Harden soft Scattermind guardrails

README.md

@@ -53,7 +53,7 @@ Candidate trace note: candidate-level `sourceItemId` / `traceId`, `sourceTitle`
 
 Soft Scattermind labels are accepted at the bridge boundary so Scattermind does not need to use harsh verdict copy in its own product surface. Lens text can say `Continue first`, `Make tangible`, `Try next`, `Evidence next`, `Hold for later`, or `Set aside`; Build Order objects and direct bridge/envelope sections can use matching camel/snake-case keys such as `continueFirst`, `evidenceNext`, `holdForLater`, and `setAside`. Ranker maps those to `doFirst / validateNext / defer / park` while preserving the softer original label in `sourceQuote` or candidate source trace.
 
-Lane safety note: explicit Scattermind `defer` / `park` hints are hard rails, not mild suggestions. Source `nonGoals` / `avoid` guardrails are also hard enough to keep conflicting candidates out of Do first / Validate next even when their local scoring hints look attractive; the result will mark the lane source as `source-non-goal` so the handoff can explain that the candidate needs guardrail resolution before active work. Handoff `source.requiresSourceTrace` is true only when a real source artifact/title is present; plain idea-only ranking still warns about a missing artifact ID when it carries prompt provenance, but it does not spam source-section/evidence warnings meant for Scattermind artifacts. Handoff `readiness` now gives downstream bridge consumers a deterministic gate: `ready`, `usable-with-warnings`, `needs-source-context`, or `blocked`, with blockers and next checks for missing evidence, source trace, duplicate IDs, or active source-non-goal conflicts. Handoff `activeSlice` (`ranker-active-slice-v1`) is the compact machine-readable continuation unit: one active item, its proof/evidence/success/kill signals, source anchor, held-back items, readiness status, and the rule that only this slice is build-ready. For tired first-screen users, `brief.decisionReceipt` repeats the one active move, first proof step, evidence question, held-back items, source anchor, and the handoff rule that only Do first is active; use it as the compact result strip before showing the full lane board. For low-friction handoff, `/api/rank-feedback` also detects a raw Scattermind/Concept Map JSON object pasted into `idea`, `ideaText`, `optionsText`, or wrapper keys such as `payload`; it expands that object before ranking and reports `input.embeddedPayloadSource` so the public form can accept copy/paste exports without a custom import screen. Exact free Snapshot JSON (`working_name`, `restated_idea`, `lenses.shape`, `questions_to_sit_with`, `reference_code`) is rankable too: Ranker derives a source-traced manual proof plus evidence-question candidates instead of rejecting the Snapshot for having fewer than two explicit next moves. If a Concept Map only carries `questions_to_sit_with` / `questionsToSitWith` / `openQuestions` and no explicit build-order lanes or action threads, Ranker converts those questions into Validate-next evidence actions with source trace instead of pretending they are software features.
+Lane safety note: explicit Scattermind `defer` / `park` hints are hard rails, not mild suggestions. Source `nonGoals` / `avoid` guardrails are also hard enough to keep conflicting candidates out of Do first / Validate next even when their local scoring hints look attractive; soft guardrail language such as “this is not a dashboard” or “keep auth/billing/workspaces out until proof” is promoted into non-goals, not merely background context. The result will mark the lane source as `source-non-goal` so the handoff can explain that the candidate needs guardrail resolution before active work. Handoff `source.requiresSourceTrace` is true only when a real source artifact/title is present; plain idea-only ranking still warns about a missing artifact ID when it carries prompt provenance, but it does not spam source-section/evidence warnings meant for Scattermind artifacts. Handoff `readiness` now gives downstream bridge consumers a deterministic gate: `ready`, `usable-with-warnings`, `needs-source-context`, or `blocked`, with blockers and next checks for missing evidence, source trace, duplicate IDs, or active source-non-goal conflicts. Handoff `activeSlice` (`ranker-active-slice-v1`) is the compact machine-readable continuation unit: one active item, its proof/evidence/success/kill signals, source anchor, held-back items, readiness status, and the rule that only this slice is build-ready. For tired first-screen users, `brief.decisionReceipt` repeats the one active move, first proof step, evidence question, held-back items, source anchor, and the handoff rule that only Do first is active; use it as the compact result strip before showing the full lane board. For low-friction handoff, `/api/rank-feedback` also detects a raw Scattermind/Concept Map JSON object pasted into `idea`, `ideaText`, `optionsText`, or wrapper keys such as `payload`; it expands that object before ranking and reports `input.embeddedPayloadSource` so the public form can accept copy/paste exports without a custom import screen. Exact free Snapshot JSON (`working_name`, `restated_idea`, `lenses.shape`, `questions_to_sit_with`, `reference_code`) is rankable too: Ranker derives a manual proof active slice plus evidence questions, carrying the Snapshot reference code/title into provenance so a Snapshot-only handoff does not need a paid Concept Map before it can produce a useful build order. If a Concept Map only carries `questions_to_sit_with` / `questionsToSitWith` / `openQuestions` and no explicit build-order lanes or action threads, Ranker converts those questions into Validate-next evidence actions with source trace instead of pretending they are software features.
 
 Recommended payload shape:
 

scripts/check-rank-feedback.mjs

@@ -171,6 +171,33 @@ try {
   assert.equal(hardRail.brief.quickGlance.topPick, 'Manual source-traced build order preview');
   assert.equal(hardRail.handoff.readiness.status, 'ready');
 
+  const softGuardrailResponse = await fetch(`${base}/api/rank-feedback`, {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({
+      schema: 'prioritix-feature-set-v1',
+      sourceName: 'Scattermind',
+      artifactId: 'concept_map_soft_guardrails',
+      idea: 'A Concept Map clarified that the continuation should stay lightweight and defend one next move.',
+      context: 'Solo builder. This is not a dashboard. Keep auth, billing, saved workspaces, and collaboration out until proof.',
+      mode: 'mvp',
+      featureSet: {
+        features: [
+          { id: 'decision-strip', title: 'One-screen decision strip', description: 'Show the active build slice with source trace and the first proof step.', evidenceNeeded: 'Can a tired user explain the first move?', rankerHints: { value: 8, effort: 2, confidence: 7, urgency: 7, risk: 2 } },
+          { id: 'saved-workspace', title: 'Saved workspace dashboard', description: 'Accounts, auth, billing, team collaboration, and saved project dashboards.', rankerHints: { value: 10, effort: 1, confidence: 9, urgency: 9, risk: 1 } },
+          { id: 'copy-brief', title: 'Copy decision brief', description: 'Copy the defended order and source anchor into notes.', recommendedLane: 'validate-next' },
+        ],
+      },
+    }),
+  });
+  assert.equal(softGuardrailResponse.status, 200);
+  const softGuardrail = await softGuardrailResponse.json();
+  assert.ok(softGuardrail.input.decisionContext.nonGoals.some(item => /not a dashboard/i.test(item)), 'soft dashboard language should become a non-goal');
+  assert.ok(softGuardrail.input.decisionContext.nonGoals.some(item => /Keep auth, billing, saved workspaces/i.test(item)), 'keep-out-until-proof language should become a non-goal');
+  assert.equal(softGuardrail.ranked.find(item => item.id === 'saved-workspace').lane.source, 'source-non-goal');
+  assert.equal(softGuardrail.buildOrder.doFirst[0], 'decision-strip');
+  assert.ok(!/dashboard/i.test(softGuardrail.brief.quickGlance.topPick));
+
   const hintedResponse = await fetch(`${base}/api/rank-feedback`, {
     method: 'POST',
     headers: { 'Content-Type': 'application/json' },

server.js

@@ -182,7 +182,9 @@ function guardrailsFromContextText(value = '') {
     const cleaned = cleanText(sentence, 180);
     if (!cleaned) continue;
     if (/^(avoid|no|do not|don't|dont|must not|never|non-goal|non goal|not yet|out of scope)\b/i.test(cleaned)) nonGoals.push(cleaned.replace(/^non[- ]goal\s*:\s*/i, ''));
-    else if (/\b(avoid|no auth|no account|no billing|no workspace|not a dashboard|without accounts|before proof|manual proof|solo builder|constraint)\b/i.test(cleaned)) constraints.push(cleaned);
+    else if (/\b(keep|leave|hold)\b[^.!?]{0,80}\b(out|later|until proof|after proof|not yet)\b/i.test(cleaned)) nonGoals.push(cleaned);
+    else if (/\b(not a dashboard|not another dashboard|no dashboard swamp|dashboard swamp)\b/i.test(cleaned)) nonGoals.push(cleaned);
+    else if (/\b(avoid|no auth|no account|no billing|no workspace|without accounts|before proof|manual proof|solo builder|constraint)\b/i.test(cleaned)) constraints.push(cleaned);
   }
   return { nonGoals: uniqueList(nonGoals), constraints: uniqueList(constraints) };
 }