feat: execute WP7-02 independent model matrices
Dada P0-A isolated Windows CI / validate-and-package (push) Failing after 54s
Dada P0-A isolated Windows CI / validate-and-package (push) Failing after 54s
This commit is contained in:
@@ -0,0 +1,151 @@
|
||||
import { createHash, randomUUID } from "node:crypto";
|
||||
import { mkdirSync, mkdtempSync, rmSync, writeFileSync } from "node:fs";
|
||||
import { tmpdir } from "node:os";
|
||||
import { join, resolve } from "node:path";
|
||||
|
||||
import { describe, expect, it } from "vitest";
|
||||
|
||||
import { CreditService } from "../../apps/api/src/credits.js";
|
||||
import { ModelConfigurationService, modelIds, type ModelConfigCandidate } from "../../apps/api/src/model-configuration.js";
|
||||
import { ModelContractEvidenceService } from "../../apps/api/src/model-contract-evidence.js";
|
||||
import { ProjectService } from "../../apps/api/src/projects.js";
|
||||
import { RegistrationService } from "../../apps/api/src/registration.js";
|
||||
import { MockResendAdapter } from "../../apps/api/src/resend-adapter.js";
|
||||
import { settleGenerationCredits } from "../../apps/worker/src/credit-settlement.js";
|
||||
import { generationErrorCategories, generationErrorRegistry } from "../../apps/worker/src/generation-error-registry.js";
|
||||
|
||||
const now = Date.parse("2026-08-04T08:00:00.000Z");
|
||||
|
||||
function matrix(modelId: string) {
|
||||
return {
|
||||
error_mapping: [...generationErrorCategories],
|
||||
execution_modes: ["sync", "async", "poll"],
|
||||
model_id: modelId,
|
||||
pure_text: { outputs: 1, status: "passed" },
|
||||
ratios: ["3:4", "1:1", "4:3", "9:16"].map((ratio) => ({ outputs: 1, ratio, status: "passed" })),
|
||||
reference_image: { outputs: 1, status: "passed" },
|
||||
};
|
||||
}
|
||||
|
||||
function editable(models: ReturnType<ModelConfigurationService["read"]>["models"]): ModelConfigCandidate[] {
|
||||
return models.map(({ config_version: _version, runtime_availability: _runtime, ...candidate }) => structuredClone(candidate));
|
||||
}
|
||||
|
||||
function writeModelEvidence(modelId: string, value: unknown) {
|
||||
const root = process.env.DADA_WP7_02_STATE_EVIDENCE_ROOT;
|
||||
if (!root) return;
|
||||
const directory = resolve(root, modelId.replaceAll(".", "_"));
|
||||
mkdirSync(directory, { recursive: true });
|
||||
writeFileSync(resolve(directory, "deterministic-state.json"), `${JSON.stringify(value, null, 2)}\n`);
|
||||
}
|
||||
|
||||
describe("TDD-WP7-EXT-001 controlled deterministic state boundaries", () => {
|
||||
it("proves nine errors, settlement replay, invalidation and full revalidation independently per model", () => {
|
||||
const evidenceIds = new Set<string>();
|
||||
for (const modelId of modelIds) {
|
||||
const root = mkdtempSync(join(tmpdir(), "dada-wp7-02-state-"));
|
||||
const databasePath = join(root, "dada.sqlite3");
|
||||
const registration = new RegistrationService({
|
||||
challengePepper: Buffer.alloc(32, 0xd1), clock: () => now,
|
||||
currentPrivacyNoticeVersion: "p0a-registration-notice-v1", databasePath,
|
||||
invitePepper: Buffer.alloc(32, 0xd2), resend: new MockResendAdapter(), sessionPepper: Buffer.alloc(32, 0xd3),
|
||||
});
|
||||
const projects = new ProjectService({ clock: () => now, databasePath });
|
||||
let credits = new CreditService({ clock: () => now, databasePath });
|
||||
try {
|
||||
const settlements = [];
|
||||
for (const outcome of ["succeeded", "failed", "rejected"] as const) {
|
||||
const userId = randomUUID();
|
||||
registration.database.prepare(`INSERT INTO users (
|
||||
user_id, normalized_email, role, status, counts_toward_stage_limit, registration_id, created_at
|
||||
) VALUES (?, ?, 'user', 'active', 1, ?, ?)`).run(userId, `${userId}@example.invalid`, randomUUID(), now);
|
||||
registration.database.prepare("INSERT INTO user_profiles (user_id, creator_name, social_id) VALUES (?, 'WP7 User', '@wp7')").run(userId);
|
||||
registration.database.prepare("INSERT INTO credit_accounts (user_id, available_balance, reserved_balance, updated_at) VALUES (?, 1, 0, ?)").run(userId, now);
|
||||
const generationId = projects.createProjectForGeneration({ ownerId: userId, prompt: "controlled fixture", ratio: "1:1", status: "queued" }).generation.generationId;
|
||||
credits.reserveGeneration({ creditCost: 1, generationId, modelId, operationKey: `generation:${generationId}:reserve`, userId });
|
||||
const input = { generationId, operationKey: `generation:${generationId}:finalize`, outcome };
|
||||
const first = settleGenerationCredits(credits, input);
|
||||
const replay = settleGenerationCredits(credits, input);
|
||||
credits.close();
|
||||
credits = new CreditService({ clock: () => now, databasePath });
|
||||
const restartReplay = settleGenerationCredits(credits, input);
|
||||
expect(replay).toEqual(first);
|
||||
expect(restartReplay).toEqual(first);
|
||||
const account = credits.readAccount(userId);
|
||||
const ledger = registration.database.prepare(`SELECT entry_type, COUNT(*) AS count FROM credit_ledger
|
||||
WHERE user_id = ? AND entry_type IN ('generation_commit', 'generation_release') GROUP BY entry_type`).get(userId);
|
||||
expect(account).toMatchObject({ availableBalance: outcome === "succeeded" ? 0 : 1, reservedBalance: 0 });
|
||||
expect(ledger).toEqual({ count: 1, entry_type: outcome === "succeeded" ? "generation_commit" : "generation_release" });
|
||||
settlements.push({
|
||||
available_after: account.availableBalance,
|
||||
ledger_entries: 1,
|
||||
outcome,
|
||||
reserved_after: account.reservedBalance,
|
||||
replay_count: 2,
|
||||
});
|
||||
}
|
||||
|
||||
const models = new ModelConfigurationService({ clock: () => now, database: registration.database });
|
||||
const contracts = new ModelContractEvidenceService({ clock: () => now, database: registration.database, models });
|
||||
const firstEvidenceHash = `sha256:${createHash("sha256").update(`${modelId}:v1`).digest("hex")}`;
|
||||
const first = contracts.recordVerified({
|
||||
actorId: "wp7-02-controlled", expectedConfigSetVersion: 1,
|
||||
evidence: {
|
||||
evidence_hash: firstEvidenceHash,
|
||||
evidence_ref: `wp7-02:${modelId}:first`,
|
||||
matrix: matrix(modelId), model_id: modelId,
|
||||
verified_at: new Date(now).toISOString(), verifier_ref: "wp7-02-controlled",
|
||||
},
|
||||
idempotencyKey: `wp7-02:${modelId}:first`,
|
||||
});
|
||||
expect(first.model.contract_validation_status).toBe("verified");
|
||||
const candidates = editable(first.configuration.models);
|
||||
const target = candidates.find((candidate) => candidate.model_id === modelId)!;
|
||||
target.route_profile = { ...target.route_profile, contract_revision: 2 };
|
||||
const changed = models.replace({
|
||||
actorId: "wp7-02-controlled", expectedConfigSetVersion: 2,
|
||||
idempotencyKey: `wp7-02:${modelId}:change`, models: candidates,
|
||||
});
|
||||
const invalidated = changed.models.find((model) => model.model_id === modelId)!;
|
||||
expect(invalidated.contract_validation_status).toBe("unverified");
|
||||
const secondEvidenceHash = `sha256:${createHash("sha256").update(`${modelId}:v2`).digest("hex")}`;
|
||||
const revalidated = contracts.recordVerified({
|
||||
actorId: "wp7-02-controlled", expectedConfigSetVersion: 3,
|
||||
evidence: {
|
||||
evidence_hash: secondEvidenceHash,
|
||||
evidence_ref: `wp7-02:${modelId}:second`,
|
||||
matrix: matrix(modelId), model_id: modelId,
|
||||
verified_at: new Date(now + 1_000).toISOString(), verifier_ref: "wp7-02-controlled",
|
||||
},
|
||||
idempotencyKey: `wp7-02:${modelId}:second`,
|
||||
});
|
||||
expect(revalidated.model.contract_validation_status).toBe("verified");
|
||||
expect(contracts.read(modelId, first.model.config_version)?.evidence_hash).toBe(firstEvidenceHash);
|
||||
expect(contracts.read(modelId, revalidated.model.config_version)?.evidence_hash).toBe(secondEvidenceHash);
|
||||
|
||||
const evidenceId = `sha256:${createHash("sha256").update(`${modelId}:deterministic-state`).digest("hex")}`;
|
||||
expect(evidenceIds.has(evidenceId)).toBe(false);
|
||||
evidenceIds.add(evidenceId);
|
||||
writeModelEvidence(modelId, {
|
||||
contract_change: {
|
||||
after_change: { config_set_version: 3, config_version: invalidated.config_version, status: invalidated.contract_validation_status },
|
||||
after_revalidation: { config_set_version: 4, config_version: revalidated.model.config_version, status: revalidated.model.contract_validation_status },
|
||||
before_change: { config_set_version: 2, config_version: first.model.config_version, status: first.model.contract_validation_status },
|
||||
full_matrix_reapplied: true,
|
||||
},
|
||||
error_scenarios: generationErrorCategories.map((category) => ({ category, ...generationErrorRegistry[category], source: "deterministic_local", status: "passed" })),
|
||||
evidence_id: evidenceId,
|
||||
model_id: modelId,
|
||||
settlements,
|
||||
status: "passed",
|
||||
});
|
||||
} finally {
|
||||
credits.close();
|
||||
projects.close();
|
||||
registration.close();
|
||||
rmSync(root, { force: true, recursive: true });
|
||||
}
|
||||
}
|
||||
expect(evidenceIds.size).toBe(3);
|
||||
});
|
||||
});
|
||||
@@ -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);
|
||||
});
|
||||
|
||||
Reference in New Issue
Block a user