feat: execute WP7-02 independent model matrices

tests/package/wp7-02-controlled-executor.test.mjs

@@ -10,6 +10,12 @@ import {
   normalizeProviderResponse,
   validateSanitizedEvidence,
 } from "../../scripts/lib/wp7-02-controlled-executor.mjs";
+import {
+  assembleControlledModelEvidence,
+  buildDeterministicExecutionEvidence,
+  createControlledReferencePng,
+  runControlledRealScenarios,
+} from "../../scripts/lib/wp7-02-controlled-matrix.mjs";
 
 const models = [
   {
@@ -82,7 +88,7 @@ test("TDD-WP7-EXT-001 builds protocol-specific requests without auth in argument
     model: "gpt-image-2",
     prompt: "controlled fixture prompt",
     response_format: "b64_json",
-    size: "1024x1536",
+    size: "1080x1920",
   });
   assert.equal("authorization" in openai.headers, false);
 });
@@ -138,3 +144,65 @@ test("TDD-WP7-EXT-001 confines the credential to the request header and discards
     return true;
   });
 });
+
+test("TDD-WP7-EXT-001 executes only five real success probes per model and keeps failed ratios blocking", async () => {
+  let fetchCalls = 0;
+  const execution = await runControlledRealScenarios({
+    fetchImpl: async () => {
+      fetchCalls += 1;
+      return new Response(JSON.stringify({
+        candidates: [{ content: { parts: [{ inlineData: { data: onePixelPng.toString("base64"), mimeType: "image/png" } }] } }],
+      }), { status: 200 });
+    },
+    maxRealCalls: 120,
+    modelConfig: models[0],
+    token: "controlled-secret-value-for-test-only",
+  });
+  assert.equal(fetchCalls, 5);
+  assert.equal(execution.real_calls, 5);
+  assert.equal(execution.status, "externally_blocked");
+  assert.equal(execution.calls.filter((call) => call.status === "passed").length, 2);
+  assert.doesNotMatch(JSON.stringify(execution), /controlled-secret|fixture prompt|iVBOR/i);
+  await assert.rejects(() => runControlledRealScenarios({ maxRealCalls: 121, modelConfig: models[0], token: "not-used" }), /WP7_02_REAL_CALL_LIMIT_INVALID/);
+});
+
+test("TDD-WP7-EXT-001 assembles independent complete evidence without retaining reference bytes", () => {
+  const reference = createControlledReferencePng();
+  assert.equal(reference.subarray(0, 8).toString("hex"), "89504e470d0a1a0a");
+  const calls = ["3:4", "1:1", "4:3", "9:16"].map((ratio, index) => ({
+    duration_ms: 1,
+    http_status: 200,
+    input: "pure_text",
+    requested_ratio: ratio,
+    response: { dimensions: { height: 1, width: 1 }, evidence_hash: `sha256:${"A".repeat(64)}`, mime: "image/png", usage_summary: { input_units: 0, output_units: 0, total_units: 0 } },
+    scenario_id: `real-${index + 1}`,
+    source: "real_gateway",
+    status: "passed",
+  }));
+  calls.push({ ...calls[1], input: "reference_image", scenario_id: "real-5" });
+  const deterministicState = {
+    contract_change: { full_matrix_reapplied: true },
+    error_scenarios: Array.from({ length: 9 }, (_, index) => ({ category: `category-${index}`, status: "passed" })),
+    model_id: models[0].model_id,
+    settlements: ["succeeded", "failed", "rejected"].map((outcome) => ({ outcome })),
+    status: "passed",
+  };
+  const evidence = assembleControlledModelEvidence({
+    deterministicState,
+    modelConfig: models[0],
+    realExecution: { calls, planned_real_calls: 5, real_calls: 5, status: "passed" },
+    runId: "wp7-02-assembly-test",
+  });
+  assert.equal(evidence.status, "passed");
+  assert.equal(evidence.manual_review.status, "pending");
+  assert.equal(evidence.matrix.error_scenarios.length, 9);
+  assert.doesNotMatch(JSON.stringify(evidence), /iVBOR|image_bytes|raw_prompt|authorization/i);
+  const execution = buildDeterministicExecutionEvidence(models[0].model_id, "wp7-02-assembly-test");
+  assert.deepEqual(execution.modes.map((entry) => entry.mode), ["sync", "async", "poll"]);
+  assert.deepEqual(execution.trace.map((entry) => `${entry.action}:${entry.before}->${entry.after}`), [
+    "start:created->pending",
+    "poll:pending->completed",
+    "poll:completed->completed",
+  ]);
+  assert.equal(execution.trace[2].replay, true);
+});