummary
from app.schemas.orchestration import OrchestratorPlan


@dataclass(slots=True)
class StubRunnerResult:
    probe_evidence: list[EvidenceSummary]
    experiments: list[ExperimentSummary]
    conclusion: ConclusionSummary
    status: str


class StubRunnerService:
    def run_initial_plan(
        self,
        *,
        session_id: str,
        observation: ObservationSummary,
        plan: OrchestratorPlan,
    ) -> StubRunnerResult:
        periodic_hint = 0.79 if "periodic" in observation.tags else 0.68
        tonal_hint = 0.74 if observation.duration_ms >= 4500 else 0.61
        classifier_top_score = 0.64 if "periodic" in observation.tags else 0.55

        probe_evidence = [
            EvidenceSummary(
                id=f"{session_id}-e1",
                producer_module_id="energy_probe",
                category="feature",
                values={
                    "rms": 0.41,
                    "peak_count": 7,
                    "duration_ms": observation.duration_ms,
                },
                confidence=0.86,
            ),
            EvidenceSummary(
                id=f"{session_id}-e2",
                producer_module_id="periodicity_probe",
                category="pattern",
                values={
                    "repeat_interval_ms": 620,
                    "stability": periodic_hint,
                },
                confidence=0.82,
            ),
            EvidenceSummary(
                id=f"{session_id}-e3",
                producer_module_id="baseline_sound_classifier",
                category="classification",
                values={
                    "top_k": [
                        {
                            "label": "periodic_alert_tone",
                            "score": classifier_top_score,
                        },
                        {
                            "label": "mechanical_click_sequence",
                            "score": 0.27,
                        },
                    ]
                },
                confidence=classifier_top_score,
            ),
        ]

        experiments: list[ExperimentSummary] = []
        parent_id: str | None = None

        for index, planned_experiment in enumerate(plan.experiments, start=1):
            experiment_id = f"{session_id}-x{index}"
            title = planned_experiment.goal
            score_total = 0.72 + (index * 0.06)
            notes = [
                f"Executed planned chain '{planned_experiment.preferred_chain_id or 'ad-hoc'}'."
            ]

            if planned_experiment.preferred_chain_id == "periodic_validation_chain":
                components = {
                    "repeat_consistency": 0.89,
                    "pattern_clarity": periodic_hint,
                    "resource_cost": 0.18,
                }
                penalties = {"mixed_source_risk": 0.05}
                notes.append("Autocorrelation and cepstrum align on repeat interval.")
                failure_reasons: list[str] = []
            elif planned_experiment.preferred_chain_id == "harmonic_validation_chain":
                components = {
                    "tonal_stability": tonal_hint,
                    "harmonic_ratio": 0.71,
                    "resource_cost": 0.22,
                }
                penalties = {"short_clip_penalty": 0.04}
                notes.append("Harmonic structure is present but not fully dominant.")
                failure_reasons = []
            else:
                components = {
                    "mixed_source_probability": 0.43,
                    "post_separation_gain": 0.18,
                    "resource_cost": 0.31,
                }
                penalties = {"separation_uncertainty": 0.08}
                notes.append(
                    "Source separation did not improve confidence enough to overtake the periodic hypothesis."
                )
                failure_reasons = []

            experiments.append(
                ExperimentSummary(
                    id=experiment_id,
                    parent_id=parent_id,
                    hypothesis_id=planned_experiment.hypothesis_id,
                    title=title,
                    status="done",
                    pipeline_summary=[
                        node.module_id for node in planned_experiment.pipeline
                    ],
                    score=ScoreCard(
                        total=round(score_total, 2),
                        components=components,
                        penalties=penalties,
                        notes=notes,
                    ),
                    failure_reasons=failure_reasons,
                )
            )
            parent_id = experiment_id

        conclusion = ConclusionSummary(
            summary=(
                "The sample most likely contains a periodic alert-like tone rather "
                "than an unstructured environmental event."
            ),
            findings=[
                RankedFinding(
                    label="Periodic alert tone",
                    confidence=0.84 if "periodic" in observation.tags else 0.76,
                    supporting_evidence=[
                        "Repeat interval remains stable near 620 ms.",
                        "Probe chain shows clear onset grouping.",
                        "Validation chains favor periodic structure over diffuse events.",
                    ],
                    contradicting_evidence=[
                        "Mixed-source contamination is not fully ruled out.",
                    ],
                )
            ],
            uncertainty=[
                "The current stub runner does not yet incorporate source separation output.",
                "A longer clip may improve confidence around tonal stability.",
            ],
            suggested_next_actions=[
                "Run harmonic validation against a longer sample window.",
                "Add a mixed-source validation branch if future captures include overlap tags.",
            ],
        )

        return StubRunnerResult(
            probe_evidence=probe_evidence,
            experiments=experiments,
            conclusion=conclusion,
            status="reviewing",
        )