feat: calibration from closed-issue actuals → forecast flips off cold-start (#1)

Close the D3 loop. The forecast now learns from the team's own estimate-vs-actual
history (the working time #5 infers from git events) instead of guessing forever.

core (@commitea/core/calibration-v0):
- fitCalibration(samples): lognormal fit on log(actual/estimate) — global +
  per-bucket (once a bucket clears the floor) + per-person bias. coldStart until
  n >= 20 closed-with-estimate issues.
- calibrationSamples(): pull those samples from the closed backlog via lifecycle
  inference (estimate label vs inferred actualWorkingDays).
- toDurationModel(): project the fit to the params forecast consumes.
- forecast() gains options.model: when past cold-start, fitted params drive the
  sim (per bucket, global fallback); otherwise the code priors do. Forecast.coldStart
  now reflects the model. nearestBucket extracted + exported.

app:
- AppShell fits calibration once from the reconciled backlog, feeds the model into
  forecastBacklog (cone), and drives the Calibration screen + Runway header.
- Focus cone footer, Runway note, and Calibration screen now say cold-start (N/20)
  vs calibrated (on N closed) from real data; Calibration scatter / bucket bias /
  per-person all fitted, degrading honestly on a thin dataset.

Known refinement: same-day closes yield 0 working-day actuals (day-granular) and
are excluded, so a fast-moving repo can sit at n=0 — honest, but a fractional
(hours-based) actual would let those count. Per-person uses gitea login, not
display name, until the person map lands.

Note: also re-lands #10 (Monte Carlo) and #5 (lifecycle) which merged into their
stacked base branches but never propagated to main (stacked-merge trap); this
branch is cut from main and carries all three so main is whole again.

Verified: 74 core tests green (9 calibration + 2 forecast-switch added), desktop
typecheck clean, 14 fixture e2e green, live spec asserts the real cold-start
calibration surface (Runway note + screen badge fitted from actuals).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

packages/core/src/calibration/calibration-v0.ts

@@ -0,0 +1,127 @@
+/**
+ * Calibration, v0 — fit the team's own estimate-vs-actual history so the
+ * forecast stops guessing (D3). The "actual" is the working time lifecycle
+ * inference derives from git events (#5), never manual tracking. Fit a
+ * lognormal on log(actual / estimate) globally and per estimate bucket; until
+ * the sample clears the cold-start threshold, the forecast keeps using the
+ * code-resident priors and this model just reports progress toward it.
+ */
+
+import { type DurationModel, type LognormalPrior, nearestBucket } from '../forecast/forecast-v0.js'
+import { inferLifecycle, type LifecycleEvent } from '../lifecycle/lifecycle-v0.js'
+import type { GiteaIssue } from '../gitea/types.js'
+
+/** Global sample size at which the fit takes over from the cold-start priors. */
+export const COLD_START_THRESHOLD = 20
+/** Minimum per-bucket sample before that bucket earns its own fit. */
+export const CALIBRATION_BUCKET_FLOOR = 3
+/** Fallback spread when a group is too small to estimate one. */
+const DEFAULT_SIGMA = 0.4
+
+/** One closed issue's estimate vs its inferred actual. */
+export interface CalibrationSample {
+  issue: number
+  estimateDays: number
+  actualWorkingDays: number
+  bucket: number
+  person: string | null
+}
+
+export interface BucketFit extends LognormalPrior {
+  n: number
+}
+
+export interface PersonBias {
+  /** Additive to global mu (log space). */
+  biasMu: number
+  n: number
+}
+
+export interface CalibrationModel {
+  /** Closed issues with an estimate + a resolvable actual. */
+  n: number
+  /** true while n < COLD_START_THRESHOLD — forecast keeps the code priors. */
+  coldStart: boolean
+  global: LognormalPrior
+  byBucket: Record<number, BucketFit>
+  byPerson: Record<string, PersonBias>
+}
+
+function mean(xs: number[]): number {
+  return xs.reduce((a, b) => a + b, 0) / xs.length
+}
+
+/** Sample standard deviation; falls back to DEFAULT_SIGMA below 2 points. */
+function stddev(xs: number[], mu: number): number {
+  if (xs.length < 2) return DEFAULT_SIGMA
+  const variance = xs.reduce((a, x) => a + (x - mu) ** 2, 0) / (xs.length - 1)
+  return Math.sqrt(variance) || DEFAULT_SIGMA
+}
+
+/** Fit a calibration model from estimate-vs-actual samples. Pure. */
+export function fitCalibration(samples: CalibrationSample[]): CalibrationModel {
+  const usable = samples.filter((s) => s.estimateDays > 0 && s.actualWorkingDays > 0)
+  const n = usable.length
+  const coldStart = n < COLD_START_THRESHOLD
+
+  const logRatios = usable.map((s) => Math.log(s.actualWorkingDays / s.estimateDays))
+  const globalMu = n ? mean(logRatios) : 0
+  const global: LognormalPrior = { mu: globalMu, sigma: n ? stddev(logRatios, globalMu) : DEFAULT_SIGMA }
+
+  const byBucket: Record<number, BucketFit> = {}
+  const byPerson: Record<string, PersonBias> = {}
+  const groups = new Map<number, number[]>()
+  const people = new Map<string, number[]>()
+  for (const s of usable) {
+    const lr = Math.log(s.actualWorkingDays / s.estimateDays)
+    ;(groups.get(s.bucket) ?? groups.set(s.bucket, []).get(s.bucket)!).push(lr)
+    if (s.person) (people.get(s.person) ?? people.set(s.person, []).get(s.person)!).push(lr)
+  }
+  for (const [bucket, lrs] of groups) {
+    if (lrs.length < CALIBRATION_BUCKET_FLOOR) continue
+    const mu = mean(lrs)
+    byBucket[bucket] = { mu, sigma: stddev(lrs, mu), n: lrs.length }
+  }
+  for (const [person, lrs] of people) {
+    if (lrs.length < CALIBRATION_BUCKET_FLOOR) continue
+    byPerson[person] = { biasMu: mean(lrs) - globalMu, n: lrs.length }
+  }
+
+  return { n, coldStart, global, byBucket, byPerson }
+}
+
+/** The subset of a model `forecast` consumes. */
+export function toDurationModel(model: CalibrationModel): DurationModel {
+  const byBucket: Record<number, LognormalPrior> = {}
+  for (const [bucket, fit] of Object.entries(model.byBucket)) {
+    byBucket[Number(bucket)] = { mu: fit.mu, sigma: fit.sigma }
+  }
+  return { coldStart: model.coldStart, global: model.global, byBucket }
+}
+
+/**
+ * Extract calibration samples from the closed backlog: each closed issue that
+ * carries an estimate and yields an inferred actual working duration.
+ */
+export function calibrationSamples(
+  issues: GiteaIssue[],
+  timelines: Record<number, LifecycleEvent[]>,
+  asOf: Date,
+): CalibrationSample[] {
+  const out: CalibrationSample[] = []
+  for (const issue of issues) {
+    if (issue.state !== 'closed') continue
+    const estimateDays = issue.facts.estimateDays
+    if (estimateDays == null) continue
+    const inf = inferLifecycle(issue, timelines[issue.number] ?? [], asOf)
+    if (inf.actualWorkingDays == null || inf.actualWorkingDays <= 0) continue
+    out.push({
+      issue: issue.number,
+      estimateDays,
+      actualWorkingDays: inf.actualWorkingDays,
+      bucket: nearestBucket(estimateDays),
+      person: issue.assignee,
+    })
+  }
+  return out
+}