fix(yield): preserve append-only release history

labs/22-yield/distribution/release-notes/2026-08-01-evaluation-surface.md

@@ -0,0 +1,9 @@
+# Evaluation cases and evidence boundary
+
+- Add six provenance-bound skill-conversion cases with their thin model-facing
+  skills and Yield workflow programs.
+- Add a reproducible source-size harness and fail-closed summary validator.
+- Keep raw transcripts, temporary repositories, and large run artifacts out of
+  Git; published behavior summaries must bind an immutable artifact digest.
+- Validate the same evaluation surface in both the source tree and projected
+  public repository.

labs/22-yield/yield/docs/examples.md

@@ -64,9 +64,9 @@ replay them.
 ## Evaluated examples
 
 The [first-party evaluation suite](../evals/) runs every example-library
-pattern through TypeScript, Python, Go, and Rust fixtures. It also checks the
-supervisor's resume, replay, divergence, blocking, and source-change behavior.
+pattern through TypeScript, Python, Go, and Rust. It also checks resume, replay,
+changed behavior, blocking, and changed source.
 
-These results establish that the tested Yield revision executes the declared
-control flow. They do not compare Yield with prose or evaluate whether an
-agent's judgment is correct.
+These results show that the tested Yield version runs the workflow steps in
+code. They do not compare Yield with prose or test whether an agent's judgment
+is correct.

labs/22-yield/yield/evals/README.md

@@ -5,18 +5,19 @@ tool, company, prompt, or skill.
 
 The suite answers two questions:
 
-1. Can each checked-in example workflow reach its declared terminal result
+1. Can each checked-in example workflow reach its expected final result
    through every supported SDK?
-2. Does the supervisor preserve its core runtime invariants when a run resumes,
-   replays, blocks, or encounters changed code?
+2. Does the runtime behave correctly when a run resumes, replays, blocks, or
+   encounters changed code?
 
 ## Current coverage
 
 - 10 workflow patterns written by this project.
 - 4 SDKs: TypeScript, Python, Go, and Rust.
-- 40 end-to-end fixture runs.
-- 5 runtime invariants: resume and complete, deterministic replay, loud replay
-  divergence, failed-requirement blocking, and source-change refusal/migration.
+- 40 end-to-end workflow tests.
+- 5 runtime checks: resume and complete, repeat the same saved step, stop when
+  behavior changes, block when a rule fails, and require approval for changed
+  source.
 
 Run the exact suite and refresh the checked-in result:
 
@@ -35,20 +36,20 @@ npm test
 
 A passing result proves that the tested Yield revision:
 
-- executes each owned workflow fixture to `completed`;
+- executes each owned workflow test to `completed`;
 - runs command steps rather than asking the model to invent their outputs;
 - presents requests in the program-defined order;
 - resumes from recorded responses;
-- reproduces a saved frontier during replay;
+- returns to the same saved step during replay;
 - stops on changed behavior or failed requirements.
 
 ## What it does not prove
 
 This suite does not prove that Yield is better than prose, that an agent's
 judgment is correct, or that illustrative commands are production-safe. The
-fixtures supply deterministic agent and human responses so the suite can test
-the coding and supervision layer in isolation.
+Fixed test data supplies agent and human responses so the suite can test only
+the code-controlled workflow layer.
 
-`results/latest.json` is a compact, website-safe result. Its source digest is
+`results/latest.json` is a compact, website-safe result. Its source hash is
 computed from the CLI, engine, protocol, SDKs, example workflows, fixtures, and
 evaluation harness. CI reruns the suite instead of trusting that file alone.

labs/22-yield/yield/evals/results/README.md

@@ -1,8 +1,8 @@
 # Published result
 
-`latest.json` records the latest first-party conformance run. It contains no
+`latest.json` records the latest first-party test run. It contains no
 model transcripts and no third-party comparison data.
 
-The result is publishable only when `npm test` reruns all workflow fixtures and
-runtime invariants successfully and reproduces the same source digest, counts,
+The result is publishable only when `npm test` reruns all workflow tests and
+runtime checks successfully and reproduces the same source hash, counts,
 and case identities.

labs/22-yield/yield/evals/results/latest.json

@@ -1,8 +1,8 @@
 {
   "schema_version": 2,
   "methodology_version": "1.0",
-  "generated_at": "2026-08-01T17:40:24.932Z",
-  "source_digest": "51a8f4ecc511324915e1cab9f8bbecbba4d9d1c21ab6d285e7a381ba733be6dd",
+  "generated_at": "2026-08-01T17:53:20.762Z",
+  "source_digest": "8426ef59771921109fa7aa2d242e089d13f391aea60f2aebf509da93b7b534ae",
   "status": "passed",
   "workflow_conformance": {
     "passed": 40,
@@ -270,7 +270,7 @@
       {
         "id": "deterministic-replay",
         "test": "TestReplayIsDeterministic",
-        "assertion": "the saved log reproduces the same frontier",
+        "assertion": "the saved log returns to the same next step",
         "status": "passed"
       },
       {
@@ -282,24 +282,24 @@
       {
         "id": "requirement-block",
         "test": "TestFailedRequirementBlocksRun",
-        "assertion": "a failed requirement produces a blocked terminal result",
+        "assertion": "a failed rule ends the run as blocked",
         "status": "passed"
       },
       {
         "id": "source-change",
         "test": "TestDigestMismatchRefusedThenMigrates",
-        "assertion": "changed source is refused until migration is explicit",
+        "assertion": "changed source is refused until the user accepts the change",
         "status": "passed"
       }
     ]
   },
   "claim_boundary": {
-    "summary": "Yield executes the tested workflows and preserves the tested supervisor invariants.",
+    "summary": "Yield executes the tested workflows and passes the tested runtime checks.",
     "compares_tools": false,
     "tests_agent_judgment": false,
     "exclusions": [
       "No claim that Yield is better than prompts or another tool.",
-      "Fixtures replace agent and human judgment with deterministic responses.",
+      "Fixed test data replaces agent and human judgment.",
       "Illustrative commands do not establish production safety for a real project."
     ]
   }

labs/22-yield/yield/evals/scripts/run.mjs

@@ -11,10 +11,10 @@ const libraryRoot = join(yieldRoot, "examples/library")
 const languages = ["typescript", "python", "go", "rust"]
 const runtimeCases = [
   ["resume-complete", "TestEndToEndRunResumeComplete", "a recorded response advances the run to completion"],
-  ["deterministic-replay", "TestReplayIsDeterministic", "the saved log reproduces the same frontier"],
+  ["deterministic-replay", "TestReplayIsDeterministic", "the saved log returns to the same next step"],
   ["replay-divergence", "TestReplayDivergenceFailsLoudly", "changed behavior stops replay instead of reusing the wrong result"],
-  ["requirement-block", "TestFailedRequirementBlocksRun", "a failed requirement produces a blocked terminal result"],
-  ["source-change", "TestDigestMismatchRefusedThenMigrates", "changed source is refused until migration is explicit"],
+  ["requirement-block", "TestFailedRequirementBlocksRun", "a failed rule ends the run as blocked"],
+  ["source-change", "TestDigestMismatchRefusedThenMigrates", "changed source is refused until the user accepts the change"],
 ]
 const excludedDirectories = new Set([
   ".git", ".yield", "node_modules", "runs", "raw", "artifacts",
@@ -105,12 +105,12 @@ async function evaluate() {
         cases: invariants,
       },
       claim_boundary: {
-        summary: "Yield executes the tested workflows and preserves the tested supervisor invariants.",
+        summary: "Yield executes the tested workflows and passes the tested runtime checks.",
         compares_tools: false,
         tests_agent_judgment: false,
         exclusions: [
           "No claim that Yield is better than prompts or another tool.",
-          "Fixtures replace agent and human judgment with deterministic responses.",
+          "Fixed test data replaces agent and human judgment.",
           "Illustrative commands do not establish production safety for a real project.",
         ],
       },
@@ -133,8 +133,8 @@ if (process.argv.includes("--write")) {
   for (const field of ["workflow_conformance", "runtime_invariants", "claim_boundary"]) {
     if (JSON.stringify(published[field]) !== JSON.stringify(result[field])) throw new Error(`published ${field} is stale`)
   }
-  console.log(`passed ${result.workflow_conformance.passed}/${result.workflow_conformance.total} workflow fixtures`)
-  console.log(`passed ${result.runtime_invariants.passed}/${result.runtime_invariants.total} runtime invariants`)
+  console.log(`passed ${result.workflow_conformance.passed}/${result.workflow_conformance.total} workflow tests`)
+  console.log(`passed ${result.runtime_invariants.passed}/${result.runtime_invariants.total} runtime checks`)
 } else {
   console.log(JSON.stringify(result, null, 2))
 }

labs/22-yield/yield/evals/scripts/validate.mjs

@@ -8,25 +8,25 @@ const fail = (message) => { throw new Error(message) }
 
 if (result.schema_version !== 2) fail("unsupported result schema")
 if (result.methodology_version !== "1.0") fail("unsupported methodology")
-if (!/^[0-9a-f]{64}$/.test(result.source_digest)) fail("source digest is invalid")
+if (!/^[0-9a-f]{64}$/.test(result.source_digest)) fail("source hash is invalid")
 if (result.status !== "passed") fail("published result is not passing")
 
 const workflows = result.workflow_conformance
 if (workflows.patterns !== 10 || workflows.languages.length !== 4) {
   fail("workflow matrix is incomplete")
 }
 if (workflows.total !== 40 || workflows.passed !== workflows.total) {
-  fail("not every workflow fixture passed")
+  fail("not every workflow test passed")
 }
 if (workflows.cases.length !== workflows.total) fail("workflow case list is incomplete")
 
 const runtime = result.runtime_invariants
 if (runtime.total !== 5 || runtime.passed !== runtime.total) {
-  fail("not every runtime invariant passed")
+  fail("not every runtime check passed")
 }
 if (runtime.cases.length !== runtime.total) fail("runtime case list is incomplete")
 
 if (result.claim_boundary.compares_tools !== false) fail("result must not compare tools")
 if (result.claim_boundary.tests_agent_judgment !== false) fail("result must isolate agent judgment")
 
-console.log(`validated ${workflows.passed} workflow fixtures and ${runtime.passed} runtime invariants`)
+console.log(`validated ${workflows.passed} workflow tests and ${runtime.passed} runtime checks`)