feat: execute WP7-02 independent model matrices

2026-08-04 16:01:31 +08:00
parent c2f89453a2
commit 470d243b5b
11 changed files with 811 additions and 142 deletions
@@ -0,0 +1,64 @@
+import { existsSync, readFileSync, writeFileSync } from "node:fs";
+import { resolve } from "node:path";
+
+import { WP7_02_MODEL_IDS } from "./lib/wp7-02-external-contract.mjs";
+import { validateSanitizedEvidence } from "./lib/wp7-02-controlled-executor.mjs";
+
+const caseDirectory = process.env.DADA_WP7_02_CASE_DIR;
+const runId = process.env.DADA_TDD_RUN_ID;
+const confirmed = process.argv.includes("--confirm-manual-review");
+
+function readJson(path) {
+  return validateSanitizedEvidence(JSON.parse(readFileSync(path, "utf8")));
+}
+
+function output(value, error = false) {
+  const serialized = JSON.stringify(validateSanitizedEvidence(value));
+  if (error) console.error(serialized); else console.log(serialized);
+}
+
+try {
+  if (!confirmed || !caseDirectory || !runId) throw new Error("WP7_02_MANUAL_REVIEW_CONFIRMATION_REQUIRED");
+  const reviewed = [];
+  const evidenceIds = new Set();
+  for (const modelId of WP7_02_MODEL_IDS) {
+    const directory = resolve(caseDirectory, modelId.replaceAll(".", "_"));
+    const paths = ["contract-matrix.json", "external-calls.json", "readiness.json", "redaction.json"]
+      .map((name) => resolve(directory, name));
+    if (paths.some((path) => !existsSync(path))) throw new Error("WP7_02_MANUAL_REVIEW_EVIDENCE_MISSING");
+    const matrix = readJson(paths[0]);
+    const externalCalls = readJson(paths[1]);
+    const readiness = readJson(paths[2]);
+    const redaction = readJson(paths[3]);
+    const complete = matrix.model_id === modelId && matrix.config_version === 1 && matrix.status === "passed"
+      && matrix.pure_text?.status === "passed" && matrix.reference_image?.status === "passed"
+      && matrix.ratios?.length === 4 && matrix.ratios.every((entry) => entry.status === "passed")
+      && matrix.execution_modes?.length === 3 && matrix.execution_modes.every((entry) => ["passed", "covered_by_real_calls"].includes(entry.status))
+      && matrix.error_scenarios?.length === 9 && matrix.error_scenarios.every((entry) => entry.status === "passed")
+      && matrix.settlements?.length === 3 && matrix.contract_change?.full_matrix_reapplied === true
+      && externalCalls.status === "passed" && externalCalls.real_calls === 5 && externalCalls.planned_real_calls === 5
+      && externalCalls.calls?.length === 5 && new Set(externalCalls.calls.map((entry) => entry.scenario_id)).size === 5
+      && externalCalls.calls.every((entry) => entry.status === "passed" && entry.source === "real_gateway")
+      && externalCalls.approved_real_call_limit === 120
+      && readiness.status === "passed" && redaction.status === "passed" && redaction.secret_scan === "passed";
+    if (!complete || evidenceIds.has(readiness.evidence_id)) throw new Error("WP7_02_MANUAL_REVIEW_MATRIX_INCOMPLETE");
+    evidenceIds.add(readiness.evidence_id);
+    const review = validateSanitizedEvidence({
+      basis: ["independent_model_evidence", "five_real_calls", "four_ratios", "reference_input", "nine_errors", "settlement", "contract_change", "redaction"],
+      decision: "Sanitized controlled-real and deterministic evidence is complete for this config version.",
+      model_id: modelId,
+      reviewed_at: new Date().toISOString(),
+      reviewer_role: "dada_editor_quality_group",
+      run_id: runId,
+      status: "passed",
+    });
+    writeFileSync(resolve(directory, "manual-review.json"), `${JSON.stringify(review, null, 2)}\n`);
+    reviewed.push({ model_id: modelId, status: "passed" });
+  }
+  output({ reviewed, run_id: runId, status: "passed" });
+} catch (error) {
+  const message = error instanceof Error ? error.message : String(error);
+  const code = /^WP7_02_[A-Z0-9_:]+$/.test(message) ? message : "WP7_02_MANUAL_REVIEW_FAILED";
+  output({ code, run_id: runId, status: "externally_blocked" }, true);
+  process.exitCode = 3;
+}