staysfixed 0.3.1 → 0.6.0

package/src/v2/observation.js

@@ -0,0 +1,938 @@
+/**
+ * Observations, paths, and the arithmetic of difference.
+ *
+ * This is the engine room. Everything the tool learns about a product — on a phone, in a
+ * terminal, over HTTP, out of the source — arrives here as `path -> value` and is compared
+ * the same way. There is exactly one comparison in the tool, and it lives in this file.
+ *
+ * Three things happen here and nothing else does:
+ *   1. Facts are made, and a malformed one is rejected AT THE SOURCE. A bad path found three
+ *      layers later is a mystery; a bad path found at `makeObservation` is a stack trace
+ *      pointing at the collector that wrote it.
+ *   2. Two captures are compared, including the paths that appeared and the paths that
+ *      vanished — the findings that matter most and the ones pixels never see.
+ *   3. The product's own noise is MEASURED, by running the same build twice, and subtracted.
+ *      There are no tolerance settings in v2 and there is no place to add one.
+ */
+
+import { StaysFixedError } from '../core/errors.js';
+
+/**
+ * @typedef {import('./types.js').Observation} Observation
+ * @typedef {import('./types.js').ObservedValue} ObservedValue
+ * @typedef {import('./types.js').ObservationMeta} ObservationMeta
+ * @typedef {import('./types.js').Channel} Channel
+ * @typedef {import('./types.js').Capture} Capture
+ * @typedef {import('./types.js').Difference} Difference
+ * @typedef {import('./types.js').DifferenceKind} DifferenceKind
+ * @typedef {import('./types.js').Wobble} Wobble
+ * @typedef {import('./types.js').WobbleEntry} WobbleEntry
+ * @typedef {import('./types.js').WobbleSubtraction} WobbleSubtraction
+ */
+
+// ---------------------------------------------------------------------------
+// Channels
+// ---------------------------------------------------------------------------
+
+/** @type {Channel[]} */
+export const CHANNELS = [
+  'meaning',
+  'effects',
+  'complaints',
+  'results',
+  'contract',
+  'counters',
+  'pixels',
+];
+
+/**
+ * What each channel is, in the words we would use to a person. Printed by `doctor` and handed
+ * to any agent that asks the tool to describe itself.
+ * @type {Record<Channel, string>}
+ */
+export const CHANNEL_NOTES = {
+  meaning: 'What the interface says a control is and does — its role, its name, whether it is on, off or disabled. Not the underlying markup.',
+  effects: 'What the product sent out into the world: calls made, files written, processes started, things saved.',
+  complaints: 'What the product complained about: console messages, errors, crashes, the code it exited with.',
+  results: 'What the product gave back: what it printed, what it answered, what it offers other code.',
+  contract: 'The doors the source code says exist: routes, exported functions, message channels. Read without running anything.',
+  counters: 'Rough counts and rough timings. Deliberately rough — precise timing is noise, not information.',
+  pixels: 'What it looked like. Used to show a person a problem another channel already found.',
+};
+
+/**
+ * @param {unknown} value
+ * @returns {value is Channel}
+ */
+export function isChannel(value) {
+  return typeof value === 'string' && /** @type {string[]} */ (CHANNELS).includes(value);
+}
+
+// ---------------------------------------------------------------------------
+// Paths — the address space the whole tool is built on
+// ---------------------------------------------------------------------------
+
+/** Longest path we will accept. A path is an address; a runaway value must never become one. */
+const MAX_PATH_LENGTH = 512;
+
+/** Deepest value we will store. Past this something is recursing, not observing. */
+const MAX_VALUE_DEPTH = 64;
+
+/**
+ * The share of its own addresses a build may disagree with itself about before the run stops
+ * counting as a measurement at all. Half is not a tuned number and nothing depends on its
+ * exact value: it is the point past which more of the comparison has been thrown away than
+ * kept, and no answer computed from what is left deserves to be called clean.
+ */
+const STORM_SHARE = 0.5;
+
+/** Below this many addresses the share means nothing — three out of four is not a storm. */
+const STORM_FLOOR = 12;
+
+/** Control characters and newlines, which would break the store and every log line. */
+const CONTROL_CHARS = /[\u0000-\u001f\u007f]/;
+
+/**
+ * The path grammar, written out so it can be printed to whoever is wiring the tool up.
+ *
+ * A path is segments joined with dots, read left to right from the widest thing to the
+ * narrowest: surface, then place, then thing, then the property of it.
+ *
+ *   api.GET./users.status
+ *   cli.build.exit
+ *   ipc.session:create.registered
+ *   screen.home.tree.button:Save.enabled
+ *
+ * The first segment names the surface so paths from a phone and paths from a terminal sit in
+ * one list without colliding. Nothing enforces the vocabulary — a project may invent its own
+ * heads — but sticking to these keeps findings readable across products.
+ */
+export const PATH_RULES = {
+  separator: '.',
+  maxLength: MAX_PATH_LENGTH,
+  minSegments: 2,
+  escaped: 'A dot inside one segment is written %2E, and a literal percent is %25. Use joinPath() and you never have to think about it.',
+  commonHeads: ['api', 'cli', 'ipc', 'screen', 'file', 'proc', 'store', 'net', 'export', 'route', 'log', 'count'],
+  examples: [
+    'api.GET./users.status',
+    'cli.build.exit',
+    'ipc.session:create.registered',
+    'screen.home.tree.button:Save.enabled',
+  ],
+};
+
+/**
+ * Make one segment safe to sit inside a path.
+ *
+ * Escaping rather than stripping matters: `v1.2` and `v12` are different names, and a
+ * stripping scheme would quietly merge two different buttons into one address.
+ *
+ * @param {string} segment
+ * @returns {string}
+ */
+export function escapeSegment(segment) {
+  return String(segment).replace(/%/g, '%25').replace(/\./g, '%2E');
+}
+
+/**
+ * @param {string} segment
+ * @returns {string}
+ */
+export function unescapeSegment(segment) {
+  return String(segment).replace(/%2E/gi, '.').replace(/%25/g, '%');
+}
+
+/**
+ * Build a path out of parts, escaping each one.
+ * @param {(string|number)[]} segments
+ * @returns {string}
+ */
+export function joinPath(segments) {
+  return segments.map((s) => escapeSegment(String(s))).join('.');
+}
+
+/**
+ * Split a path back into its unescaped parts.
+ * @param {string} path
+ * @returns {string[]}
+ */
+export function splitPath(path) {
+  return String(path).split('.').map(unescapeSegment);
+}
+
+/**
+ * Is this a usable path? Returns the reason it is not, or null when it is fine.
+ *
+ * Kept separate from `assertPath` so a collector can filter a noisy source without throwing
+ * on every stray line.
+ *
+ * @param {unknown} path
+ * @returns {string|null}
+ */
+export function pathProblem(path) {
+  if (typeof path !== 'string') return `a path must be a string, got ${typeof path}`;
+  if (path.length === 0) return 'a path cannot be empty';
+  if (path.length > MAX_PATH_LENGTH) return `a path cannot be longer than ${MAX_PATH_LENGTH} characters (this one is ${path.length})`;
+  if (path !== path.trim()) return 'a path cannot start or end with a space';
+  // If this fires, a value has leaked into the address.
+  if (CONTROL_CHARS.test(path)) return 'a path cannot contain control characters or newlines';
+  if (path.startsWith('.') || path.endsWith('.')) return 'a path cannot start or end with a dot';
+  if (path.includes('..')) return 'a path cannot contain an empty segment (two dots in a row)';
+  const segments = path.split('.');
+  if (segments.length < PATH_RULES.minSegments) {
+    return `a path needs at least ${PATH_RULES.minSegments} parts — an address, not a name. Try something like "cli.${path}" or "screen.home.${path}"`;
+  }
+  for (const s of segments) {
+    if (s.trim().length === 0) return 'a path cannot contain a blank segment';
+  }
+  return null;
+}
+
+/**
+ * @param {unknown} path
+ * @returns {string}
+ */
+export function assertPath(path) {
+  const problem = pathProblem(path);
+  if (problem) {
+    throw new StaysFixedError(`Bad observation path: ${problem}.`, {
+      hint: `Paths look like ${PATH_RULES.examples[0]}. ${PATH_RULES.escaped}`,
+    });
+  }
+  return /** @type {string} */ (path);
+}
+
+/**
+ * Match a path against a pattern.
+ *
+ * `*` matches anything inside one segment. `**` matches any number of segments. Used by the
+ * normalisation rules and by anything that wants to talk about a family of paths at once.
+ *
+ *   matchPath('api.GET./users.status', 'api.*.*.status')            -> true
+ *   matchPath('screen.home.tree.button:Save.enabled', 'screen.**')  -> true
+ *
+ * @param {string} path
+ * @param {string} pattern
+ * @returns {boolean}
+ */
+export function matchPath(path, pattern) {
+  if (pattern === '**' || pattern === path) return true;
+  return matchFrom(path.split('.'), 0, pattern.split('.'), 0);
+}
+
+/**
+ * @param {string[]} p
+ * @param {number} startPi
+ * @param {string[]} g
+ * @param {number} startGi
+ * @returns {boolean}
+ */
+function matchFrom(p, startPi, g, startGi) {
+  let pi = startPi;
+  let gi = startGi;
+  while (gi < g.length) {
+    const part = g[gi];
+    if (part === '**') {
+      // `**` at the end swallows the rest; otherwise try every split point.
+      if (gi === g.length - 1) return true;
+      for (let k = pi; k <= p.length; k++) {
+        if (matchFrom(p, k, g, gi + 1)) return true;
+      }
+      return false;
+    }
+    if (pi >= p.length) return false;
+    if (!matchSegment(p[pi], part)) return false;
+    pi++;
+    gi++;
+  }
+  return pi === p.length;
+}
+
+/**
+ * @param {string} segment
+ * @param {string} pattern
+ * @returns {boolean}
+ */
+function matchSegment(segment, pattern) {
+  if (pattern === '*') return true;
+  if (!pattern.includes('*')) return segment === pattern;
+  const source = '^' + pattern.split('*').map(escapeRegExp).join('[^.]*') + '$';
+  return new RegExp(source).test(segment);
+}
+
+/**
+ * @param {string} s
+ * @returns {string}
+ */
+function escapeRegExp(s) {
+  return s.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+}
+
+/**
+ * A stable order for paths, so two captures always list the same way and a diff of two
+ * reports is readable.
+ *
+ * Segment by segment, and a segment that is all digits compares as a number — otherwise
+ * `item.10` sorts before `item.2` and every report reads wrong. Plain `<` rather than
+ * `localeCompare` on purpose: locale collation varies with the ICU build, and an order that
+ * changes with the machine is exactly the kind of noise this tool exists to remove.
+ *
+ * @param {string} a
+ * @param {string} b
+ * @returns {number}
+ */
+export function comparePaths(a, b) {
+  if (a === b) return 0;
+  const as = a.split('.');
+  const bs = b.split('.');
+  const n = Math.min(as.length, bs.length);
+  for (let i = 0; i < n; i++) {
+    const x = as[i];
+    const y = bs[i];
+    if (x === y) continue;
+    const xn = /^\d+$/.test(x);
+    const yn = /^\d+$/.test(y);
+    if (xn && yn) {
+      const d = Number(x) - Number(y);
+      if (d !== 0) return d < 0 ? -1 : 1;
+      // '007' and '7' are the same number — fall back to byte order so it is still stable.
+      return x < y ? -1 : 1;
+    }
+    if (xn !== yn) return xn ? -1 : 1; // numbers before words, consistently
+    return x < y ? -1 : 1;
+  }
+  if (as.length === bs.length) return 0;
+  return as.length < bs.length ? -1 : 1;
+}
+
+// ---------------------------------------------------------------------------
+// Values — canonical form, equality, and how far apart two of them are
+// ---------------------------------------------------------------------------
+
+/**
+ * @param {unknown} v
+ * @returns {v is Record<string, unknown>}
+ */
+function isPlainObject(v) {
+  return typeof v === 'object' && v !== null && !Array.isArray(v);
+}
+
+/**
+ * A canonical string for a value: object keys sorted, so two objects that say the same thing
+ * in a different key order are equal.
+ *
+ * Non-finite numbers are written out as text rather than left to `JSON.stringify`, which
+ * turns NaN and Infinity into `null` — and a NaN that reads as null is a real difference
+ * hidden by a serialiser, which is the one thing this tool must never do.
+ *
+ * @param {ObservedValue} value
+ * @returns {string}
+ */
+export function canonicalJson(value) {
+  return JSON.stringify(canonicalise(value, 0)) ?? 'null';
+}
+
+/**
+ * @param {ObservedValue} value
+ * @param {number} depth
+ * @returns {unknown}
+ */
+function canonicalise(value, depth) {
+  if (depth > MAX_VALUE_DEPTH) return '<too deep>';
+  if (typeof value === 'number' && !Number.isFinite(value)) return `<number:${String(value)}>`;
+  if (Array.isArray(value)) return value.map((v) => canonicalise(v, depth + 1));
+  if (isPlainObject(value)) {
+    /** @type {Record<string, unknown>} */
+    const out = {};
+    for (const key of Object.keys(value).sort()) {
+      out[key] = canonicalise(/** @type {ObservedValue} */ (value[key]), depth + 1);
+    }
+    return out;
+  }
+  return value;
+}
+
+/**
+ * @param {ObservedValue|undefined} a
+ * @param {ObservedValue|undefined} b
+ * @returns {boolean}
+ */
+export function sameValue(a, b) {
+  if (a === undefined || b === undefined) return a === b;
+  if (a === b) return true;
+  return canonicalJson(a) === canonicalJson(b);
+}
+
+/**
+ * Roughly how far apart two values are, 0 (identical) to 1 (nothing in common).
+ *
+ * READ THIS BEFORE USING IT: the number is for sorting a list and for writing a sentence a
+ * person can read. It is NOT a threshold and nothing in v2 compares it against one. Whether
+ * something differs is decided by equality; whether it counts is decided by measured wobble.
+ * Distance only decides what to show first.
+ *
+ * @param {ObservedValue|undefined} a
+ * @param {ObservedValue|undefined} b
+ * @param {number} [depth]
+ * @returns {number}
+ */
+export function valueDistance(a, b, depth = 0) {
+  if (a === undefined || b === undefined) return a === b ? 0 : 1;
+  if (sameValue(a, b)) return 0;
+  if (depth > 8) return 1;
+
+  if (typeof a === 'number' && typeof b === 'number') {
+    if (!Number.isFinite(a) || !Number.isFinite(b)) return 1;
+    const scale = Math.max(Math.abs(a), Math.abs(b), 1);
+    return clamp01(Math.abs(a - b) / scale);
+  }
+  if (typeof a === 'string' && typeof b === 'string') return stringDistance(a, b);
+  if (Array.isArray(a) && Array.isArray(b)) {
+    const n = Math.max(a.length, b.length);
+    if (n === 0) return 0;
+    let sum = 0;
+    for (let i = 0; i < n; i++) {
+      sum += i < a.length && i < b.length ? valueDistance(a[i], b[i], depth + 1) : 1;
+    }
+    return clamp01(sum / n);
+  }
+  if (isPlainObject(a) && isPlainObject(b)) {
+    const keys = new Set([...Object.keys(a), ...Object.keys(b)]);
+    if (keys.size === 0) return 0;
+    let sum = 0;
+    for (const k of keys) {
+      const av = /** @type {ObservedValue|undefined} */ (a[k]);
+      const bv = /** @type {ObservedValue|undefined} */ (b[k]);
+      sum += av === undefined || bv === undefined ? 1 : valueDistance(av, bv, depth + 1);
+    }
+    return clamp01(sum / keys.size);
+  }
+  // Different shapes entirely — a string where a number used to be. As far apart as it gets.
+  return 1;
+}
+
+/**
+ * How much of two strings is shared at their ends. Cheap on purpose: a real edit distance is
+ * quadratic and stdout observations run to megabytes.
+ * @param {string} a
+ * @param {string} b
+ * @returns {number}
+ */
+function stringDistance(a, b) {
+  const total = a.length + b.length;
+  if (total === 0) return 0;
+  let prefix = 0;
+  while (prefix < a.length && prefix < b.length && a[prefix] === b[prefix]) prefix++;
+  let suffix = 0;
+  while (
+    suffix < a.length - prefix &&
+    suffix < b.length - prefix &&
+    a[a.length - 1 - suffix] === b[b.length - 1 - suffix]
+  ) {
+    suffix++;
+  }
+  return clamp01(1 - (2 * (prefix + suffix)) / total);
+}
+
+/**
+ * @param {number} n
+ * @returns {number}
+ */
+function clamp01(n) {
+  if (!Number.isFinite(n)) return 1;
+  return n < 0 ? 0 : n > 1 ? 1 : n;
+}
+
+// ---------------------------------------------------------------------------
+// Making an observation
+// ---------------------------------------------------------------------------
+
+/**
+ * Check a value is something we can store and compare, and say plainly what is wrong if not.
+ * @param {unknown} value
+ * @param {string} where     Where inside the value we are, for the error message.
+ * @param {number} depth
+ * @param {Set<unknown>} seen
+ * @returns {string|null}
+ */
+function valueProblem(value, where, depth, seen) {
+  if (depth > MAX_VALUE_DEPTH) return `${where} nests deeper than ${MAX_VALUE_DEPTH} levels`;
+  if (value === null) return null;
+  const t = typeof value;
+  if (t === 'string' || t === 'number' || t === 'boolean') return null;
+  if (t === 'undefined') {
+    return `${where} is undefined — a fact we do not have is an absent path, not a path holding nothing`;
+  }
+  if (t === 'function' || t === 'symbol' || t === 'bigint') return `${where} is a ${t}, which cannot be stored or compared`;
+  if (value instanceof Date) return `${where} is a Date — write it out as a string first, and let the clock rule normalise it`;
+  if (Array.isArray(value) || isPlainObject(value)) {
+    if (seen.has(value)) return `${where} contains itself`;
+    seen.add(value);
+    /** @type {[string, unknown][]} */
+    const entries = Array.isArray(value)
+      ? value.map((v, i) => /** @type {[string, unknown]} */ ([`${where}[${i}]`, v]))
+      : Object.entries(value).map(([k, v]) => /** @type {[string, unknown]} */ ([`${where}.${k}`, v]));
+    for (const [childWhere, child] of entries) {
+      const problem = valueProblem(child, childWhere, depth + 1, seen);
+      if (problem) return problem;
+    }
+    seen.delete(value);
+    return null;
+  }
+  return `${where} is a ${Object.prototype.toString.call(value)}, which cannot be stored or compared`;
+}
+
+/**
+ * Make one observation, and refuse to make a broken one.
+ *
+ * Validation lives here rather than at the store, at the diff or in a report, because a bad
+ * path found later is a mystery and a bad path found here is a stack trace pointing straight
+ * at the collector that produced it.
+ *
+ * @param {string} path
+ * @param {Channel} channel
+ * @param {ObservedValue} value
+ * @param {ObservationMeta} [meta]
+ * @returns {Observation}
+ */
+export function makeObservation(path, channel, value, meta) {
+  const safePath = assertPath(path);
+  if (!isChannel(channel)) {
+    throw new StaysFixedError(`Unknown observation channel "${String(channel)}" for ${safePath}.`, {
+      hint: `The channels are: ${CHANNELS.join(', ')}.`,
+    });
+  }
+  const problem = valueProblem(value, 'the value', 0, new Set());
+  if (problem) {
+    throw new StaysFixedError(`Cannot observe ${safePath}: ${problem}.`, {
+      hint: 'Observations hold strings, numbers, booleans, null, and arrays or plain objects of those.',
+    });
+  }
+  /** @type {Observation} */
+  const observation = { path: safePath, channel, value };
+  if (meta && Object.keys(meta).length > 0) observation.meta = meta;
+  return observation;
+}
+
+/**
+ * Put a list of observations in the canonical order.
+ * @param {Observation[]} observations
+ * @returns {Observation[]}
+ */
+export function sortObservations(observations) {
+  return [...observations].sort((a, b) => comparePaths(a.path, b.path));
+}
+
+// ---------------------------------------------------------------------------
+// Indexing and diffing
+// ---------------------------------------------------------------------------
+
+/**
+ * @param {Capture|Observation[]} x
+ * @returns {Observation[]}
+ */
+function observationsOf(x) {
+  return Array.isArray(x) ? x : x.observations;
+}
+
+/**
+ * @param {Capture|Observation[]} x
+ * @returns {string|undefined}
+ */
+function journeyOf(x) {
+  return Array.isArray(x) ? undefined : x.journey;
+}
+
+/**
+ * Index observations by path.
+ *
+ * The FIRST observation at a path wins. Two facts at one address is a bug in whatever
+ * collected them — see `findDuplicatePaths` — and letting the last one win would hide it
+ * behind whatever order the collector happened to emit in.
+ *
+ * @param {Observation[]} observations
+ * @returns {Map<string, Observation>}
+ */
+export function indexByPath(observations) {
+  /** @type {Map<string, Observation>} */
+  const map = new Map();
+  for (const o of observations) {
+    if (!map.has(o.path)) map.set(o.path, o);
+  }
+  return map;
+}
+
+/**
+ * Paths a capture claimed twice with two different answers. A collector bug, always —
+ * reported rather than thrown so one bad address does not lose a whole run.
+ *
+ * @param {Observation[]} observations
+ * @returns {{path: string, values: ObservedValue[]}[]}
+ */
+export function findDuplicatePaths(observations) {
+  /** @type {Map<string, ObservedValue[]>} */
+  const seen = new Map();
+  for (const o of observations) {
+    const list = seen.get(o.path);
+    if (list) list.push(o.value);
+    else seen.set(o.path, [o.value]);
+  }
+  /** @type {{path: string, values: ObservedValue[]}[]} */
+  const out = [];
+  for (const [path, values] of seen) {
+    if (values.length < 2) continue;
+    /** @type {ObservedValue[]} */
+    const distinct = [];
+    for (const v of values) {
+      if (!distinct.some((d) => sameValue(d, v))) distinct.push(v);
+    }
+    if (distinct.length > 1) out.push({ path, values: distinct });
+  }
+  return out.sort((a, b) => comparePaths(a.path, b.path));
+}
+
+/**
+ * Compare a reference capture against a candidate capture.
+ *
+ * Three kinds come out, and the last two are what this tool exists for:
+ *   changed   — the same address now answers differently
+ *   appeared  — an address that did not exist before
+ *   vanished  — an address that has stopped existing. A door that closed. No screenshot
+ *               comparison has ever noticed one of these.
+ *
+ * @param {Capture|Observation[]} reference
+ * @param {Capture|Observation[]} candidate
+ * @returns {Difference[]}
+ */
+export function diffCaptures(reference, candidate) {
+  const ref = indexByPath(observationsOf(reference));
+  const cand = indexByPath(observationsOf(candidate));
+  const journey = journeyOf(candidate) ?? journeyOf(reference);
+
+  /** @type {Difference[]} */
+  const out = [];
+
+  for (const [path, r] of ref) {
+    const c = cand.get(path);
+    if (!c) {
+      out.push(difference(path, r.channel, 'vanished', r.value, undefined, journey, r, undefined));
+      continue;
+    }
+    if (!sameValue(r.value, c.value)) {
+      out.push(difference(path, c.channel, 'changed', r.value, c.value, journey, r, c));
+    }
+  }
+
+  for (const [path, c] of cand) {
+    if (!ref.has(path)) {
+      out.push(difference(path, c.channel, 'appeared', undefined, c.value, journey, undefined, c));
+    }
+  }
+
+  return out.sort((a, b) => comparePaths(a.path, b.path));
+}
+
+/**
+ * @param {string} path
+ * @param {Channel} channel
+ * @param {DifferenceKind} kind
+ * @param {ObservedValue|undefined} referenceValue
+ * @param {ObservedValue|undefined} candidateValue
+ * @param {string|undefined} journey
+ * @param {Observation|undefined} refObs
+ * @param {Observation|undefined} candObs
+ * @returns {Difference}
+ */
+function difference(path, channel, kind, referenceValue, candidateValue, journey, refObs, candObs) {
+  /** @type {Difference} */
+  const d = {
+    path,
+    channel,
+    kind,
+    distance: valueDistance(referenceValue, candidateValue),
+  };
+  if (referenceValue !== undefined) d.reference = referenceValue;
+  if (candidateValue !== undefined) d.candidate = candidateValue;
+  if (journey) d.journey = journey;
+
+  const describe = candObs?.meta?.describe ?? refObs?.meta?.describe;
+  if (describe) d.describe = describe;
+  const evidence = candObs?.meta?.evidence ?? refObs?.meta?.evidence;
+  if (evidence) d.evidence = evidence;
+
+  // One address arriving on two different channels is worth saying out loud: two collectors
+  // are both claiming it, and one of them is wrong about what it is looking at.
+  if (refObs && candObs && refObs.channel !== candObs.channel) {
+    const prefix = d.describe ? d.describe + ' ' : '';
+    d.describe = `${prefix}(this address was observed as ${refObs.channel} before and ${candObs.channel} now — two collectors are claiming it)`;
+  }
+  return d;
+}
+
+// ---------------------------------------------------------------------------
+// Wobble — the product arguing with itself
+// ---------------------------------------------------------------------------
+
+/**
+ * Measure what a build disagrees with itself about, by comparing two runs of the SAME build.
+ *
+ * This replaces every tolerance setting the old tool had. A tolerance is a guess about how
+ * much noise a product makes; this is the measurement. Two runs, same bytes, same machine,
+ * minutes apart — anything that differs was not caused by anybody's change.
+ *
+ * @param {Capture|Observation[]} runA
+ * @param {Capture|Observation[]} runB
+ * @returns {Wobble}
+ */
+export function measureWobble(runA, runB) {
+  const a = Array.isArray(runA) ? undefined : runA;
+  const b = Array.isArray(runB) ? undefined : runB;
+
+  if (a && b && a.build.id !== b.build.id) {
+    throw new StaysFixedError(
+      `Wobble has to be measured on one build, but these captures are of different builds (${a.build.id} and ${b.build.id}).`,
+      { hint: 'Run the same build twice. Comparing two different builds gives a difference, not a wobble.' },
+    );
+  }
+  if (a && b && a.journey !== b.journey) {
+    throw new StaysFixedError(
+      `Wobble has to be measured on one journey, but these captures walked "${a.journey}" and "${b.journey}".`,
+    );
+  }
+
+  const indexA = indexByPath(observationsOf(runA));
+  const indexB = indexByPath(observationsOf(runB));
+
+  /** @type {WobbleEntry[]} */
+  const entries = [];
+  let steady = 0;
+
+  for (const [path, oa] of indexA) {
+    const ob = indexB.get(path);
+    if (!ob) {
+      entries.push({ path, channel: oa.channel, kind: 'vanished', a: oa.value, distance: 1 });
+      continue;
+    }
+    if (sameValue(oa.value, ob.value)) {
+      steady++;
+      continue;
+    }
+    entries.push({
+      path,
+      channel: ob.channel,
+      kind: 'changed',
+      a: oa.value,
+      b: ob.value,
+      distance: valueDistance(oa.value, ob.value),
+    });
+  }
+
+  for (const [path, ob] of indexB) {
+    if (!indexA.has(path)) {
+      entries.push({ path, channel: ob.channel, kind: 'appeared', b: ob.value, distance: 1 });
+    }
+  }
+
+  entries.sort((x, y) => comparePaths(x.path, y.path));
+
+  return {
+    buildId: a?.build.id ?? b?.build.id ?? '',
+    journey: a?.journey ?? b?.journey ?? '',
+    runs: [a?.id ?? 'a', b?.id ?? 'b'],
+    entries,
+    unstable: entries.map((e) => e.path),
+    steady,
+    measured: true,
+  };
+}
+
+/**
+ * A wobble record for the case where the build was only run once.
+ *
+ * It subtracts nothing, and it exists so the rest of the pipeline never has to ask whether it
+ * has a measurement — it asks `measured`, and a run that could not measure says so in its
+ * summary instead of pretending its list is clean.
+ *
+ * @param {string} buildId
+ * @param {string} journey
+ * @returns {Wobble}
+ */
+export function unmeasuredWobble(buildId, journey) {
+  return { buildId, journey, runs: ['', ''], entries: [], unstable: [], steady: 0, measured: false };
+}
+
+/**
+ * Fold several journeys' wobble into one record, so a whole-product run subtracts in one go.
+ *
+ * One journey that could not be measured twice makes the whole record unmeasured, because the
+ * alternative is a summary claiming a clean subtraction over a list that is partly raw.
+ *
+ * @param {Wobble[]} wobbles
+ * @returns {Wobble}
+ */
+export function mergeWobble(wobbles) {
+  if (wobbles.length === 1) return wobbles[0];
+  /** @type {WobbleEntry[]} */
+  const entries = [];
+  const seen = new Set();
+  let steady = 0;
+  let measured = wobbles.length > 0;
+  for (const w of wobbles) {
+    if (!w.measured) measured = false;
+    steady += w.steady;
+    for (const e of w.entries) {
+      if (seen.has(e.path)) continue;
+      seen.add(e.path);
+      entries.push(e);
+    }
+  }
+  entries.sort((a, b) => comparePaths(a.path, b.path));
+  return {
+    buildId: wobbles[0]?.buildId ?? '',
+    journey: '*',
+    runs: [wobbles[0]?.runs[0] ?? '', wobbles[0]?.runs[1] ?? ''],
+    entries,
+    unstable: entries.map((e) => e.path),
+    steady,
+    measured,
+  };
+}
+
+/**
+ * When a wobble measurement stops being a measurement.
+ *
+ * Subtraction is set subtraction: a difference at an address the build cannot answer the same
+ * way twice is dropped. That is right, and it has one failure shape, which is the worst shape
+ * this tool has. If the second run of the new build FALLS OVER — the app crashed half way, a
+ * port was taken, a device went to sleep, a first run wrote a cache the second one read — then
+ * most of the addresses the first run answered are missing from the second, every one of them
+ * is filed as unsteady, every real difference at them is subtracted, and the run ends
+ * "nothing that already worked has changed". Confident, clean, and about nothing.
+ *
+ * So the share is looked at. A product that disagrees with itself about a handful of addresses
+ * is normal — a timestamp, an id, a port. A product that disagrees with itself about MOST of
+ * them did not wobble; something went wrong with the run. This is not a tolerance: no number
+ * here decides whether any difference is real. It decides one thing only — whether this run is
+ * entitled to say the word "clean".
+ *
+ * @param {Wobble} wobble
+ * @returns {{stormy: boolean, share: number, looked: number, vanished: number, why: string}}
+ */
+export function wobbleStorm(wobble) {
+  const unstable = wobble.unstable.length;
+  const looked = unstable + wobble.steady;
+  const vanished = wobble.entries.filter((e) => e.kind === 'vanished').length;
+  const share = looked === 0 ? 0 : unstable / looked;
+  if (!wobble.measured || looked < STORM_FLOOR || share <= STORM_SHARE) {
+    return { stormy: false, share, looked, vanished, why: '' };
+  }
+  const percent = Math.round(share * 100);
+  const why =
+    `The new build was run twice and gave a different answer at ${unstable} of the ${looked} addresses it was asked about — ${percent}% of them` +
+    (vanished > 0 ? `, and ${vanished} address${vanished === 1 ? '' : 'es'} the first run answered were missing from the second altogether` : '') +
+    '. That is not a product wobbling; that is a run that went wrong. Everything it disagreed with itself about is dropped before anything is compared, so on this run the comparison covered almost nothing. This is not a pass and not a failure — there is no answer here. Run it again on a quiet machine, and if it happens twice, something in the product or its setup does not survive being started a second time.';
+  return { stormy: true, share, looked, vanished, why };
+}
+
+/**
+ * Subtract the measured noise from the differences.
+ *
+ * The rule is set subtraction and nothing cleverer: if a path will not sit still between two
+ * runs of the same build, a difference at that path proves nothing, whatever its size. Any
+ * "but it changed by MORE than the wobble did" rule is a tolerance wearing a disguise, and
+ * tolerances are how tools like this die — too loose to catch the real thing, too tight to
+ * leave switched on.
+ *
+ * The third result is the one no other tool produces. A path that was steady in the reference
+ * and wobbles now is a finding in itself: the change made something unpredictable. Nothing is
+ * "wrong" at that address and it still needs fixing.
+ *
+ * @param {Difference[]} differences
+ * @param {Wobble} wobble                                Measured on the candidate build.
+ * @param {{referenceWobble?: Wobble, steadyInReference?: string[]}} [opts]
+ * @returns {WobbleSubtraction}
+ */
+export function subtractWobble(differences, wobble, opts = {}) {
+  const unstableNow = new Set(wobble.unstable);
+  // NOT symmetric, and that is deliberate. Subtracting the OLD build's wobble as well was
+  // tried on 2026-08-30 and taken straight back out: a path the old build answered randomly
+  // and the new build now answers the same way every time is a REAL change — somebody made
+  // something deterministic, or hard-coded what used to vary — and subtracting the old
+  // build's wobble is exactly what would hide it. Where both builds wobble at a path, the
+  // candidate's own wobble already covers it, so nothing is lost by leaving this alone.
+
+  /** @type {Difference[]} */
+  const real = [];
+  /** @type {Difference[]} */
+  const noise = [];
+
+  for (const d of differences) {
+    const wobbling = unstableNow.has(d.path);
+    // Copy rather than mutate: the caller's list is often the stored diff, and a flag written
+    // into it becomes a fact nobody can trace back to whoever decided it.
+    const flagged = { ...d, real: !wobbling, wobbling };
+    if (wobbling) noise.push(flagged);
+    else real.push(flagged);
+  }
+
+  const referenceWobble = opts.referenceWobble;
+  const steadyBefore = opts.steadyInReference ? new Set(opts.steadyInReference) : null;
+  const couldTell = Boolean((referenceWobble && referenceWobble.measured) || steadyBefore);
+
+  /** @type {WobbleEntry[]} */
+  let newlyUnstable = [];
+  if (couldTell) {
+    const unstableBefore = new Set(referenceWobble?.unstable ?? []);
+    newlyUnstable = wobble.entries.filter((e) => {
+      if (unstableBefore.has(e.path)) return false;
+      // With an explicit steady list we only claim the paths it names. Without one, anything
+      // the reference did not record as unstable counts.
+      return steadyBefore ? steadyBefore.has(e.path) : true;
+    });
+  }
+
+  const storm = wobbleStorm(wobble);
+  /** @type {WobbleSubtraction} */
+  const out = {
+    real,
+    noise,
+    newlyUnstable,
+    couldTellNewlyUnstable: couldTell,
+    note: subtractionNote(wobble, couldTell, real.length, noise.length, newlyUnstable.length),
+  };
+  if (storm.stormy) {
+    out.couldNotTell = true;
+    out.couldNotTellWhy = storm.why;
+    out.note = `${storm.why} ${out.note}`;
+  }
+  return out;
+}
+
+/**
+ * The sentence that goes in the summary. Written here so every caller says the same honest
+ * thing rather than inventing its own wording.
+ *
+ * @param {Wobble} wobble
+ * @param {boolean} couldTell
+ * @param {number} realCount
+ * @param {number} noiseCount
+ * @param {number} newlyUnstableCount
+ * @returns {string}
+ */
+function subtractionNote(wobble, couldTell, realCount, noiseCount, newlyUnstableCount) {
+  if (!wobble.measured) {
+    return `The new build was only run once, so none of its own noise has been subtracted. All ${realCount} difference${realCount === 1 ? '' : 's'} here may include things that change on every run. Run it twice for a clean list.`;
+  }
+  const parts = [
+    `Running the new build twice showed ${wobble.unstable.length} address${wobble.unstable.length === 1 ? '' : 'es'} that will not sit still, and ${wobble.steady} that ${wobble.steady === 1 ? 'does' : 'do'}.`,
+    `${noiseCount} difference${noiseCount === 1 ? '' : 's'} landed on the unsteady ones and ${noiseCount === 1 ? 'was' : 'were'} dropped; ${realCount} remain${realCount === 1 ? 's' : ''}.`,
+  ];
+  if (!couldTell) {
+    parts.push('There is no record of how steady the old build was, so nothing can be reported as newly unpredictable.');
+  } else if (newlyUnstableCount > 0) {
+    parts.push(`${newlyUnstableCount} address${newlyUnstableCount === 1 ? ' was' : 'es were'} steady before and unpredictable now — the change made something non-deterministic.`);
+  } else {
+    parts.push('Nothing that used to be steady has become unpredictable.');
+  }
+  return parts.join(' ');
+}