@openwop/openwop-conformance 2.46.19 → 2.46.20

package/CHANGELOG.md

@@ -1,3 +1,9 @@
+## [2.46.20] — 2026-10-24 — RFC 0081 eval envelope legs; RFC 0249 Draft
+
+- **Two new legs in `v2-eval-trajectory-trials` (RFC 0081 §C)**, on the existing `conformance-eval-trajectory` run, so no extra run: `0081.eval-envelope` (one `eval.started` and one `eval.completed` bracket every `eval.scored`, `eval.completed` precedes `run.completed`, and its counts equal the summary's) and `0081.summary-served` (the served `EvalSummary` validates and its counts match its rows). Judges in `lib/eval-envelope-witness.ts`, whose self-tests fail each wrong host answer. Before this, no v2 leg read `eval.completed`.
+- **Host impact:** hosts that advertise `agents.evalSuite.trajectory` and list the fixture (openwop-app, MyndHyve) run both legs. A host that omits `eval.completed` fails `0081.eval-envelope`.
+- RFC 0249 (v2 proposal operations and an install read-back for RFC 0096) is filed as `Draft`. Its gap register lands in the packed `spec/v1/gaps.json`, which opened this cycle.
+
 ## [2.46.19] — 2026-10-22 — RFC 0248: ctx.callAI attempt accounting and retry policy
 
 - **New scenario `v2-callai-attempt-accounting`** with eight legs: `0248.attempts-count`, `output-tokens-sum`, `usage-incomplete`, `structured-output`, `attempts-replayed`, `max-attempts-cap`, `no-budget-growth` and `retry-malformed-refused`. They are gated on `aiProviders.callAccounting.attempts` / `.retryPolicy` and the fixture.

package/dist/spec-artifacts.lock.json

@@ -1,5 +1,5 @@
 {
   "package": "@openwop/spec-artifacts",
-  "version": "2.46.19",
-  "stampSha256": "8b0bfeeda6c1a1e8bbac41bc73b01f70e71ad88ca5dc98aee4a373b6611db396"
+  "version": "2.46.20",
+  "stampSha256": "426586e18ceca126f915b8517b442ecc6178e3f2cecde3325bf00dcfd76fb37d"
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@openwop/openwop-conformance",
-  "version": "2.46.19",
+  "version": "2.46.20",
   "description": "Black-box conformance suite for OpenWOP hosts (v2 current; v1 scenarios retained for the frozen v1 tree).",
   "repository": {
     "type": "git",
@@ -56,6 +56,6 @@
     "@openwop/spec-artifacts": "file:../spec-artifacts"
   },
   "peerDependencies": {
-    "@openwop/spec-artifacts": "2.46.19"
+    "@openwop/spec-artifacts": "2.46.20"
   }
 }

package/requirements.json

@@ -3,10 +3,10 @@
   "generatedFrom": "src/scenarios/*.test.ts",
   "counts": {
     "files": 747,
-    "tests": 2822,
-    "withStableId": 2822,
+    "tests": 2824,
+    "withStableId": 2824,
     "interpolatedTitles": 0,
-    "explicitIds": 2695
+    "explicitIds": 2697
   },
   "records": [
     {
@@ -31092,7 +31092,7 @@
     {
       "id": "openwop.it.v2-eval-trajectory-trials.the-rfc-0244-suite-summary-and-event-fields-validate-and-the-negatives-are-refus",
       "file": "v2-eval-trajectory-trials.test.ts",
-      "line": 150,
+      "line": 162,
       "title": "the RFC 0244 suite, summary and event fields validate and the negatives are refused",
       "explicitId": "openwop.requirement.0244.wire-shape",
       "citations": [
@@ -31131,7 +31131,7 @@
     {
       "id": "openwop.it.v2-eval-trajectory-trials.the-rfc-0244-additions-are-content-free-eval-summary-no-content-leak",
       "file": "v2-eval-trajectory-trials.test.ts",
-      "line": 173,
+      "line": 185,
       "title": "the RFC 0244 additions are content-free (eval-summary-no-content-leak)",
       "explicitId": "openwop.requirement.0244.content-free",
       "citations": [
@@ -31148,7 +31148,7 @@
     {
       "id": "openwop.it.v2-eval-trajectory-trials.the-eval-suite-fixtures-validate-against-the-v2-suite-schema",
       "file": "v2-eval-trajectory-trials.test.ts",
-      "line": 180,
+      "line": 192,
       "title": "the eval-suite fixtures validate against the v2 suite schema",
       "explicitId": "openwop.requirement.0244.fixtures-valid",
       "citations": [
@@ -31162,7 +31162,7 @@
     {
       "id": "openwop.it.v2-eval-trajectory-trials.the-rfc-0246-gating-fields-validate-and-stay-in-0-1",
       "file": "v2-eval-trajectory-trials.test.ts",
-      "line": 187,
+      "line": 199,
       "title": "the RFC 0246 gating fields validate and stay in 0–1",
       "explicitId": "openwop.requirement.0246.gated-fields-shape",
       "citations": [
@@ -31184,7 +31184,7 @@
     {
       "id": "openwop.it.v2-eval-trajectory-trials.a-fixture-entry-may-be-marked-iserror-and-only-as-a-boolean-rfc-0245-a",
       "file": "v2-eval-trajectory-trials.test.ts",
-      "line": 195,
+      "line": 207,
       "title": "a fixture entry may be marked isError, and only as a boolean (RFC 0245 §A)",
       "explicitId": "openwop.requirement.0245.fixture-error-shape",
       "citations": [
@@ -31206,7 +31206,7 @@
     {
       "id": "openwop.it.v2-eval-trajectory-trials.a-trajectory-suite-is-graded-per-the-reference-grader",
       "file": "v2-eval-trajectory-trials.test.ts",
-      "line": 205,
+      "line": 217,
       "title": "a trajectory suite is graded per the reference grader",
       "explicitId": "openwop.requirement.0244.trajectory-graded",
       "citations": [
@@ -31237,10 +31237,38 @@
         }
       ]
     },
+    {
+      "id": "openwop.it.v2-eval-trajectory-trials.an-eval-run-brackets-its-scores-with-one-eval-started-and-one-eval-completed-rfc",
+      "file": "v2-eval-trajectory-trials.test.ts",
+      "line": 251,
+      "title": "an eval run brackets its scores with one eval.started and one eval.completed (RFC 0081 §C)",
+      "explicitId": "openwop.requirement.0081.eval-envelope",
+      "citations": [
+        {
+          "section": "run-event-payloads evalStarted / evalCompleted, eval-summary.schema.json (RFC 0081 §C)",
+          "requirement": null,
+          "interpolated": true
+        }
+      ]
+    },
+    {
+      "id": "openwop.it.v2-eval-trajectory-trials.the-served-evalsummary-validates-and-its-counts-match-its-rows-rfc-0081-c",
+      "file": "v2-eval-trajectory-trials.test.ts",
+      "line": 259,
+      "title": "the served EvalSummary validates and its counts match its rows (RFC 0081 §C)",
+      "explicitId": "openwop.requirement.0081.summary-served",
+      "citations": [
+        {
+          "section": "run-event-payloads evalStarted / evalCompleted, eval-summary.schema.json (RFC 0081 §C)",
+          "requirement": null,
+          "interpolated": true
+        }
+      ]
+    },
     {
       "id": "openwop.it.v2-eval-trajectory-trials.repeated-trials-emit-one-eval-scored-per-task-per-trial",
       "file": "v2-eval-trajectory-trials.test.ts",
-      "line": 241,
+      "line": 280,
       "title": "repeated trials emit one eval.scored per task per trial",
       "explicitId": "openwop.requirement.0244.trials-per-task",
       "citations": [
@@ -31258,7 +31286,7 @@
     {
       "id": "openwop.it.v2-eval-trajectory-trials.the-per-task-trials-block-and-trialstats-follow-the-b-formulas",
       "file": "v2-eval-trajectory-trials.test.ts",
-      "line": 252,
+      "line": 291,
       "title": "the per-task trials block and trialStats follow the §B formulas",
       "explicitId": "openwop.requirement.0244.trial-stats-formula",
       "citations": [
@@ -31297,7 +31325,7 @@
     {
       "id": "openwop.it.v2-eval-trajectory-trials.a-fixture-miss-under-onmiss-fail-fails-the-trial-and-is-counted",
       "file": "v2-eval-trajectory-trials.test.ts",
-      "line": 275,
+      "line": 314,
       "title": "a fixture miss under onMiss \"fail\" fails the trial and is counted",
       "explicitId": "openwop.requirement.0244.fixture-miss-fails",
       "citations": [
@@ -31334,7 +31362,7 @@
     {
       "id": "openwop.it.v2-eval-trajectory-trials.a-suite-using-an-unadvertised-rfc-0244-feature-is-refused-at-createrun",
       "file": "v2-eval-trajectory-trials.test.ts",
-      "line": 295,
+      "line": 334,
       "title": "a suite using an unadvertised RFC 0244 feature is refused at createRun",
       "explicitId": "openwop.requirement.0244.unadvertised-feature-refused",
       "citations": [
@@ -31352,7 +31380,7 @@
     {
       "id": "openwop.it.v2-eval-trajectory-trials.under-passrule-mean-a-task-whose-trials-failed-their-trajectory-does-not-pass-rf",
       "file": "v2-eval-trajectory-trials.test.ts",
-      "line": 311,
+      "line": 350,
       "title": "under passRule \"mean\" a task whose trials failed their trajectory does not pass (RFC 0246)",
       "explicitId": "openwop.requirement.0246.mean-gated",
       "citations": [
@@ -31366,7 +31394,7 @@
     {
       "id": "openwop.it.v2-eval-trajectory-trials.a-suite-whose-tasks-failed-their-trajectory-does-not-pass-on-the-output-scores-r",
       "file": "v2-eval-trajectory-trials.test.ts",
-      "line": 324,
+      "line": 363,
       "title": "a suite whose tasks failed their trajectory does not pass on the output scores (RFC 0246)",
       "explicitId": "openwop.requirement.0246.suite-gated",
       "citations": [
@@ -31380,7 +31408,7 @@
     {
       "id": "openwop.it.v2-eval-trajectory-trials.a-fixture-entry-marked-iserror-is-replayed-as-a-tool-error-not-a-success-rfc-024",
       "file": "v2-eval-trajectory-trials.test.ts",
-      "line": 335,
+      "line": 374,
       "title": "a fixture entry marked isError is replayed as a tool error, not a success (RFC 0245)",
       "explicitId": "openwop.requirement.0245.fixture-error-replayed",
       "citations": [
@@ -31394,7 +31422,7 @@
     {
       "id": "openwop.it.v2-eval-trajectory-trials.a-suite-using-iserror-is-refused-at-createrun-without-fixtureerrors-rfc-0245-c",
       "file": "v2-eval-trajectory-trials.test.ts",
-      "line": 349,
+      "line": 388,
       "title": "a suite using isError is refused at createRun without fixtureErrors (RFC 0245 §C)",
       "explicitId": "openwop.requirement.0245.unadvertised-refused",
       "citations": [

package/schemas/CORPUS-STAMP.json

@@ -1,17 +1,17 @@
 {
   "_comment": "Provenance of @openwop/spec-artifacts (RFC 0168 §D.2). files: SHA-256 per file; the conformance suite compares the installed peer against dist/spec-artifacts.lock.json at start.",
   "package": "@openwop/spec-artifacts",
-  "version": "2.46.19",
-  "corpusTag": "v2.46.19",
+  "version": "2.46.20",
+  "corpusTag": "v2.46.20",
   "files": {
     "api/.redocly.lint-ignore.yaml": "bf5a8350b88a72fa43f59605ed8d903ed24b6cfccda5e45509c9f6ed9ee4e712",
     "api/asyncapi.yaml": "d5ecb9ee6114582be3b1f662c84bfac9ae96dae7bacb853e461168f70a8e1c7d",
     "api/grpc/openwop.proto": "c3e72bb17cba514ee98feb6434e6c9b6ea6795bfd086489ec69fd882dd1ad977",
     "api/openapi.yaml": "ac6de5303872664b78ff419c035915593e01599e02835e777a0123de1baeb398",
     "api/redocly.yaml": "b0604c89b2ca6d5076ec25725c539dad44a741a811fe524439ee6daef8baa09f",
-    "api/seams-v2.yaml": "9133614d3cad4ff6ecf2758f2f986dfd257dc5a3a2b10110e7537114a2c03a65",
-    "api/v2/asyncapi.yaml": "5e44771306b5dfa1824967a177208228d12dca66a36c10de5bb0441b8bd26c72",
-    "api/v2/openapi.yaml": "2f7e4863b8ba09d77e5ae8a7c977311b6138878ba910d7a41d50546034cad7c2",
+    "api/seams-v2.yaml": "14325981296aaf0ab2dcce8aa96527bd68935c1130bd9bf6103ee6b6c4df4f8e",
+    "api/v2/asyncapi.yaml": "b5f5369bbb35a2c3c75ff6c4f6c9281e77b141dd1d21e9495fef670d72b74a5b",
+    "api/v2/openapi.yaml": "3b17112c01e94987b0cef6a328cc9a6d3a0c0521bbafb14b1fc8abb4e10d9091",
     "api/v2/redocly.yaml": "1e66b60e6118ad11a823bb620678be464d99dfe50a40e3e6f93ec9429b88b34c",
     "schemas/README.md": "276fc1d51eb08e14e9127b9fd27761d7046ee33c12000f3a64a801322db8bf58",
     "schemas/a2a-task-state.schema.json": "75d5049dea8bd873ff0e7546f1c60c8d36c219bec7084264360be54a8be30ae1",
@@ -208,14 +208,14 @@
     "schemas/workspace-file.schema.json": "464de85c2a068243084ee9c1d969bc7cd5d8f7948574e58450d6493c38a0e1e4",
     "spec/v1/alias-detectors.json": "22444c22bb88c568339ef512553f9fc38146e22b566d11ab8608f655d688d19c",
     "spec/v1/capability-declaration-classes.json": "e7729aed5c4b4e1dd02abab0530f14cc95f5d4070fe51fb139e7f5cccefa00c6",
-    "spec/v1/core-standard-manifest.json": "6b5a2e873b59e22f27bd9832559382a42ad23e29cac8f9156d592bb19e168fb1",
+    "spec/v1/core-standard-manifest.json": "d70cc792bd36812bfcec84c6a8b0d2cd096a3fbed19967afbc8f1dd1a7d26d87",
     "spec/v1/deprecations.json": "e723bc11d7272c71789ad1303bb7f21809a8a024abf57594f9d5c35dc199ece9",
     "spec/v1/deprecations.schema.json": "3e393c405d2a41b467d8c5e3c468549078df1ce6a6d2588fc488097b95d9b55b",
     "spec/v1/end-of-support.json": "35f0bf3b70fe78b16e2a2fd81a153fc27e703f6c16376c6fe7cdf1f4f2c4e6d4",
     "spec/v1/event-codemap.json": "3da60d884157793a360da532a9fcbbfb5285636db325a74cec94b34622186d97",
     "spec/v1/event-codemap.schema.json": "d05933b2e88103aff51a2f774df97b9bda065e0fc88dbd9b114cc47a32189174",
     "spec/v1/extensions.json": "79a60754aa16cbdbdb604f8e5c00038af535a1690a4d07bd5ff7c38f0256a7bc",
-    "spec/v1/gaps.json": "52777098a223e678d12751507b1ac0d02eef7465d6263c25175489734055cc3e",
+    "spec/v1/gaps.json": "4655f3087d9917f3b6b7612af23c86a1b338c732a9c1a72ee4a705a1b4f4f7a9",
     "spec/v1/gaps.schema.json": "8fd83259f556553c9df0f53e7a82ca8c2a4771d471197f69b4e2ae0ceeacfd66",
     "spec/v1/migrations.json": "4c50760be70cb644d02427bcaf5749606700edf4513f423903cf3b1d12f5c300",
     "spec/v1/migrations.schema.json": "886779aa6c22e646db097f5df210adb018a4dd14a7b815465a18c8a7056c8f72",
@@ -314,9 +314,9 @@
     "spec/v2/path-manifest.json": "7b5679a4dad98c80d855361d88a98c8a4e045af15e310862a05b285b815dc9fc",
     "spec/v2/peer-dependency-aliases.json": "d10299280abee08258502925bc327293ee413e0108cd6e6ec75ff6110653308d",
     "spec/v2/profiles.json": "b382b261c7844798575f7c1b37c934f0da19846016a2337d883d4f5ef6e05479",
-    "spec/v2/release.json": "9870141128549af0cac3a0e7c4906c87a5ce65047591a3824690b9e20dcf59c1",
+    "spec/v2/release.json": "a8b6d20a4ed90b21b4f9feb57a9c2ec5fdb13c74a74f857e5e3aec5ee8e75e40",
     "spec/v2/retention-floors.json": "eaf3722d95c79947af1d4269ef85117e126518c588cfcf1a2b21b97269f51624",
-    "spec/v2/surface-baseline.json": "5ba73afbabe33b983c3ac74670a6aacfe32042a0d40de2faf6cdb151b055a1ae"
+    "spec/v2/surface-baseline.json": "c5aee6bbf8011e11351a4f9e0aafaf444d60af55116c543f04b55b88878cdf83"
   },
-  "corpusCommit": "d49a2818d766f78f7c3fc9d55403ed42a404ba71"
+  "corpusCommit": "3ed6fc93890b4d00c44aafc0fe3ce75a24b8dcc7"
 }

package/src/lib/eval-envelope-witness.ts

@@ -0,0 +1,74 @@
+/**
+ * RFC 0081 §C — the `eval.*` envelope of an eval run, and the summary it
+ * serves. A pure judge over one eval run, for
+ * `v2-eval-trajectory-trials.test.ts`.
+ *
+ * The rules are the `run-event-payloads` descriptions, which the `agents`
+ * family lists as normative text:
+ *   - `evalStarted` — emitted ONCE, at the start of the run;
+ *   - `evalScored` — once per task (per trial under RFC 0244 §B);
+ *   - `evalCompleted` — emitted ONCE, after all tasks and before
+ *     `run.completed`; its aggregates describe the same run the summary does.
+ * The served `EvalSummary` validates against `eval-summary.schema.json`.
+ *
+ * Until 2.46.20 no v2 leg read `eval.completed`; the frozen v1
+ * `agent-eval-run.test.ts` did (RFC 0081 correction note, 2026-10-24).
+ *
+ * @see RFCS/0081-agent-evaluation-and-scorecards.md
+ */
+
+export interface EnvelopeFinding { readonly ok: boolean; readonly doc: string; readonly message: string }
+
+type Rec = Record<string, unknown>;
+interface Event { readonly type: string; readonly sequence: number; readonly payload: Rec }
+
+const isRec = (v: unknown): v is Rec => typeof v === 'object' && v !== null && !Array.isArray(v);
+const f = (ok: boolean, doc: string, message: string): EnvelopeFinding => ({ ok, doc, message });
+
+const C = 'run-event-payloads evalStarted / evalScored / evalCompleted (RFC 0081 §C)';
+
+export interface EnvelopeOptions {
+  /** Validates an `eval.completed` payload against `run-event-payloads#/$defs/evalCompleted`. */
+  readonly completedValid: (payload: unknown) => { ok: boolean; errors: string };
+}
+
+export function judgeEvalEnvelope(events: readonly Event[], summary: Rec, opts: EnvelopeOptions): EnvelopeFinding[] {
+  const ordered = [...events].sort((a, b) => a.sequence - b.sequence);
+  const at = (type: string): number[] => ordered.flatMap((e, i) => (e.type === type ? [i] : []));
+  const started = at('eval.started');
+  const scored = at('eval.scored');
+  const completed = at('eval.completed');
+  const runCompleted = at('run.completed');
+  const out: EnvelopeFinding[] = [];
+
+  out.push(f(started.length === 1, C, `eval.started MUST be emitted once (got ${started.length})`));
+  out.push(f(completed.length === 1, C, `eval.completed MUST be emitted once (got ${completed.length})`));
+  out.push(f(scored.length > 0, C, 'the run MUST emit eval.scored for its tasks'));
+  if (started.length === 1 && scored.length > 0) {
+    out.push(f(started[0]! < scored[0]!, C, 'eval.started MUST precede every eval.scored'));
+  }
+  if (completed.length !== 1) return out;
+  const c = completed[0]!;
+  if (scored.length > 0) out.push(f(c > scored[scored.length - 1]!, C, 'eval.completed MUST follow every eval.scored ("after all tasks")'));
+  if (runCompleted.length > 0) out.push(f(c < runCompleted[0]!, C, 'eval.completed MUST precede run.completed'));
+
+  const payload = ordered[c]!.payload;
+  const v = opts.completedValid(payload);
+  out.push(f(v.ok, C, `eval.completed MUST validate: ${v.errors}`));
+  for (const k of ['taskCount', 'passedCount', 'passed'] as const) {
+    if (k in summary) out.push(f(payload[k] === summary[k], C, `eval.completed.${k} MUST equal the served summary's (event ${JSON.stringify(payload[k])}, summary ${JSON.stringify(summary[k])})`));
+  }
+  return out;
+}
+
+/** The served summary validates, and its per-task rows agree with its own counts. */
+export function judgeSummaryServed(summary: Rec, valid: (s: unknown) => { ok: boolean; errors: string }): EnvelopeFinding[] {
+  const out: EnvelopeFinding[] = [];
+  const v = valid(summary);
+  out.push(f(v.ok, 'eval-summary.schema.json (RFC 0081 §C)', `the served EvalSummary MUST validate: ${v.errors}`));
+  const tasks = Array.isArray(summary['tasks']) ? (summary['tasks'] as unknown[]).filter(isRec) : [];
+  out.push(f(summary['taskCount'] === tasks.length, 'eval-summary.schema.json (RFC 0081 §C)', `taskCount MUST equal the number of tasks (got ${JSON.stringify(summary['taskCount'])}, ${tasks.length} tasks)`));
+  const passed = tasks.filter((t) => t['passed'] === true).length;
+  out.push(f(summary['passedCount'] === passed, 'eval-summary.schema.json (RFC 0081 §C)', `passedCount MUST count the passed tasks (got ${JSON.stringify(summary['passedCount'])}, ${passed} passed)`));
+  return out;
+}

package/src/scenarios/v2-eval-trajectory-trials.test.ts

@@ -32,6 +32,13 @@
  *      lists `conformance-eval-trials` refuses it at createRun with
  *      `400 validation_error`.
  *
+ * RFC 0081 §C, on the `conformance-eval-trajectory` run (2.46.20). Judges in
+ * `lib/eval-envelope-witness.ts`.
+ *   - `eval.started` and `eval.completed` are emitted once each, bracketing
+ *     every `eval.scored`, and `eval.completed` precedes `run.completed`; its
+ *     counts equal the served summary's.
+ *   - The served `EvalSummary` validates, and its counts match its rows.
+ *
  * RFC 0246 (safety fix): a trial that failed its trajectory counts 0 in the
  * two pass decisions, while every reported score is unchanged. Judges in
  * `lib/eval-gate-witness.ts`.
@@ -73,6 +80,7 @@ import { majorProfile } from '../lib/major-profile.js';
 import { isFixtureAdvertised } from '../lib/fixtures.js';
 import { FIXTURES_DIR } from '../lib/paths.js';
 import { pollTerminal, readEvents, isRecord, type Ev } from '../lib/fixture-run-observer.js';
+import { judgeEvalEnvelope, judgeSummaryServed } from '../lib/eval-envelope-witness.js';
 import { gradeTrajectory, passAtK, passHatK, scoreStderr, withinTolerance, type TrajectorySpec } from '../lib/eval-trajectory.js';
 import { judgeMeanGated, judgeSuiteGated } from '../lib/eval-gate-witness.js';
 import { judgeFixtureErrorReplayed } from '../lib/eval-fixture-error-witness.js';
@@ -97,6 +105,9 @@ const DOC_0246 = 'eval-summary.schema.json gatedScore / gatedAggregateScore (RFC
 const ID_FIXTURE_ERROR_SHAPE = 'openwop.requirement.0245.fixture-error-shape';
 const ID_FIXTURE_ERROR = 'openwop.requirement.0245.fixture-error-replayed';
 const ID_FIXTURE_ERROR_REFUSED = 'openwop.requirement.0245.unadvertised-refused';
+const ID_ENVELOPE = 'openwop.requirement.0081.eval-envelope';
+const ID_SUMMARY = 'openwop.requirement.0081.summary-served';
+const DOC_0081 = 'run-event-payloads evalStarted / evalCompleted, eval-summary.schema.json (RFC 0081 §C)';
 const DOC_0245 = 'agents.schema.json §evalSuite.fixtureErrors (RFC 0245)';
 
 interface Task { taskId: string; input: { script: Array<{ tool: string; arguments?: unknown }>; output: string }; expected: { kind: string; trajectory?: TrajectorySpec }; fixtures?: { toolResponses?: Array<Record<string, unknown>> } }
@@ -107,6 +118,7 @@ const suiteSchema = v2Validator('agent-eval-suite');
 const summarySchema = v2Validator('eval-summary');
 const scored = v2RefValidator('run-event-payloads.schema.json#/$defs/evalScored');
 const started = v2RefValidator('run-event-payloads.schema.json#/$defs/evalStarted');
+const completedPayload = v2RefValidator('run-event-payloads.schema.json#/$defs/evalCompleted');
 
 async function http(fn: () => Promise<OpenWOPResponse>): Promise<OpenWOPResponse | null> {
   try { return await fn(); } catch { return null; }
@@ -225,6 +237,33 @@ describe('v2 RFC 0244 — host behaviour', () => {
     expect(calls, req(ID_TRAJECTORY, DOC, `the log MUST carry one agent.tool-called per scripted call (${scripted})`)).toBe(scripted);
   });
 
+  async function trajectoryRun(): Promise<{ skip: [Parameters<typeof softSkip>[0], string] } | { o: Extract<Observed, { kind: 'ok' }> }> {
+    const { doc, facet } = await evalSuiteFacet();
+    if (!doc) return { skip: ['blocked', UNREADABLE] };
+    if (facet?.['trajectory'] !== true) return { skip: ['inapplicable', 'the host does not advertise agents.evalSuite.trajectory'] };
+    const id = 'conformance-eval-trajectory';
+    if (!isFixtureAdvertised(id)) return { skip: ['inapplicable', `the host does not advertise fixture ${id}`] };
+    const o = await runSuite(id);
+    if (o.kind === 'skip') return { skip: [o.disposition, o.reason] };
+    return { o };
+  }
+
+  it('an eval run brackets its scores with one eval.started and one eval.completed (RFC 0081 §C)', async () => {
+    const r = await trajectoryRun();
+    if ('skip' in r) return softSkip(...r.skip);
+    for (const x of judgeEvalEnvelope(r.o.events, r.o.summary, { completedValid: completedPayload })) {
+      expect(x.ok, req(ID_ENVELOPE, DOC_0081, x.message)).toBe(true);
+    }
+  });
+
+  it('the served EvalSummary validates and its counts match its rows (RFC 0081 §C)', async () => {
+    const r = await trajectoryRun();
+    if ('skip' in r) return softSkip(...r.skip);
+    for (const x of judgeSummaryServed(r.o.summary, summarySchema)) {
+      expect(x.ok, req(ID_SUMMARY, DOC_0081, x.message)).toBe(true);
+    }
+  });
+
   async function trialsRun(): Promise<{ skip: [Parameters<typeof softSkip>[0], string] } | { o: Extract<Observed, { kind: 'ok' }>; suite: Suite; n: number; k: number }> {
     const { doc, facet } = await evalSuiteFacet();
     if (!doc) return { skip: ['blocked', UNREADABLE] };