NODEDC_MISSION_CORE/apps/control-station/test/e46fDashCamBakeoff.test.mjs

import assert from "node:assert/strict";
import { readFile } from "node:fs/promises";
import test, { after, before } from "node:test";
import { createServer } from "vite";

let server;
let fetchE46FDashCamBakeoff;

before(async () => {
  server = await createServer({
    server: { middlewareMode: true },
    appType: "custom",
    logLevel: "silent",
  });
  ({ fetchE46FDashCamBakeoff } = await server.ssrLoadModule(
    "/src/core/laboratory/e46fDashCamBakeoff.ts",
  ));
});

after(async () => {
  await server?.close();
});

function metrics(overrides = {}) {
  return {
    frame_count: 4489,
    route_duration_seconds: 458.713353,
    detection_observation_count: 23478,
    track_observation_count: 24937,
    detection_box_clipped_count: 0,
    track_box_clipped_count: 1630,
    unique_track_count: 545,
    mean_tracked_objects_per_frame: 5.555135,
    zero_detection_frame_count: 39,
    zero_track_frame_count: 25,
    tracker_recovered_frame_count: 29,
    full_layer_blackout_event_count: 1,
    route_id_gap_event_count: 0,
    short_track_count: 37,
    short_track_fraction: 0.06789,
    track_class_switch_count: 0,
    ...overrides,
  };
}

test("E46F preserves temporal gains but admits the semantic rejection", async () => {
  const resultId = `e46f-dashcam-bakeoff-${"a".repeat(64)}`;
  const baselineId = `e46e-ready-stack-${"b".repeat(64)}`;
  const identity = "c".repeat(64);
  const fetcher = async () => new Response(JSON.stringify({
    schema_version: "missioncore.e46f-dashcam-bakeoff-catalog/v1",
    items: [{
      schema_version: "missioncore.e46f-dashcam-bakeoff-view/v1",
      result_id: resultId,
      created_at_utc: "2026-08-04T09:00:00Z",
      source_session_id: "20260720T065719Z_viewer_live",
      camera_source_id: "sensor.camera.right",
      metrics: metrics(),
      acceptance: {
        full_route_accounted: true,
        stock_detector_tracker_executed: true,
        controlled_detector_only_change: true,
        visual_overlay_available: true,
        independent_truth_available: false,
        navigation_or_safety_accepted: false,
      },
      method: {
        schema_version: "missioncore.laboratory-method/v1",
        completeness: "complete",
        execution_class: "hybrid",
        pipeline_id: "e46f-deepstream-dashcamnet-detectnet-v2-nvdcf/v1",
        components: [{
          kind: "model",
          name: "NVIDIA DashCamNet",
          version: "pruned_onnx_v1.0.4",
          role: "moving-camera traffic-object detection",
          identity_sha256: identity,
        }],
      },
      limitations: ["not independent truth"],
      comparison: {
        controlled_change: "detector-only",
        baseline_result_id: baselineId,
        baseline_metrics: metrics({
          unique_track_count: 909,
          zero_track_frame_count: 28,
          full_layer_blackout_event_count: 2,
          short_track_fraction: 0.093509,
        }),
        delta: {
          zero_track_frame_count: -3,
          full_layer_blackout_event_count: -1,
          unique_track_count: -364,
          short_track_fraction: -0.025619,
        },
        large_box_visual_triage: {
          area_ratio_threshold: 0.2,
          candidate: {
            observation_count: 2912,
            frame_count: 2173,
            track_id_count: 23,
            class_observations: { person: 2912 },
          },
          baseline: {
            observation_count: 191,
            frame_count: 191,
            track_id_count: 4,
            class_observations: { car: 191 },
          },
          interpretation: "diagnostic visual triage; not precision/recall",
        },
        visual_review: {
          status: "rejected-semantic-regression",
          sample_video_seconds: [4.2, 9, 22, 44, 264, 418],
          finding: "fisheye rim becomes huge person tracks",
          next_action: "rectify valid FOV and repeat A/B",
        },
        verdict: "reject-dashcamnet-on-unrectified-fisheye",
      },
      video: {
        url: `/api/v1/laboratory/e46f/results/${resultId}/overlay.mp4`,
        media_type: "video/mp4",
        byte_length: 150513080,
        sha256: identity,
        width: 800,
        height: 600,
      },
      ground_truth: false,
    }],
  }), { status: 200, headers: { "Content-Type": "application/json" } });

  const result = await fetchE46FDashCamBakeoff({ fetcher });
  assert.equal(result.resultId, resultId);
  assert.equal(result.comparison.verdict, "reject-dashcamnet-on-unrectified-fisheye");
  assert.equal(result.comparison.delta.zeroTrackFrameCount, -3);
  assert.equal(result.comparison.largeBoxVisualTriage.candidate.observationCount, 2912);
  assert.equal(result.comparison.visualReview.status, "rejected-semantic-regression");
  assert.equal(result.acceptance.navigationOrSafetyAccepted, false);
  assert.equal(result.video.sha256, identity);
  assert.doesNotMatch(JSON.stringify(result), /Users|D:\\|runtime\/experiments/);
});

test("E46F uses the fixed LAB anatomy and exposes the frozen full video", async () => {
  const [resultView, visual] = await Promise.all([
    readFile(new URL("../src/workspaces/laboratory/E46FDashCamBakeoffResult.tsx", import.meta.url), "utf8"),
    readFile(new URL("../src/workspaces/laboratory/E46EReadyStackVisual.tsx", import.meta.url), "utf8"),
  ]);
  assert.match(resultView, /LaboratorySummary/);
  assert.match(resultView, /LaboratoryEvidence/);
  assert.match(resultView, /LaboratoryResultSummary/);
  assert.match(resultView, /semantic regression/);
  assert.match(visual, /LaboratoryEvidenceViewer/);
  assert.match(visual, /<video/);
  assert.match(visual, /controls/);
});