fix(wp7-02): preserve independent mixed model evidence
Dada P0-A isolated Windows CI / validate-and-package (push) Failing after 2m11s

This commit is contained in:
suyx
2026-08-04 17:35:38 +08:00
parent 382d50058c
commit 05949230ca
5 changed files with 76 additions and 35 deletions
+2
View File
@@ -222,7 +222,9 @@ export function assembleControlledModelEvidence({ deterministicState, modelConfi
evidence_id: evidenceId,
external_calls: {
approved_real_call_limit: WP7_02_CONTROLLED_REAL_LIMIT,
attempts: realExecution.attempts,
calls: realExecution.calls,
maximum_real_calls: realExecution.maximum_real_calls,
mode: "controlled_real",
planned_real_calls: realExecution.planned_real_calls,
real_calls: realExecution.real_calls,
+4 -1
View File
@@ -206,7 +206,10 @@ export function validateIndependentEvidenceSet(evidence) {
const passed = entry.status === "passed" && entry.matrix?.status === "passed"
&& entry.external_calls?.status === "passed" && entry.external_calls.real_calls > 0
&& entry.manual_review?.status === "passed" && entry.redaction?.status === "passed";
if (entry.matrix?.model_id !== entry.model_id || (!blocked && !passed)
const pendingReview = entry.status === "passed" && entry.matrix?.status === "passed"
&& entry.external_calls?.status === "passed" && entry.external_calls.real_calls > 0
&& entry.manual_review?.status === "pending" && entry.redaction?.status === "passed";
if (entry.matrix?.model_id !== entry.model_id || (!blocked && !passed && !pendingReview)
|| /\"verified\"\s*:/i.test(JSON.stringify(entry))) {
throw new Error("WP7_02_BLOCKED_EVIDENCE_INVALID");
}
+54
View File
@@ -0,0 +1,54 @@
const productDimensions = Object.freeze({
"3:4": [1080, 1440],
"1:1": [1080, 1080],
"4:3": [1440, 1080],
"9:16": [1080, 1920],
});
function modelEvidenceComplete(entry) {
const { externalCalls, matrix, modelId, readiness, redaction } = entry;
return matrix.model_id === modelId && Number.isSafeInteger(matrix.config_version) && matrix.config_version > 0 && matrix.status === "passed"
&& matrix.pure_text?.status === "passed" && matrix.reference_image?.status === "passed"
&& matrix.ratios?.length === 4 && matrix.ratios.every((row) => row.status === "passed")
&& matrix.execution_modes?.length === 3 && matrix.execution_modes.every((row) => ["passed", "covered_by_real_calls"].includes(row.status))
&& matrix.error_scenarios?.length === 9 && matrix.error_scenarios.every((row) => row.status === "passed")
&& matrix.settlements?.length === 3 && matrix.contract_change?.full_matrix_reapplied === true
&& externalCalls.status === "passed" && externalCalls.real_calls >= 5 && externalCalls.real_calls <= 6
&& externalCalls.planned_real_calls === 5 && externalCalls.maximum_real_calls === 6
&& externalCalls.attempts?.length === externalCalls.real_calls
&& externalCalls.calls?.length === 5 && new Set(externalCalls.calls.map((row) => row.scenario_id)).size === 5
&& externalCalls.calls.every((row) => row.status === "passed" && row.source === "real_gateway")
&& externalCalls.calls.every((row) => {
const [width, height] = productDimensions[row.requested_ratio] ?? [];
return row.response?.dimensions?.width === width && row.response?.dimensions?.height === height;
})
&& externalCalls.approved_real_call_limit === 120
&& readiness.status === "passed" && redaction.status === "passed" && redaction.secret_scan === "passed";
}
export function reviewIndependentModelEvidence(entries, { reviewedAt, runId }) {
if (!Array.isArray(entries) || entries.length !== 3 || !runId || !reviewedAt) {
throw new Error("WP7_02_MANUAL_REVIEW_EVIDENCE_INVALID");
}
const evidenceIds = entries.map((entry) => entry.readiness?.evidence_id);
if (evidenceIds.some((id) => typeof id !== "string") || new Set(evidenceIds).size !== entries.length) {
throw new Error("WP7_02_MANUAL_REVIEW_EVIDENCE_NOT_INDEPENDENT");
}
const reviews = entries.map((entry) => modelEvidenceComplete(entry) ? {
basis: ["independent_model_evidence", "five_scenarios_bounded_attempts", "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: entry.modelId,
reviewed_at: reviewedAt,
reviewer_role: "dada_editor_quality_group",
run_id: runId,
status: "passed",
} : {
decision: "Independent model evidence remains incomplete or externally blocked.",
model_id: entry.modelId,
reviewed_at: reviewedAt,
reviewer_role: "dada_editor_quality_group",
run_id: runId,
status: "blocked",
});
return { reviews, status: reviews.every((review) => review.status === "passed") ? "passed" : "externally_blocked" };
}
+10 -33
View File
@@ -3,11 +3,11 @@ 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";
import { reviewIndependentModelEvidence } from "./lib/wp7-02-manual-review.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");
const productDimensions = { "3:4": [1080, 1440], "1:1": [1080, 1080], "4:3": [1440, 1080], "9:16": [1080, 1920] };
function readJson(path) {
return validateSanitizedEvidence(JSON.parse(readFileSync(path, "utf8")));
@@ -20,8 +20,7 @@ function output(value, error = false) {
try {
if (!confirmed || !caseDirectory || !runId) throw new Error("WP7_02_MANUAL_REVIEW_CONFIRMATION_REQUIRED");
const reviewed = [];
const evidenceIds = new Set();
const entries = [];
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"]
@@ -31,37 +30,15 @@ try {
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.real_calls <= 6 && externalCalls.planned_real_calls === 5
&& externalCalls.maximum_real_calls === 6 && externalCalls.attempts?.length === externalCalls.real_calls
&& 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.calls.every((entry) => {
const [width, height] = productDimensions[entry.requested_ratio] ?? [];
return entry.response?.dimensions?.width === width && entry.response?.dimensions?.height === height;
})
&& 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"), `${JSON.stringify(review, null, 2)}\n`);
reviewed.push({ model_id: modelId, status: "passed" });
entries.push({ directory, externalCalls, matrix, modelId, readiness, redaction });
}
output({ reviewed, run_id: runId, status: "passed" });
const result = reviewIndependentModelEvidence(entries, { reviewedAt: new Date().toISOString(), runId });
for (let index = 0; index < entries.length; index += 1) {
const review = validateSanitizedEvidence(result.reviews[index]);
writeFileSync(resolve(entries[index].directory, "manual-review.json"), `${JSON.stringify(review, null, 2)}\n`);
}
output({ reviewed: result.reviews.map(({ model_id, status }) => ({ model_id, status })), run_id: runId, status: result.status }, result.status !== "passed");
if (result.status !== "passed") process.exitCode = 3;
} catch (error) {
const message = error instanceof Error ? error.message : String(error);
const code = /^WP7_02_[A-Z0-9_:]+$/.test(message) ? message : "WP7_02_MANUAL_REVIEW_FAILED";
+6 -1
View File
@@ -137,13 +137,18 @@ if (!externalExitCodesValid) {
process.exit(1);
}
if (controlledReal && externalCommands.every((command) => command.exit_code === 0)) {
if (controlledReal) {
const manualConfirmed = process.env.DADA_WP7_02_MANUAL_REVIEW_CONFIRMATION === "confirmed";
const manual = run("manual-review", process.execPath, ["scripts/record-wp7-02-manual-review.mjs", ...(manualConfirmed ? ["--confirm-manual-review"] : [])], {
env: { DADA_TDD_RUN_ID: runId, DADA_WP7_02_CASE_DIR: caseDirectory },
logicalCommand: `pnpm review:wp7-02${manualConfirmed ? " -- --confirm-manual-review" : ""}`,
});
commands.push(manual);
if (![0, 3].includes(manual.exit_code)) {
writeFileSync(resolve(caseDirectory, "commands.json"), `${JSON.stringify({ commands, run_id: runId }, null, 2)}\n`);
console.error(JSON.stringify({ code: "WP7_02_MANUAL_REVIEW_COMMAND_FAILED", command: manual.command, exit_code: manual.exit_code, real_calls: 0, run_id: runId, status: "failed" }));
process.exit(1);
}
}
const modelEvidence = WP7_02_MODEL_IDS.map((modelId) => {