fix(SatisfactionCapture): numbered list items must not parse as a 1/10 rating

LifeOS/install/hooks/SatisfactionCapture.hook.ts

@@ -75,6 +75,11 @@ const RATINGS_FILE = join(SIGNALS_DIR, 'ratings.jsonl');
 const LAST_RESPONSE_CACHE = join(BASE_DIR, 'MEMORY', 'STATE', 'last-response.txt');
 const MIN_PROMPT_LENGTH = 3;
 
+// Longest trailing comment accepted after a BARE leading digit ("8 nice", "8 - nice").
+// Real rating comments are a few words; a long tail means prose that merely starts
+// with a digit. Does not apply to "N/10" or word forms — those are unambiguous.
+const MAX_BARE_DIGIT_COMMENT = 80;
+
 // Sentence-starters that mean a leading number is describing work, not rating it
 // (e.g. "2/10 items done", "3 of the files"). Shared by the fraction and generic parsers.
 const SENTENCE_STARTERS = /^(items?|things?|steps?|files?|lines?|bugs?|issues?|errors?|times?|minutes?|hours?|days?|seconds?|percent|%|th\b|st\b|nd\b|rd\b|of\b|in\b|at\b|to\b|the\b|a\b|an\b)/i;
@@ -109,7 +114,7 @@ const WORD_NUMBERS: Record<string, number> = {
 
 // ── Explicit Rating Detection ──
 
-function parseExplicitRating(prompt: string): { rating: number; comment?: string } | null {
+export function parseExplicitRating(prompt: string): { rating: number; comment?: string } | null {
   const trimmed = prompt.trim();
 
   // Check word-form ratings first (e.g., "ten", "Eight")
@@ -140,11 +145,26 @@ function parseExplicitRating(prompt: string): { rating: number; comment?: string
 
   if (rating < 1 || rating > 10) return null;
 
+  // ")" and "]" right after the digit mark a numbered list ("1) do this,
+  // 2) do that"), not a rating. Prefer a missed rating to a false positive:
+  // a miss loses one data point, whereas rating <= 3 writes a permanent
+  // failure capture and a satisfaction-ledger row that never happened.
+  // Not covered: whitespace before the delimiter ("1 ) item") still parses
+  // as a rating — tightening that would also catch "8 . nice", so it is
+  // left alone rather than widened past this one concern.
   const afterNumber = trimmed.slice(match[1].length);
-  if (afterNumber.length > 0 && /^[/.\dA-Za-z]/.test(afterNumber)) return null;
+  if (afterNumber.length > 0 && /^[/.)\]\dA-Za-z]/.test(afterNumber)) return null;
 
   if (rest && SENTENCE_STARTERS.test(rest)) return null;
 
+  // A bare leading digit is the weakest rating evidence in this function — it is
+  // also exactly how a numbered list item starts, and "-" is an accepted rating
+  // separator, so "2 - add the header, but also back the file up first" is
+  // indistinguishable from "8 - nice" by shape alone. Length separates them:
+  // real rating comments are a few words. The "N/10" and word forms above are
+  // unambiguous rating syntax and are deliberately NOT capped.
+  if (rest && rest.length > MAX_BARE_DIGIT_COMMENT) return null;
+
   return { rating, comment: rest };
 }
 

LifeOS/install/hooks/SatisfactionCapture.test.ts

@@ -0,0 +1,113 @@
+#!/usr/bin/env bun
+/**
+ * SatisfactionCapture.test.ts - parseExplicitRating() coverage
+ *
+ * No test runner is wired up in this repo yet, so this file is a
+ * self-contained harness: `bun run SatisfactionCapture.test.ts` prints a
+ * readable PASS/FAIL line per case and exits non-zero on any failure;
+ * `bun test SatisfactionCapture.test.ts` executes the same top-level code
+ * and surfaces the same exit code.
+ *
+ * Covers the numbered-list false-positive fix (a "1) ..." prompt must never
+ * parse as rating 1) plus every explicit-rating form that must keep working
+ * byte-identically.
+ */
+
+import { parseExplicitRating } from './SatisfactionCapture.hook.ts';
+
+type Expected = { rating: number; comment?: string } | null;
+
+interface Case {
+  label: string;
+  input: string;
+  expected: Expected;
+}
+
+const cases: Case[] = [
+  // ── Numbered-list prompts must never parse as a rating (the bug) ──
+  {
+    label: 'multi-item ") " list is not rating 1',
+    input: "1) install the CLI, 2) run the setup script, 3) restart the daemon",
+    expected: null,
+  },
+  {
+    label: 'single ") " list item is not rating 1',
+    input: "1) what does this flag do?",
+    expected: null,
+  },
+  {
+    label: 'multi-line "." list is not rating 1',
+    input: "1. first thing\n2. second thing",
+    expected: null,
+  },
+  {
+    label: 'lone ") " item with no siblings is not rating 1',
+    input: "1) only one item with no siblings",
+    expected: null,
+  },
+  {
+    label: 'bracket form "] " is not rating 2',
+    input: "2] bracket form",
+    expected: null,
+  },
+
+  // ── Bare digit + long prose is a list item, not a rating ──
+  // "-" is an accepted rating separator, so these are shape-identical to "8 - nice"
+  // and only comment length tells them apart.
+  {
+    label: 'bare digit + "-" + long instruction is not a rating',
+    input: "2 - config.md -- add the do-not-edit header, and also copy the file somewhere outside the install directory first",
+    expected: null,
+  },
+  {
+    label: 'bare digit + "-" + multi-item instruction is not a rating',
+    input: "1 - we don't have the premium licence tier.  2 - the zone is set to office IPs, the datacenter IP and my home IP.  Run those checks",
+    expected: null,
+  },
+  {
+    label: 'bare digit + prose referencing other items is not a rating',
+    input: "4 and 5 -- both look like things we already moved into a private directory earlier, so they can go",
+    expected: null,
+  },
+  {
+    label: 'unambiguous N/10 form is NOT length-capped',
+    input: "8/10 really solid work here, especially the part where you caught the edge case before it shipped",
+    expected: { rating: 8, comment: 'really solid work here, especially the part where you caught the edge case before it shipped' },
+  },
+  {
+    label: 'unambiguous word form is NOT length-capped',
+    input: "eight really solid work here, especially the part where you caught the edge case before it shipped",
+    expected: { rating: 8, comment: 'really solid work here, especially the part where you caught the edge case before it shipped' },
+  },
+
+  // ── Legitimate rating forms must keep working, byte-identically ──
+  { label: 'bare digit', input: '8', expected: { rating: 8 } },
+  { label: 'bare ten', input: '10', expected: { rating: 10 } },
+  { label: 'word form', input: 'ten', expected: { rating: 10 } },
+  { label: 'word form with comment', input: 'Eight great work', expected: { rating: 8, comment: 'great work' } },
+  { label: 'N/10 with comment', input: '10/10, thank you', expected: { rating: 10, comment: 'thank you' } },
+  { label: 'N / 10 spaced', input: '9 / 10', expected: { rating: 9 } },
+  { label: 'out of 10 with comment', input: '8 out of 10 nice', expected: { rating: 8, comment: 'nice' } },
+  { label: 'dash separator', input: '8 - nice', expected: { rating: 8, comment: 'nice' } },
+  { label: 'colon separator', input: '8: nice', expected: { rating: 8, comment: 'nice' } },
+  { label: 'space separator', input: '9 solid work', expected: { rating: 9, comment: 'solid work' } },
+  { label: 'fraction sentence-starter is not a rating', input: '2/10 items', expected: null },
+  { label: 'sentence-starter is not a rating', input: '2 items', expected: null },
+  { label: 'sentence-starter is not a rating (files)', input: '3 files changed', expected: null },
+];
+
+let failures = 0;
+
+for (const { label, input, expected } of cases) {
+  const actual = parseExplicitRating(input);
+  const pass = JSON.stringify(actual) === JSON.stringify(expected);
+  if (pass) {
+    console.log(`PASS: ${label}`);
+  } else {
+    failures++;
+    console.error(`FAIL: ${label} — input=${JSON.stringify(input)} expected=${JSON.stringify(expected)} actual=${JSON.stringify(actual)}`);
+  }
+}
+
+console.log(`\n${cases.length - failures}/${cases.length} passed`);
+if (failures > 0) process.exitCode = 1;