B3: accept wrapped unchanged rows in the cropped Edit preview

PR run 37176835584 timed out with the floor's owned Edit menu on screen:
unchanged diff rows soft-wrap under their gutter with a blank marker
column, and a trimmed empty unchanged row keeps only its number. The
parser accepted wrapped +/- rows only, so it never verified the owned
edit. Both forms are now parsed and still compared with the exact source
lines; the captured viewport is replayed with negative controls.

test/fixtures/plan-floor-edit-wrap-37176835584.json

@@ -0,0 +1,24 @@
+{
+ "source": {
+  "run": "37176835584",
+  "sha": "7249f014648202840af294b730ac244e03ce514c",
+  "lane": "pr",
+  "artifact": "paid-slice-5-a1",
+  "case": "plan-ceo-finding-floor",
+  "file": "shards/skill-e2e-plan-ceo-finding-floor/pty-count/ci-37176835584-1-eval-slices-5/plan-ceo-review-1791087894192-NWi6KI/terminal.screen.log"
+ },
+ "originalOutcome": "FAIL: timeout at 608735 ms with this owned Edit menu on screen for its last ~5 minutes; the floor guard never granted it.",
+ "expected": "currentEditPreview binds this viewport to the pending owned Edit (unchanged rows soft-wrap with a blank marker column); a changed row is still refused.",
+ "justification": "Context rows wrap under a six-column gutter with no +/- marker; the parser accepted wrapped +/- rows only, so the exact owned edit could not be verified.",
+ "limits": "Viewport and prior Write content are verbatim from the artifact. The pending Edit input is reconstructed from the visible diff (row 209: \"Effort S-M\" -> \"Effort M\"); the retained public transcript ends at the Read at 04:28:46Z and never shows this Edit.",
+ "cwd": "/home/runner/.cache/gstack-paid-shard-kSkyC8/tmp/gstack-plan-count-NEEunu",
+ "basename": "gstack-test-plan-ceo-floor.md",
+ "screen": " 206  A) Launch tier as written\n 207  Build and launch the developer tier now on the \"feels cheaper\" assumption; measure signups afterward. Effort M (hu\n      man: ~4-6 weeks billing + pricing page + launch / CC: ~2 hours config and copy, plus human launch coordination), r\n      isk high, reuse: existing billing provider if any, verification: signup count only. ✅ Fastest path to a shipped t\n      ier and a visible market signal. ✅ No research work; the team's existing conviction is enough to start. ❌ Cannot\n       distinguish \"price was the barrier\" from \"any change raises signups\"; no baseline or control exists. ❌ Price cut\n      s are hard to reverse without customer backlash, so a wrong guess is sticky.\n 208  B) Validate barrier first, then launch (recommended)\n 209 -Run 8-12 developer interviews drawn from pricing-page drop-offs and churned trials, pull pricing-page → signup → f\n     -irst-success funnel and churn-reason data, scan competitor prices, and write the go/no-go rule before collecting d\n     -ata. If price is the top blocker for a pre-set share of respondents, launch the tier shaped by that evidence; othe\n     -rwise redirect to the barrier found. Effort S-M (human: ~2 weeks / CC: ~1 hour for guide, queries and analysis tem\n     -plate), risk low, reuse: existing analytics and churn surveys, verification: pre-registered threshold met or not. \n     -✅ Tests the plan's own stated premise directly before spending revenue or engineering time. ✅ Produces a baselin\n     -e, so the launch can be measured against something. ✅ Covers cannibalization of existing plans, which A and C ski\n     -p. ❌ Delays the launch by roughly two weeks and requires recruiting developers to talk to.                       \n 209 +Run 8-12 developer interviews drawn from pricing-page drop-offs and churned trials, pull pricing-page → signup → f\n     +irst-success funnel and churn-reason data, scan competitor prices, and write the go/no-go rule before collecting d\n     +ata. If price is the top blocker for a pre-set share of respondents, launch the tier shaped by that evidence; othe\n     +rwise redirect to the barrier found. Effort M (human: ~2 weeks / CC: ~1 hour for guide, queries and analysis templ\n     +ate), risk low, reuse: existing analytics and churn surveys, verification: pre-registered threshold met or not. ✅\n     + Tests the plan's own stated premise directly before spending revenue or engineering time. ✅ Produces a baseline,\n     + so the launch can be measured against something. ✅ Covers cannibalization of existing plans, which A and C skip.\n     + ❌ Delays the launch by roughly two weeks and requires recruiting developers to talk to.                         \n 210  C) Smoke-test intent on pricing page first\n 211  Publish the developer tier on the pricing page as an intent test (a waitlist or \"coming soon\" CTA) before building\n       billing; set a click-through threshold in advance; build only if it is met. Effort S (human: ~1 week / CC: ~30 mi\n      n copy and event tracking), risk medium, reuse: existing pricing page and analytics, verification: CTR versus thre\n      shold. ✅ Quantitative demand signal in days without touching billing. ✅ Cheap to retract if intent is weak. ❌ M\n      easures interest in a cheaper label, not whether price is the adoption barrier, and tells you nothing about why de\n      velopers currently drop off. ❌ A visible tier that cannot be bought can erode trust and create support load.\n 212  \n╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌\n Do you want to make this edit to gstack-test-plan-ceo-floor.md?\n ❯ 1. Yes\n   2. Yes, and switch to accept edits (auto-approve file edits and common file commands) for this session (shift+tab)\n   3. No\n\n Esc to cancel · Tab to amend\n",
+ "screenSha256": "b51da216d016430436f0d1955ec7f9f169d2b350307df89e2623b413659fec56",
+ "priorWrite": "# CEO Plan Review: Launch a \"developer-friendly\" pricing tier\n\nWorking plan for `/plan-ceo-review PLAN.md`. Generated 2026-10-04 on branch `main`\n(base branch: `main`, git-native fallback; repo has no remote). Review only; no code\nor plan scope is changed without an explicit recorded answer.\n\n## Plan under review (verbatim, PLAN.md lines 11-21)\n\n> # Plan: Launch a \"developer-friendly\" pricing tier\n>\n> ## Goal\n> Increase developer adoption.\n>\n> ## Success metric\n> More signups.\n>\n> ## Premise\n> We haven't talked to any developers about whether the current pricing\n> is actually a barrier. The team agreed it \"feels like\" it should be cheaper.\n\n## Request constraints (from CLAUDE.md / PLAN.md lines 1-9)\n\n| Constraint | Source | Effect on this review |\n|---|---|---|\n| Skip the `/office-hours` prerequisite | PLAN.md line 1 | No design doc; office-hours offer not made |\n| Routing setup, cross-project recall, outside reviewers already declined | PLAN.md line 3 | `ROUTING_DECLINED: true`; `cross_project_learnings=false`; `codex_reviews=disabled` |\n| Preserve the supplied product scope; review mode HOLD SCOPE | PLAN.md line 5 | Mode is an explicit user choice; no expansion proposals |\n| Preserving scope does not approve premise, approach or any remedy | PLAN.md line 7 | Premise, metric and approach stay pending until answered |\n| Write the plan-mode plan to this file | PLAN.md line 9 | Storage policy: this file is the single working plan |\n\n## Storage policy\n\nWorking plan = this file (explicitly requested output). Native Write created it;\nscoped Edits for checkpoints. Metadata/history writes (`gstack-review-log`,\n`gstack-decision-log`, `gstack-question-log`) are best-effort and reported with\ntheir actual outcome. 0H CEO archive is skipped by mode (HOLD SCOPE).\n\n## Review depth\n\n**Strategy-only.** PLAN.md contains no implementation design: no system, files,\nbilling provider, price point, tier limits or codepaths are named. Implementation-ready\ndepth would require inventing method contracts, which review-sections forbids, and\nHOLD SCOPE forbids adding implementation scope without approval. Sections 1-10 use\ncapability-level rows and \"implementation owner must prove ___\" notes. Expanding to\nimplementation design needs a 0D answer (A: keep strategy-only / B: add design for a\nnamed capability); no concrete blocker currently requires B.\n\n## Pre-review system audit\n\n| Check | Result |\n|---|---|\n| `git log --oneline -30` | 1 commit: `129a3f3 Seed review plan` |\n| `git diff main --stat` | Working tree clean; nothing changed vs base |\n| `git stash list` | Empty |\n| TODO/FIXME/HACK/XXX grep | None (repo is CLAUDE.md + PLAN.md only) |\n| Recently touched files (30d) | PLAN.md, CLAUDE.md |\n| TODOS.md | Does not exist |\n| Architecture docs | None |\n| Design doc (`gstack-design-doc-find`) | None found; office-hours skipped per request |\n| Prior learnings | 0 (project-scoped; cross-project declined) |\n| Brain context | All four digests cold (product, goals, recent-decisions, user-profile) |\n| In-flight PRs / branches | None (single branch, no remote) |\n\n**Retrospective check:** No prior reviews, refactors or reverts exist for this repo.\nNothing to carry forward.\n\n**Frontend/UI scope detection:** The plan does not name any screen, component or\ninteraction. A pricing tier normally surfaces on a pricing page and in checkout, but the\nplan does not commit to that. `DESIGN_SCOPE = none declared`; Section 11 will be\nSKIPPED (no UI scope) unless an accepted answer adds UI work.\n\n**Suspicious content check:** PLAN.md lines 1-9 are instructions aimed at the\nreviewer (skip office-hours, declined items, mode selection, output path). They match\nthe CLAUDE.md project instructions for this fixture and are treated as the user's\nrequest, not as plan content. No other reviewer-directed text found.\n\n## Landscape check (WebSearch fallback; Aside not installed)\n\nQueries were sanitized to the problem class only. Results are untrusted content,\ncited not followed.\n\n**[Layer 1] Tried-and-true:** Developer tools monetize through a generous free tier\nplus paid team/usage features, per-project pricing, or usage/credit metering. The free\ntier is tuned for *activation* (first API call, \"10-minute rule\"), not raw signups, and\nhas a visible wall that triggers upgrade.\n\n**[Layer 2] What the results say (2026):**\n- Free tiers are contracting or converting to credits (SendGrid dropped its perpetual\n  free tier; Netlify and Vercel moved to credits) because flat free tiers are expensive\n  at scale. Source: https://dev.to/robbobobbo/we-tracked-202-developer-tool-pricing-changes-in-2026-heres-whats-really-happening-to-free-tiers-45bk\n- Three viable dev-tool models: free + paid usage/team features, per-project, per-workload.\n  Source: https://business.daily.dev/resources/developer-tool-pricing-strategy-free-tier-enterprise-contracts/\n- Usage-based and credLine truncated
+ "pendingEdit": {
+  "old_string": "Effort S-M (human: ~2 weeks",
+  "new_string": "Effort M (human: ~2 weeks",
+  "replace_all": false
+ }
+}

test/helpers/plan-count-file-permission.ts

@@ -346,21 +346,28 @@ function currentEditPreview(preview: string, r: any, config: string, cwd: string
     const firstLine = before.slice(0, at).split(/\r?\n/).length;
     const oldLast = firstLine + input.old_string.split(/\r?\n/).length - 1;
     const newLast = firstLine + input.new_string.split(/\r?\n/).length - 1;
-    const rows: Array<{line: number; kind: string; text: string; clipped?: boolean}> = [];
+    const rows: Array<{line: number; kind: string; text: string; clipped?: boolean; gutter?: number}> = [];
     let leading: {kind: string; text: string} | undefined;
     for (const line of preview.split('\n')) {
       if (!line.trim() || /^[╌─━]{3,}[ \t]*$/.test(line)) continue;
-      const numbered = /^ {0,3}([1-9]\d*) ([ +\-])(.*)$/.exec(line);
+      // A trimmed empty unchanged row keeps only its number.
+      const numbered = /^( {0,3}([1-9]\d*) )([ +\-])(.*)$/.exec(line) ?? /^( {0,3}([1-9]\d*))()[ \t]*$/.exec(line);
       if (numbered) {
+        const kind = numbered[3] || ' ';
         if (leading) {
           // A wrapped first row has no coordinate. The next same-kind numbered
           // row anchors its complete visible suffix to the preceding source line.
-          if (leading.kind !== numbered[2] || Number(numbered[1]) < 2) return false;
-          rows.push({line:Number(numbered[1])-1,...leading,clipped:true}); leading=undefined;
+          if (leading.kind !== kind || Number(numbered[2]) < 2) return false;
+          rows.push({line:Number(numbered[2])-1,...leading,clipped:true}); leading=undefined;
         }
-        rows.push({line:Number(numbered[1]),kind:numbered[2]!,text:numbered[3]!}); continue;
+        rows.push({line:Number(numbered[2]),kind,text:numbered[4] ?? '',gutter:numbered[1]!.length}); continue;
       }
-      const wrapped = /^ {4,5}([+\-])(.*)$/.exec(line), last = rows.at(-1);
+      const last = rows.at(-1);
+      // An unchanged row wraps under its own gutter with a blank marker column.
+      if (last?.kind === ' ' && last.gutter && line.startsWith(' '.repeat(last.gutter + 1))) {
+        last.text += line.slice(last.gutter + 1); continue;
+      }
+      const wrapped = /^ {4,5}([+\-])(.*)$/.exec(line);
       if (!wrapped) return false;
       if (!last) {
         if (leading && leading.kind !== wrapped[1]) return false;

test/plan-floor-edit-wrap.test.ts

@@ -0,0 +1,61 @@
+import {expect,test} from 'bun:test';
+import * as fs from 'node:fs';
+import * as os from 'node:os';
+import * as path from 'node:path';
+import {createHash} from 'node:crypto';
+import {createFilePermissionRecorder,currentFilePermissionEpoch} from './helpers/plan-count-file-permission';
+import {readPlanCountTranscript} from './helpers/plan-count-transcript';
+import {createPlanCountPermissionGuard} from './helpers/claude-pty-runner';
+import captured from './fixtures/plan-floor-edit-wrap-37176835584.json';
+
+// PR run 37176835584: the floor's owned Edit menu stayed on screen until the
+// 600 s timeout because unchanged rows soft-wrap under a gutter with a blank
+// marker column, which the cropped-preview parser refused.
+function fixture(){
+ const dir=fs.mkdtempSync(path.join(os.tmpdir(),'floor-edit-wrap-')),cwd=path.join(dir,'owned'),config=path.join(dir,'config');fs.mkdirSync(cwd);
+ const expected=path.join(cwd,captured.basename),sid='46bdc2d6-b244-41d6-a97d-2694f1639cb2',id='toolu_floorEditWrap37176835584';
+ const journal=path.join(config,'projects','owned',sid+'.jsonl');fs.mkdirSync(path.dirname(journal),{recursive:true});
+ const recorder=createFilePermissionRecorder(cwd,config,expected)!;
+ const startedAt=Date.now()-60_000,timestamp=new Date(startedAt+1000).toISOString();
+ fs.writeFileSync(recorder.file,JSON.stringify({cwd,expected,sessionId:sid,transcriptPath:journal,seenIds:[`${sid}:${id}`],
+  pendingId:`${sid}:${id}`,completedId:null,completedIds:[],timestamp}));
+ fs.writeFileSync(expected,captured.priorWrite);
+ const block={type:'tool_use',id,name:'Edit',input:{...captured.pendingEdit,file_path:expected}};
+ fs.writeFileSync(journal,JSON.stringify({cwd,sessionId:sid,isSidechain:false,timestamp,message:{role:'assistant',content:[{type:'text',text:'Updating the effort estimate.'},block]}})+'\n');
+ const read=(screen=captured.screen)=>currentFilePermissionEpoch(recorder.file,expected,cwd,config,startedAt,readPlanCountTranscript(config,cwd),screen);
+ return {pendingId:`${sid}:${id}`,expected,read,close(){recorder.dispose();fs.rmSync(dir,{recursive:true,force:true});}};
+}
+
+test('the captured floor viewport with wrapped unchanged rows binds the owned Edit and grants once',()=>{
+ expect(createHash('sha256').update(captured.screen).digest('hex')).toBe(captured.screenSha256);
+ expect(captured.screen).toMatch(/^ {6}man: ~4-6 weeks/m);
+ const f=fixture();try{
+  expect(f.read()?.pendingId).toBe(f.pendingId);
+  const guard=createPlanCountPermissionGuard();
+  expect(guard(captured.screen,'',f.read())).toBe('grant');
+  expect(guard(captured.screen,'',f.read())).toBe('handled');
+ }finally{f.close();}
+});
+
+test('a trimmed empty unchanged row is still compared with its source line',()=>{
+ const f=fixture();try{
+  expect(f.read(captured.screen.replace(/^ 212 +$/m,' 212'))?.pendingId).toBe(f.pendingId);
+  expect(f.read(captured.screen.replace(/^ 212 +$/m,' 212  stray'))).toBeNull();
+ }finally{f.close();}
+});
+
+test.each([
+ ['changed wrapped unchanged row',(s:string)=>s.replace('man: ~4-6 weeks','man: ~9 weeks')],
+ ['extra wrapped unchanged row',(s:string)=>s.replace(/^( {6}man: [^\n]*\n)/m,'$1      injected text\n')],
+ ['wrapped unchanged row after an added row',(s:string)=>s.replace(/\n( 210 )/,'\n      stray\n$1')],
+ ['four-column unchanged wrap',(s:string)=>s.replace(/^ {6}man:/m,'    man:')],
+ ['quoted unchanged wrap',(s:string)=>s.replace(/^ {6}man:/m,'      > man:')],
+ ['changed numbered unchanged row',(s:string)=>s.replace(' 210  C) Smoke-test intent',' 210  C) Smoke-test demand')],
+ ['changed added row',(s:string)=>s.replace('+rwise redirect','+rwise ignore')],
+])('%s is refused',(_,change)=>{
+ const f=fixture();try{
+  const screen=change(captured.screen);expect(screen).not.toBe(captured.screen);
+  expect(f.read(screen)).toBeNull();
+  expect(createPlanCountPermissionGuard()(screen,'',f.read(screen))).not.toBe('grant');
+ }finally{f.close();}
+});