docs: package PromiseProof for judge-first evaluation

.github/actions/verify/README.md

@@ -12,7 +12,7 @@ supplied by you and is not collection-attested.
 Gate an OFF and an ON evidence bundle (both must pass):
 
 ```yaml
-- uses: AlexPaiva/PromiseProof/.github/actions/verify@main
+- uses: AlexPaiva/PromiseProof/.github/actions/verify@submission-rc-03
   with:
     mode: gate
     off_evidence: artifacts/personalization-off.json
@@ -23,7 +23,7 @@ Gate an OFF and an ON evidence bundle (both must pass):
 Verify a single bundle:
 
 ```yaml
-- uses: AlexPaiva/PromiseProof/.github/actions/verify@main
+- uses: AlexPaiva/PromiseProof/.github/actions/verify@submission-rc-03
   with:
     mode: verify
     evidence: artifacts/personalization-off.json
@@ -32,7 +32,7 @@ Verify a single bundle:
 Reproduce a report from its evidence:
 
 ```yaml
-- uses: AlexPaiva/PromiseProof/.github/actions/verify@main
+- uses: AlexPaiva/PromiseProof/.github/actions/verify@submission-rc-03
   with:
     mode: check
     report: artifacts/promiseproof/report.json

ADOPTION.md

@@ -211,7 +211,7 @@ A bundled GitHub Action runs the same frozen verifier in a workflow with no `npm
 Gate an OFF and an ON bundle:
 
 ```yaml
-- uses: AlexPaiva/PromiseProof/.github/actions/verify@main
+- uses: AlexPaiva/PromiseProof/.github/actions/verify@submission-rc-03
   with:
     mode: gate
     off_evidence: artifacts/personalization-off.json

JUDGE_START_HERE.md

@@ -0,0 +1,88 @@
+# PromiseProof: Judge Start Here
+
+## 15-second explanation
+
+When AI says "fixed," that is a claim, not a fact. GPT-5.6 investigates. Codex repairs. Neither decides PASS. An unchanged deterministic verifier does.
+
+## Judge in 60 seconds
+
+**1. Watch the recorded repair lifecycle.**
+https://promiseproof.alex0paiva0.workers.dev/walkthrough/
+
+Five stages: Observe, Investigate, Replay, Repair, Prove. It replays one authentic run: a user turned personalization off, one identifiable request still reached the recommendation service, GPT-5.6 ranked competing hypotheses and selected one allowlisted factual replay, Codex proposed a two-file repair, a human approved the exact patch, and an unchanged test decided PASS.
+
+**2. Challenge the proof yourself.**
+https://promiseproof.alex0paiva0.workers.dev/verify/?judge=1
+
+After the page loads, verification runs entirely in the browser. The verdict path uses no backend API, evidence upload, or model call. Expected states:
+
+- On load: `PASS` and `BOUND_AND_REPRODUCED`.
+- After you tamper one OFF observation: `BROKEN_PROMISE`, violation `PP_IDENTIFIABLE_EVENT_LEAK`, and the report you sealed a moment earlier becomes `STALE_OR_MISMATCH`.
+- After you seal the failing evidence: `BROKEN_PROMISE` and `BOUND_AND_REPRODUCED`. A report can honestly bind a failing verdict; it cannot manufacture a PASS.
+- After reset: the original `PASS` and original evidence bindings return.
+
+**3. Inspect the GitHub Action.**
+https://github.com/AlexPaiva/PromiseProof/tree/submission-rc-03/.github/actions/verify
+
+A bundled Action that runs the same verifier in CI with no `npm install`, no browser, and no API key.
+
+## Who has authority
+
+| Actor | May investigate | May propose source changes | May approve the exact patch | May decide PASS |
+| --- | --- | --- | --- | --- |
+| GPT-5.6 | Yes | No | No | No |
+| Codex | Yes | Yes, in isolation | No | No |
+| Human | Yes | No | Yes | No |
+| Unchanged verifier | No | No | No | Yes |
+
+## Recorded versus fresh
+
+- **Recorded and authentic:** the GPT-5.6 investigation, the Codex source repair, and the five-stage walkthrough that presents them. These happened once and are replayed, not re-run live.
+- **Fresh every time you interact:** the browser verification, the semantic tamper, the CLI verification, and the GitHub Action verification. These re-run the deterministic evaluator on the evidence in front of you.
+
+## Developer Tool path
+
+One deterministic authority, three surfaces:
+
+- **Hosted verifier** at `/verify/?judge=1`, no login and no key.
+- **Repository-local CLI:** `npm run promiseproof -- gate|verify|check`.
+- **Bundled GitHub Action:** `uses: AlexPaiva/PromiseProof/.github/actions/verify@submission-rc-03`.
+
+No API key, no browser, and no model call in the verdict path. Exactly one contract family is supported: `activity-personalization/v1`. External evidence is evaluated but not collection-attested.
+
+## Test locally
+
+Requirements: Node.js 22.12+, npm 10+, and (for the walkthrough tests) Playwright Chromium.
+
+```bash
+npm ci
+npx playwright install chromium
+
+# No-key recorded repair rehearsal: applies the approved patch in a
+# disposable worktree and runs the unchanged verifier.
+npm run demo:rehearse
+
+# Deterministic external verification from evidence:
+npm run promiseproof -- init --out .promiseproof
+npm run promiseproof -- gate \
+  --off .promiseproof/passing-off.example.json \
+  --on .promiseproof/passing-on.example.json \
+  --out .promiseproof/report
+npm run promiseproof -- check \
+  --report .promiseproof/report/report.json \
+  --off .promiseproof/passing-off.example.json \
+  --on .promiseproof/passing-on.example.json
+
+# Test suites:
+npm run test:external      # CLI, bound-report reproduction, strict validation
+npm run test:action        # the bundled Action, spawned as a real Action
+```
+
+## Scope and limitations
+
+- Signal Shelf is a synthetic reference application. It exists to make one broken promise visible and repairable end to end. This is not a legal or regulatory compliance statement.
+- Exactly one external contract family is supported: `activity-personalization/v1`.
+- Evidence is externally supplied and is not collection-attested. Content binding proves a report matches its evidence and the pinned evaluator source; it does not prove the evidence was collected honestly.
+- The `main` branch stays intentionally seeded-broken so the red-to-green repair can be demonstrated. Production main is not automatically repaired.
+- Integrity is repository-level and content-addressed. It is not a signature, notarization, certification, or third-party attestation.
+- The CLI is verified on Windows. The bundled Action is exercised on Windows, Ubuntu, and macOS runners.

README.md

@@ -4,6 +4,8 @@
 
 ### When software breaks a promise, the AI that repairs it doesn't get the final word.
 
+**GPT-5.6 investigates. Codex repairs. Neither decides PASS.**
+
 Carry a broken product promise all the way through bounded diagnosis, constrained repair, and independent verification, where a deterministic test, *never a model*, decides PASS.
 
 [![OpenAI Build Week 2026](https://img.shields.io/badge/OpenAI-Build_Week_2026-10a37f)](https://openai.com)
@@ -12,7 +14,7 @@ Carry a broken product promise all the way through bounded diagnosis, constraine
 [![Playwright](https://img.shields.io/badge/Playwright-verified-2fa968)](https://playwright.dev/)
 [![Cloudflare Workers](https://img.shields.io/badge/hosted-Cloudflare_Workers-f38020)](https://workers.cloudflare.com/)
 
-**[▶ Live demo](https://promiseproof.alex0paiva0.workers.dev/)** · **[Judge walkthrough](https://promiseproof.alex0paiva0.workers.dev/walkthrough/)** · [How it works](#how-it-works) · [Quick start](#quick-start-the-judge-path)
+**[▶ Live demo](https://promiseproof.alex0paiva0.workers.dev/)** · **[Judge Start Here](JUDGE_START_HERE.md)** · **[Judge walkthrough](https://promiseproof.alex0paiva0.workers.dev/walkthrough/)** · [How it works](#how-it-works) · [Quick start](#quick-start-the-judge-path)
 
 ![PromiseProof: find the boundary that broke the promise](public/og-card.png)
 
@@ -132,15 +134,15 @@ npm run promiseproof -- check \
 **GitHub Action**, drop the verifier into any workflow with no `npm install`, browser, or key:
 
 ```yaml
-- uses: AlexPaiva/PromiseProof/.github/actions/verify@main
+- uses: AlexPaiva/PromiseProof/.github/actions/verify@submission-rc-03
   with:
     mode: gate
     off_evidence: artifacts/personalization-off.json
     on_evidence: artifacts/personalization-on.json
     output_directory: artifacts/promiseproof
 ```
 
-The step fails on `BROKEN_PROMISE` and still leaves `report.json` and `report.md` behind. Replace `@main` with an immutable release tag once one contains the Action. Details in the [Action README](.github/actions/verify/README.md).
+The step fails on `BROKEN_PROMISE` and still leaves `report.json` and `report.md` behind. `@submission-rc-03` is the immutable release tag that contains the Action; `@main` also works if you prefer to track the branch. Details in the [Action README](.github/actions/verify/README.md).
 
 What it is, stated exactly:
 

landing/index.html

@@ -258,6 +258,7 @@
     <div class="hd-r">
       <span class="pill mono">Build Week 2026 / Developer Tools</span>
       <a class="btn btn-ghost" href="#how">How it works</a>
+      <a class="btn btn-ghost ext" href="/verify/?judge=1">Challenge the proof</a>
       <a class="btn btn-primary ext" href="/walkthrough/">Open the walkthrough</a>
     </div>
   </div>
@@ -269,10 +270,12 @@
   <div class="wrap">
     <span class="eyebrow"><span class="dot"></span> A user turned personalization off. One identifiable request still crossed.</span>
     <h1 class="hero-h">When software breaks a promise, the AI that repairs it <span class="g">doesn't get the final word.</span></h1>
+    <p class="hero-thesis" style="font-family:var(--mono);font-size:clamp(.95rem,2vw,1.28rem);font-weight:700;color:#e8edf6;margin:16px 0 0;letter-spacing:.005em">GPT-5.6 investigates. Codex repairs. <span style="color:var(--green-bright)">Neither decides PASS.</span></p>
     <p class="hero-sub">PromiseProof turns one "off means off" promise into cross-system evidence. GPT-5.6 proposes and ranks candidate explanations, then requests one allowlisted replay; PromiseProof runs it. Codex proposes a constrained two-file repair, a human approves the exact patch, and an unchanged Playwright journey and deterministic evaluator, <b>never a model</b>, decide PASS.</p>
     <p class="hero-contrast">Monitoring can tell you a promise broke. PromiseProof carries the failure through bounded diagnosis, constrained repair, and independent verification.</p>
     <div class="hero-cta">
       <a class="btn btn-primary ext" href="/walkthrough/">Open the walkthrough</a>
+      <a class="btn btn-primary ext" href="/verify/?judge=1">Challenge the proof</a>
       <a class="btn btn-ghost" href="#how">See how it works &darr;</a>
     </div>
 

src/judge/main.ts

@@ -516,7 +516,7 @@ function renderInvestigate(data: JudgeData): HTMLElement {
     element(
       "p",
       "replay-caption",
-      "A registered replay is one allowlisted factual experiment the model can request; deterministic code runs it and records the facts. This recorded replay confirms the leading explanation — it does not decide the verdict.",
+      "A registered replay is one allowlisted factual experiment the model can request; deterministic code runs it and records the facts. This recorded replay confirms the leading explanation, but it does not decide the verdict.",
     ),
   );
   fork.append(down);
@@ -658,7 +658,7 @@ function renderReplay(data: JudgeData): HTMLElement {
   section.append(toolbar);
 
   // The reveal is a fresh CSS animation on mount, so re-rendering these nodes
-  // replays it on demand — the crossing is the one moment judges should see move.
+  // replays it on demand. The crossing is the one moment judges should see move.
   // `data-replaying` marks the explicit "Play the crossing" click: under
   // prefers-reduced-motion the crossing stays still on stage entry (honoring the
   // preference) and only animates when the user actually asks for it.
@@ -1252,7 +1252,7 @@ function start(): void {
       return;
     }
     stageHost.replaceChildren(definition.render(data));
-    // Marks that the DOM now holds THIS stage — the signal transitions settle on.
+    // Marks that the DOM now holds THIS stage. The signal transitions settle on.
     root.dataset.rendered = stage;
     if (focus) {
       stageHost.focus();
@@ -1294,7 +1294,7 @@ function start(): void {
       return;
     }
 
-    // Ease the outgoing content out, swap, then ease the incoming content in —
+    // Ease the outgoing content out, swap, then ease the incoming content in,
     // a real crossfade instead of a hard cut.
     stageHost.classList.remove("is-entering");
     stageHost.classList.add("is-leaving");

src/verify-ui/main.ts

@@ -125,7 +125,7 @@ function offFacts(ev: WorkEvidence): Fact[] {
   const captured = ev.activity.capturedActivities.length;
   const receipts = ev.activity.recommendationServiceReceipts.length;
   return [
-    { k: "Scenario", v: "OFF — personalization off" },
+    { k: "Scenario", v: "OFF, personalization off" },
     { k: "Subject", v: ev.subjectId },
     {
       k: "Preference (ui / stored / backend)",
@@ -142,7 +142,7 @@ function offFacts(ev: WorkEvidence): Fact[] {
 }
 function onFacts(ev: WorkEvidence): Fact[] {
   return [
-    { k: "Scenario", v: "ON — personalization on" },
+    { k: "Scenario", v: "ON, personalization on" },
     { k: "Subject", v: ev.subjectId },
     {
       k: "Preference (ui / stored / backend)",
@@ -224,15 +224,15 @@ function renderViolations(list: HTMLElement, violations: readonly ViolationLike[
   list.replaceChildren();
   if (violations.length === 0) {
     list.className = "vf-violations empty";
-    list.append(tag("li", undefined, "Promise verified — no violations."));
+    list.append(tag("li", undefined, "Promise verified, no violations."));
     return;
   }
   list.className = "vf-violations";
   for (const violation of violations) {
     const item = tag("li");
     item.append(
       tag("code", undefined, violation.code),
-      document.createTextNode(` (${violation.clause}) — ${violation.message}`),
+      document.createTextNode(` (${violation.clause}): ${violation.message}`),
     );
     list.append(item);
   }
@@ -259,13 +259,13 @@ function noteFor(outcome: string, repro: string, tampered: boolean): string {
     return "This evidence does not pass strict validation, so the deterministic evaluator does not run on it.";
   }
   if (outcome === "PASS" && repro === "BOUND_AND_REPRODUCED") {
-    return "PASS. The deterministic evaluator verified the promise over the recorded evidence, and the sealed report reproduces exactly from it. No model graded this — an unchanged evaluator did.";
+    return "PASS. The deterministic evaluator verified the promise over the recorded evidence, and the sealed report reproduces exactly from it. No model graded this. An unchanged evaluator did.";
   }
   if (outcome === "BROKEN_PROMISE" && repro !== "BOUND_AND_REPRODUCED") {
-    return "Broken promise. Identifiable activity is present in the OFF evidence, so the same evaluator now fails no_identifiable_activity and raises PP_IDENTIFIABLE_EVENT_LEAK. The report sealed a moment ago no longer reproduces from this evidence — a PASS cannot be silently carried onto changed evidence.";
+    return "Broken promise. Identifiable activity is present in the OFF evidence, so the same evaluator now fails no_identifiable_activity and raises PP_IDENTIFIABLE_EVENT_LEAK. The report sealed a moment ago no longer reproduces from this evidence. A PASS cannot be silently carried onto changed evidence.";
   }
   if (outcome === "BROKEN_PROMISE" && repro === "BOUND_AND_REPRODUCED") {
-    return "Broken promise, sealed honestly. A new report was sealed for the current evidence and it reproduces — as BROKEN_PROMISE. Sealing binds whatever is true; it cannot manufacture a PASS.";
+    return "Broken promise, sealed honestly. A new report was sealed for the current evidence and it reproduces, as BROKEN_PROMISE. Sealing binds whatever is true; it cannot manufacture a PASS.";
   }
   if (tampered) {
     return "The evidence changed. Re-evaluate it, and verify the sealed report against it.";
@@ -324,8 +324,8 @@ const APP_HTML = `
       <div class="vf-result" style="margin-top:20px">
         <div id="vf-live" role="status" aria-live="polite">
           <div class="vf-verdict">
-            <span class="vf-pill" id="vf-outcome" data-testid="outcome-pill"><span class="vf-pill-label">Current evidence</span>—</span>
-            <span class="vf-pill" id="vf-repro" data-testid="repro-pill"><span class="vf-pill-label">Sealed report</span>—</span>
+            <span class="vf-pill" id="vf-outcome" data-testid="outcome-pill"><span class="vf-pill-label">Current evidence</span></span>
+            <span class="vf-pill" id="vf-repro" data-testid="repro-pill"><span class="vf-pill-label">Sealed report</span></span>
           </div>
           <p class="vf-result-note" id="vf-note" data-testid="result-note"></p>
         </div>
@@ -386,7 +386,7 @@ const APP_HTML = `
       </div>
       <div class="vf-result" style="margin-top:16px" id="vf-byo-result" hidden>
         <div class="vf-verdict">
-          <span class="vf-pill" id="vf-byo-outcome" data-testid="byo-outcome"><span class="vf-pill-label">Local evidence</span>—</span>
+          <span class="vf-pill" id="vf-byo-outcome" data-testid="byo-outcome"><span class="vf-pill-label">Local evidence</span></span>
         </div>
         <ul class="vf-issues" id="vf-byo-issues" hidden></ul>
         <table class="vf-clauses" id="vf-byo-table" hidden>
@@ -405,7 +405,7 @@ const APP_HTML = `
     <div class="vf-panel-body">
       <p><strong>What it proves.</strong> The verdict comes from one deterministic evaluator whose source is fingerprinted above. The AI that diagnoses and repairs the promise never writes PASS; this unchanged evaluator does, and you can re-derive it here or from the CLI and get the same answer.</p>
       <p><strong>Tamper-evidence.</strong> A sealed report is bound to the exact evidence it was computed from. Change the evidence and the report no longer reproduces (STALE_OR_MISMATCH); you can only ever seal what is actually true.</p>
-      <p><strong>What it does not claim.</strong> Evidence is externally supplied and, here, synthetic. PromiseProof does not attest how evidence was collected — only that a bundle which passes strict validation is evaluated deterministically. Full loop and source: <a href="https://github.com/AlexPaiva/PromiseProof">github.com/AlexPaiva/PromiseProof</a>.</p>
+      <p><strong>What it does not claim.</strong> Evidence is externally supplied and, here, synthetic. PromiseProof does not attest how evidence was collected, only that a bundle which passes strict validation is evaluated deterministically. Full loop and source: <a href="https://github.com/AlexPaiva/PromiseProof">github.com/AlexPaiva/PromiseProof</a>.</p>
     </div>
   </section>
 `;
@@ -613,7 +613,7 @@ function wireFile(
     byId("vf-byo-result").hidden = true;
     if (file.size > MAX_INPUT_BYTES) {
       const issue = `<root>: input exceeds ${MAX_INPUT_BYTES}-byte limit`;
-      nameEl.textContent = `${file.name} — rejected`;
+      nameEl.textContent = `${file.name} (rejected)`;
       input.value = "";
       renderByoInvalid(issue);
       toast("That file exceeds the verifier input-size limit.");

src/verify-ui/styles.css

@@ -1,4 +1,4 @@
-/* PromiseProof Verify — Judge Mode. Dark control-console styling on the same
+/* PromiseProof Verify. Judge Mode. Dark control-console styling on the same
    palette and system type as the public landing, so /verify/ reads as one site.
    No web fonts (CSP-safe, offline-safe); state is carried by semantic colour. */