@worca/app 1.2.0 → 1.3.0-rc.2

package/src/shared/team-metrics/aggregate.mjs

@@ -0,0 +1,341 @@
+// src/shared/team-metrics/aggregate.mjs
+// Team metrics aggregation (team-metrics-design.md §4.9). PURE: no I/O, no node: imports —
+// shared by the server (read.mjs) and the browser (the page re-aggregates per range/group/filter).
+
+export const SUPPORTED_RECORD_VERSION = 1;
+export const RANGES = Object.freeze(['this-month', 'last-month', 'quarter', 'year', 'all', 'custom']);
+// Only dimensions where a run has exactly ONE key: a run's cost is one number for the whole
+// pipeline, so it is never split across the projects it touched (`project` stays a breakdown and
+// a filter — "runs that touched X" — but not a stack).
+export const GROUP_BYS = Object.freeze(['workflow', 'result', 'actor']);
+export const FILTER_DIMS = Object.freeze(['workflow', 'source', 'actor', 'project', 'models', 'result']);
+const DAY = 86_400_000;
+const WEEK = 7 * DAY;
+
+// UTC ISO-8601 only (what record.mjs writes). Anything else — "2026-09-01 00:30" (parsed in LOCAL time,
+// so server and browser would bucket it differently), "+200000-01-01…" (expanded years) — is malformed.
+const ISO_UTC_RE = /^\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}(\.\d{1,9})?Z$/;
+
+/** @returns {{record:object}|{malformed:true}|{unknownV:true}} */
+export function parseRecordLine(line) {
+  let v;
+  try { v = JSON.parse(line); } catch { return { malformed: true }; }
+  if (!v || typeof v !== 'object' || Array.isArray(v) || !Number.isInteger(v.v)) return { malformed: true };
+  if (v.v > SUPPORTED_RECORD_VERSION) return { unknownV: true };
+  if (v.v < 1 || typeof v.id !== 'string' || typeof v.startedAt !== 'string'
+    || !ISO_UTC_RE.test(v.startedAt) || !Number.isFinite(Date.parse(v.startedAt))) return { malformed: true };
+  return { record: v };
+}
+
+const MAX_CUSTOM_DAYS = 3660; // ~10 years: bounds the weekly series
+const MAX_WEEKS = Math.ceil(MAX_CUSTOM_DAYS / 7) + 2;
+
+function parseDay(s) {
+  if (typeof s !== 'string' || !/^\d{4}-\d{2}-\d{2}$/.test(s)) return null;
+  const ms = Date.parse(`${s}T00:00:00Z`);
+  // Reject roll-overs such as 2026-02-31 (Date.parse would silently give Mar 3).
+  return Number.isFinite(ms) && new Date(ms).toISOString().startsWith(s) ? ms : null;
+}
+
+/**
+ * UTC window + the previous window for deltas. startMs inclusive, endMs exclusive.
+ * Deltas are like-for-like: while `now` is inside the current window (this month / quarter /
+ * year so far), the previous window is cut to the same elapsed length, so "vs prev. month" on
+ * Sep 16 compares Sep 1–16 with Aug 1–16, not with all of August.
+ */
+export function resolveRange(range = 'this-month', { now = Date.now(), from = null, to = null } = {}) {
+  const d = new Date(now);
+  const y = d.getUTCFullYear();
+  const m = d.getUTCMonth();
+  const win = (startMs, endMs, prevStartMs) => {
+    // `>=`, not `>`: at exactly 00:00 on the 1st the window is open with 0 elapsed, so the previous
+    // window must be 0 long too. With `>` it fell through to the closed-window branch and compared
+    // "nothing so far this month" against ALL of last month for one millisecond.
+    const elapsed = now >= startMs && now < endMs ? now - startMs : startMs - prevStartMs;
+    return { range, startMs, endMs, prevStartMs, prevEndMs: Math.min(startMs, prevStartMs + elapsed) };
+  };
+  switch (range) {
+    case 'this-month': return win(Date.UTC(y, m, 1), Date.UTC(y, m + 1, 1), Date.UTC(y, m - 1, 1));
+    case 'last-month': return win(Date.UTC(y, m - 1, 1), Date.UTC(y, m, 1), Date.UTC(y, m - 2, 1));
+    case 'quarter': { const q = Math.floor(m / 3) * 3; return win(Date.UTC(y, q, 1), Date.UTC(y, q + 3, 1), Date.UTC(y, q - 3, 1)); }
+    case 'year': return win(Date.UTC(y, 0, 1), Date.UTC(y + 1, 0, 1), Date.UTC(y - 1, 0, 1));
+    case 'all': return { range, startMs: null, endMs: null, prevStartMs: null, prevEndMs: null };
+    case 'custom': {
+      const s = parseDay(from);
+      const e = parseDay(to);
+      if (s == null || e == null || e < s) throw new RangeError('custom range needs from and to as valid YYYY-MM-DD dates with from <= to');
+      if ((e - s) / DAY > MAX_CUSTOM_DAYS) throw new RangeError(`custom range is limited to ${MAX_CUSTOM_DAYS} days`);
+      const endMs = e + DAY;
+      return { range, startMs: s, endMs, prevStartMs: s - (endMs - s), prevEndMs: s };
+    }
+    default: throw new RangeError(`unknown range "${range}" (expected ${RANGES.join(' | ')})`);
+  }
+}
+
+/** Monday 00:00 UTC of the week containing ms. */
+export function weekStartMs(ms) {
+  const d = new Date(ms);
+  return Date.UTC(d.getUTCFullYear(), d.getUTCMonth(), d.getUTCDate() - ((d.getUTCDay() + 6) % 7));
+}
+
+const startOf = (r) => Date.parse(r.startedAt);
+const usdOf = (r) => (Number.isFinite(r?.cost?.usd) ? r.cost.usd : 0);
+const isNum = (v) => typeof v === 'number' && Number.isFinite(v);
+const sum = (a) => a.reduce((s, x) => s + x, 0);
+const round2 = (v) => Math.round(v * 100) / 100;
+const round1 = (v) => Math.round(v * 10) / 10;
+
+function percentile(values, p) {
+  const xs = values.filter(isNum).sort((a, b) => a - b);
+  if (!xs.length) return null;
+  return xs[Math.min(xs.length - 1, Math.max(0, Math.ceil(p * xs.length) - 1))]; // nearest rank
+}
+
+const NONE = '__none__';
+
+/** Dimension keys for filtering/breakdowns. Each returns [{key,label,sub?}] (a run may touch several projects). */
+export const DIMENSIONS = {
+  workflow: (r) => [{ key: r.workflow?.id || r.workflow?.name || NONE, label: r.workflow?.name || r.workflow?.id || 'Unknown workflow' }],
+  // A hand-written or future record may carry no `result`: without the guard it became the stack
+  // key "undefined" with an undefined label.
+  result: (r) => [{ key: typeof r.result === 'string' && r.result ? r.result : NONE, label: typeof r.result === 'string' && r.result ? r.result : 'Unknown result' }],
+  actor: (r) => [{ key: r.actor || NONE, label: r.actor || 'No actor' }],
+  source: (r) => (r.source
+    ? [{ key: `${r.source.type}:${r.source.ref ?? r.source.url ?? ''}`, label: [r.source.ref, r.source.title].filter(Boolean).join(' ') || r.source.url || r.source.type, sub: r.source.type }]
+    : [{ key: NONE, label: 'No ticket', sub: 'ad-hoc prompts' }]),
+  models: (r) => {
+    const ms = Array.isArray(r.agents?.models) ? [...r.agents.models].sort() : [];
+    return [{ key: ms.join(' + ') || NONE, label: ms.join(' + ') || 'Unknown models' }];
+  },
+  project: (r) => {
+    if (r.target?.kind === 'workspace') {
+      const t = Array.isArray(r.target.touched) ? r.target.touched : [];
+      // `files`: this member's own changed-file count (target.touchedFiles, additive); null on a
+      // record that predates it — the run's total is NOT this project's, so the row says "–".
+      const tf = r.target.touchedFiles && typeof r.target.touchedFiles === 'object' ? r.target.touchedFiles : null;
+      return t.length
+        ? t.map((p) => ({ key: p, label: p, files: tf && Number.isInteger(tf[p]) ? tf[p] : null }))
+        : [{ key: NONE, label: 'No project touched', files: null }];
+    }
+    return [{ key: r.target?.project || NONE, label: r.target?.project || 'Unknown project' }];
+  },
+};
+
+export function applyFilter(records, filter = {}) {
+  const active = Object.entries(filter || {}).filter(([dim, key]) => FILTER_DIMS.includes(dim) && key != null && key !== '');
+  if (!active.length) return records;
+  return records.filter((r) => active.every(([dim, key]) => DIMENSIONS[dim](r).some((k) => k.key === key)));
+}
+
+function computeKpis(rs, now) {
+  const n = rs.length;
+  const count = (res) => rs.filter((r) => r.result === res).length;
+  const usd = sum(rs.map(usdOf));
+  const paired = rs.filter((r) => isNum(r.wallMs) && isNum(r.activeMs));
+  const wallSum = sum(paired.map((r) => r.wallMs));
+  const withPr = rs.filter((r) => r.pr && (r.pr.url || r.pr.number != null));
+  const reviews = rs.map((r) => r.cycles?.review).filter(isNum);
+  const d = new Date(now);
+  const monthStart = Date.UTC(d.getUTCFullYear(), d.getUTCMonth(), 1);
+  return {
+    spendUsd: round2(usd),
+    spendThisMonthUsd: round2(sum(rs.filter((r) => startOf(r) >= monthStart).map(usdOf))),
+    runs: n,
+    done: count('done'), failed: count('failed'), stopped: count('stopped'),
+    successRate: n ? count('done') / n : null,
+    costPerRunUsd: n ? round2(usd / n) : null,
+    costPerRunMedianUsd: percentile(rs.map(usdOf), 0.5),
+    costPerRunP90Usd: percentile(rs.map(usdOf), 0.9),
+    runsWithPr: withPr.length,
+    costPerRunWithPrUsd: withPr.length ? round2(sum(withPr.map(usdOf)) / withPr.length) : null,
+    durationMedianMs: percentile(rs.map((r) => r.wallMs), 0.5),
+    machineMs: sum(rs.map((r) => r.activeMs).filter(isNum)),
+    autonomy: wallSum > 0 ? sum(paired.map((r) => r.activeMs)) / wallSum : null,
+    interventionsPerRun: n ? round1(sum(rs.map((r) => (r.interventions?.questions | 0) + (r.interventions?.pauses | 0))) / n) : null,
+    reviewCyclesMean: reviews.length ? round1(sum(reviews) / reviews.length) : null,
+    convergeInOneRate: reviews.length ? reviews.filter((c) => c <= 1).length / reviews.length : null,
+    filesChanged: sum(rs.map((r) => r.git?.filesChanged).filter(isNum)),
+  };
+}
+
+const pct = (cur, prev) => (isNum(cur) && isNum(prev) && prev > 0 ? (cur - prev) / prev : null);
+
+function deltas(k, p) {
+  return {
+    spendPct: pct(k.spendUsd, p.spendUsd),
+    runsPct: pct(k.runs, p.runs),
+    costPerRunPct: pct(k.costPerRunUsd, p.costPerRunUsd),
+    durationPct: pct(k.durationMedianMs, p.durationMedianMs),
+    autonomyPts: isNum(k.autonomy) && isNum(p.autonomy) ? Math.round((k.autonomy - p.autonomy) * 100) : null,
+    reviewCycles: isNum(k.reviewCyclesMean) && isNum(p.reviewCyclesMean) ? round1(k.reviewCyclesMean - p.reviewCyclesMean) : null,
+  };
+}
+
+function breakdown(rs, dim) {
+  const rows = new Map();
+  for (const r of rs) {
+    for (const k of DIMENSIONS[dim](r)) {
+      const row = rows.get(k.key) || { key: k.key, label: k.label, sub: k.sub ?? null, runs: 0, usd: 0, done: 0, reviews: [], filesChanged: 0, filesUnknown: 0 };
+      row.runs += 1;
+      row.usd += usdOf(r);                       // "a run touching two projects counts in both"
+      if (r.result === 'done') row.done += 1;
+      if (isNum(r.cycles?.review)) row.reviews.push(r.cycles.review);
+      // A key that carries its own count (a touched member) uses it; every other dimension is
+      // whole-run, so the run's total applies. A member without a count is counted as unknown.
+      const files = 'files' in k ? k.files : (isNum(r.git?.filesChanged) ? r.git.filesChanged : null);
+      if (isNum(files)) row.filesChanged += files; else row.filesUnknown += 1;
+      rows.set(k.key, row);
+    }
+  }
+  // `share` is normalised against THIS column's own sum, not against the run total. For every
+  // single-key dimension the two are equal; for `project` on a workspace scope a run touching two
+  // projects deliberately counts its full spend in BOTH rows (decision 19, the mockup subtitle),
+  // so dividing by the run total made the bars sum to 200%. The table answers "what did the runs
+  // that touched this project cost"; no chart splits a run's cost across projects (GROUP_BYS).
+  const colTotal = [...rows.values()].reduce((a, r) => a + r.usd, 0);
+  // `runShare`: the fraction of the runs in range that touched this key — the per-project bar
+  // (spend is not a per-project fact, so the project table shows this instead of a spend share).
+  const runTotal = rs.length;
+  return [...rows.values()]
+    .map((row) => ({
+      key: row.key, label: row.label, sub: row.sub, runs: row.runs, usd: round2(row.usd),
+      runShare: runTotal > 0 ? row.runs / runTotal : 0,
+      perRunUsd: row.runs ? round2(row.usd / row.runs) : null,
+      successRate: row.runs ? row.done / row.runs : null,
+      cyclesMean: row.reviews.length ? round1(sum(row.reviews) / row.reviews.length) : null,
+      share: colTotal > 0 ? row.usd / colTotal : 0,
+      // null when no run in the row carries a count: the cell shows "–", never a wrong number.
+      filesChanged: row.filesUnknown === row.runs ? null : row.filesChanged,
+      filesUnknown: row.filesUnknown,
+      isNone: row.key === NONE,
+    }))
+    // Projects order by what the table shows (runs touched, then files); every other dimension by spend.
+    .sort((a, b) => (a.isNone - b.isNone) || (dim === 'project'
+      ? (b.runs - a.runs) || ((b.filesChanged ?? -1) - (a.filesChanged ?? -1)) || a.label.localeCompare(b.label)
+      : (b.usd - a.usd) || (b.runs - a.runs)));
+}
+
+function weekSpan(rs, win, now) {
+  // reduce, not Math.min(...spread): a spread over ~100k+ records overflows the call stack.
+  let minStart = Infinity;
+  let maxStart = -Infinity;
+  for (const r of rs) { const t = startOf(r); if (t < minStart) minStart = t; if (t > maxStart) maxStart = t; }
+  const first = win.startMs ?? (rs.length ? minStart : null);
+  if (first == null) return [];
+  // Series stop at `now`, but never before the newest record in range (a teammate's clock may run
+  // ahead) — by at most 31 days: records come from a branch anyone with push access can write,
+  // and one record dated year 20000 must not produce ~900k weeks (a 100 MB response / Map overflow).
+  const cap = Math.max(now, rs.length ? Math.min(maxStart, now + 31 * DAY) : -Infinity);
+  const last = Math.min(win.endMs != null ? win.endMs - 1 : cap, cap);
+  const to = weekStartMs(Math.max(first, last));
+  const from = Math.max(weekStartMs(first), to - (MAX_WEEKS - 1) * WEEK); // at most ~10 years of columns
+  const weeks = [];
+  for (let w = from; w <= to; w += WEEK) weeks.push(w);
+  return weeks;
+}
+
+function series(rs, win, groupBy, now) {
+  const weeks = weekSpan(rs, win, now);
+  const idx = new Map(weeks.map((w, i) => [w, i]));
+  // Null-prototype maps: keys are record-supplied strings ("constructor", "__proto__" must not collide).
+  const spend = weeks.map((w) => ({ weekStartMs: w, totalUsd: 0, stacks: Object.create(null) }));
+  const runs = weeks.map((w) => ({ weekStartMs: w, done: 0, failed: 0, stopped: 0 }));
+  const labels = Object.create(null);
+  const totals = Object.create(null);
+  for (const r of rs) {
+    const i = idx.get(weekStartMs(startOf(r)));
+    if (i == null) continue;
+    if (r.result === 'done' || r.result === 'failed' || r.result === 'stopped') runs[i][r.result] += 1;
+    spend[i].totalUsd += usdOf(r);
+    for (const k of DIMENSIONS[groupBy](r)) {
+      const v = usdOf(r);                          // one key per run for every GROUP_BYS dimension
+      spend[i].stacks[k.key] = (spend[i].stacks[k.key] || 0) + v;
+      totals[k.key] = (totals[k.key] || 0) + v;
+      labels[k.key] = k.label;
+    }
+  }
+  for (const pt of spend) {
+    pt.totalUsd = round2(pt.totalUsd);
+    // Back to a plain object (Object.fromEntries defines own data properties, so "__proto__" survives)
+    // — callers and deepStrictEqual expect ordinary objects.
+    pt.stacks = Object.fromEntries(Object.entries(pt.stacks).map(([key, v]) => [key, round2(v)]));
+  }
+  const stackKeys = Object.keys(totals).sort((a, b) => totals[b] - totals[a])
+    .map((key) => ({ key, label: labels[key], totalUsd: round2(totals[key]) }));
+  return { spend, runs, stackKeys };
+}
+
+/** Records come from a branch anyone with push access can write: only http(s) links survive. */
+export function safeHttpUrl(u) {
+  if (typeof u !== 'string' || !u) return null;
+  try { const x = new URL(u); return x.protocol === 'https:' || x.protocol === 'http:' ? x.href : null; } catch { return null; }
+}
+
+function toRunRow(r) {
+  return {
+    id: r.id, title: r.title ?? '(untitled)', startedAt: r.startedAt,
+    workflow: r.workflow?.name ?? r.workflow?.id ?? null, result: r.result,
+    usd: usdOf(r), wallMs: isNum(r.wallMs) ? r.wallMs : null, activeMs: isNum(r.activeMs) ? r.activeMs : null,
+    reviewCycles: isNum(r.cycles?.review) ? r.cycles.review : null,
+    pr: r.pr && (r.pr.url || r.pr.number != null) ? { number: Number.isInteger(r.pr.number) ? r.pr.number : null, url: safeHttpUrl(r.pr.url) } : null,
+    actor: r.actor ?? null,
+    source: r.source ? [r.source.ref, r.source.title].filter(Boolean).join(' ') : null,
+    projects: r.target?.kind === 'workspace'
+      ? (Array.isArray(r.target.touched) ? r.target.touched.filter((p) => typeof p === 'string') : [])
+      : [r.target?.project].filter((p) => typeof p === 'string' && p),
+  };
+}
+
+/**
+ * @param {object[]} records  v1 records (already parsed; unknown v / malformed already dropped)
+ * @param {{range?:string, from?:string|null, to?:string|null, groupBy?:string, filter?:object, now?:number}} [opts]
+ */
+export function aggregate(records, { range = 'this-month', from = null, to = null, groupBy = 'workflow', filter = {}, now = Date.now() } = {}) {
+  if (!GROUP_BYS.includes(groupBy)) throw new RangeError(`unknown groupBy "${groupBy}" (expected ${GROUP_BYS.join(' | ')})`);
+  const win = resolveRange(range, { now, from, to });
+  const usable = (records || []).filter((r) => r && Number.isFinite(startOf(r)));
+  const filtered = applyFilter(usable, filter);
+  const inWin = (s, e) => (r) => (s == null || startOf(r) >= s) && (e == null || startOf(r) < e);
+  const inRange = filtered.filter(inWin(win.startMs, win.endMs));
+  const kpis = computeKpis(inRange, now);
+  const prevKpis = win.prevStartMs == null ? null : computeKpis(filtered.filter(inWin(win.prevStartMs, win.prevEndMs)), now);
+  const hasActor = inRange.some((r) => r.actor);
+  const hasWorkspace = inRange.some((r) => r.target?.kind === 'workspace');
+  return {
+    range: win, groupBy, filter: { ...filter },
+    kpis, prev: prevKpis, deltas: prevKpis ? deltas(kpis, prevKpis) : null,
+    series: series(inRange, win, groupBy, now),
+    breakdowns: {
+      workflow: breakdown(inRange, 'workflow'),
+      source: breakdown(inRange, 'source'),
+      actor: hasActor ? breakdown(inRange, 'actor') : null,
+      project: hasWorkspace ? breakdown(inRange, 'project') : null,
+      models: breakdown(inRange, 'models'),
+    },
+    runs: [...inRange].sort((a, b) => startOf(b) - startOf(a)).map(toRunRow),
+    totalRecords: usable.length,
+  };
+}
+
+export const CSV_COLUMNS = Object.freeze(['startedAt', 'title', 'workflow', 'result', 'costUsd', 'wallMs', 'activeMs', 'reviewCycles', 'prNumber', 'prUrl', 'actor', 'source', 'projects', 'id']);
+
+function csvCell(v) {
+  if (v == null) return '';
+  let s = String(v);
+  if (/^[=+\-@\t\r]/.test(s)) s = `'${s}`;            // spreadsheet formula injection guard
+  return /[",\r\n]/.test(s) ? `"${s.replace(/"/g, '""')}"` : s;
+}
+
+/** CSV of aggregate().runs rows (the filtered run table). */
+export function toCsv(runRows) {
+  const lines = [CSV_COLUMNS.join(',')];
+  for (const r of runRows) {
+    lines.push([
+      r.startedAt, r.title, r.workflow, r.result, r.usd, r.wallMs, r.activeMs, r.reviewCycles,
+      r.pr?.number, r.pr?.url, r.actor, r.source, (r.projects || []).join(' '), r.id,
+    ].map(csvCell).join(','));
+  }
+  // BOM: Excel otherwise reads the file as the local code page and mangles every non-ASCII
+  // actor/title ("Siniša Đukić"). Sheets and LibreOffice ignore it.
+  return '' + lines.join('\r\n') + '\r\n';
+}

package/src/shared/team-metrics/workspace-match.mjs

@@ -0,0 +1,13 @@
+// src/shared/team-metrics/workspace-match.mjs
+// Does a workspace run record belong to workspace `ws`? Records written since
+// `target.workspaceId` exists match on the stable id; older records (and hand-written
+// ones) carry only the display name, so they fall back to a case-insensitive name
+// match. Matching on the name alone orphaned a workspace's whole history on rename and
+// pooled two same-named workspaces of different teammates.
+export function matchesWorkspace(target, ws) {
+  if (!target || target.kind !== 'workspace' || !ws) return false;
+  const id = typeof target.workspaceId === 'string' && target.workspaceId ? target.workspaceId : null;
+  if (id && typeof ws.id === 'string' && ws.id) return id === ws.id;
+  const name = typeof ws.name === 'string' ? ws.name.toLowerCase() : '';
+  return !!name && String(target.workspace || '').toLowerCase() === name;
+}

package/ui/public/about-links.mjs

@@ -0,0 +1,21 @@
+// ui/public/about-links.mjs
+// The About card's outbound links. Split out of app.js's paintAbout so a test can
+// drive it against the real #about-card markup without booting the whole app —
+// app.js has no exports at all. Pure DOM in, pure DOM out; no fetch, no listeners.
+
+/**
+ * Point the two feedback anchors at `info.bugsUrl` (package.json bugs.url, served
+ * as `app.bugsUrl` by GET /api/settings). A missing or empty bugsUrl leaves the
+ * static hrefs in index.html alone, so the links work before /api/settings lands.
+ *
+ * @param {ParentNode} root  any node containing #aboutBugLink / #aboutIdeaLink
+ * @param {{bugsUrl?:string}} info
+ */
+export function paintAboutInto(root, info = {}) {
+  if (!root || !info || typeof info.bugsUrl !== 'string' || !info.bugsUrl) return;
+  const base = info.bugsUrl.replace(/\/+$/, '');
+  const bug = root.querySelector('#aboutBugLink');
+  const idea = root.querySelector('#aboutIdeaLink');
+  if (bug) bug.href = `${base}/new?labels=bug`;
+  if (idea) idea.href = `${base}/new?labels=enhancement`;
+}