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>

apps/desktop/src/renderer/src/components/screens/calibration-screen.tsx

@@ -1,11 +1,12 @@
 import React from 'react'
 
-import { CALIBRATION } from '../../data/fixtures.js'
+import { CALIBRATION, type CalibrationData } from '../../data/fixtures.js'
 import { Badge, Card, Icon } from '../ui/index.js'
 
-// Calibration report — estimate-vs-actual evidence behind the cones
-export function CalibrationScreen({ onBack }: { onBack: () => void }) {
-  const c = CALIBRATION
+// Calibration report — estimate-vs-actual evidence behind the cones.
+// `data` (real fit from closed-issue actuals) overrides the demo fixture.
+export function CalibrationScreen({ onBack, data }: { onBack: () => void; data?: CalibrationData }) {
+  const c = data ?? CALIBRATION
 
   // scatter chart geometry
   const W = 420,
@@ -68,7 +69,11 @@ export function CalibrationScreen({ onBack }: { onBack: () => void }) {
             <h1 style={{ font: 'var(--text-display)', color: 'var(--ink-1)', margin: 0 }}>Calibration</h1>
             <p style={{ font: 'var(--text-data)', color: 'var(--ink-3)', margin: '6px 0 0', whiteSpace: 'nowrap' }}>{c.n} closed issues with estimates · evidence, not opinion</p>
           </div>
-          <Badge tone="ok" dot>curve active · n ≥ 20</Badge>
+          {c.active ? (
+            <Badge tone="ok" dot>curve active · n ≥ 20</Badge>
+          ) : (
+            <Badge tone="warn" dot>cold-start · {c.n}/20</Badge>
+          )}
         </header>
       </div>
 
@@ -170,7 +175,9 @@ export function CalibrationScreen({ onBack }: { onBack: () => void }) {
               <span style={{ font: '500 12.5px var(--font-mono)', color: 'var(--ink-1)', whiteSpace: 'nowrap' }}>{c.effect.banded}</span>
             </div>
             <p style={{ font: 'var(--text-agent)', color: 'var(--ink-2)', margin: '10px 0 0' }}>
-              You are not bad at estimating; you are optimistic in a very stable way. Stable, I can work with.
+              {c.active
+                ? 'You are not bad at estimating; you are optimistic in a very stable way. Stable, I can work with.'
+                : 'Not enough closed history yet — I’m forecasting from cold-start priors and widening the cone to stay honest. The curve takes over at 20.'}
             </p>
           </Card>
         </div>

apps/desktop/src/renderer/src/components/screens/focus-screen.tsx

@@ -109,7 +109,9 @@ export function FocusScreen({
         <BurnUpCone data={forecast?.cone} />
         <p style={{ font: 'var(--text-agent)', color: 'var(--ink-2)', margin: '10px 0 0' }}>
           {forecast
-            ? `${forecast.scope} open ${forecast.scope === 1 ? 'issue' : 'issues'} in scope. Cold-start priors — the cone tightens as the team closes work.`
+            ? forecast.coldStart
+              ? `${forecast.scope} open ${forecast.scope === 1 ? 'issue' : 'issues'} in scope. Cold-start priors — ${forecast.calibratedN}/20 estimated closes so far; the cone tightens as the team closes work.`
+              : `${forecast.scope} open ${forecast.scope === 1 ? 'issue' : 'issues'} in scope, calibrated on ${forecast.calibratedN} closed ${forecast.calibratedN === 1 ? 'issue' : 'issues'} of your own.`
             : 'The cone has narrowed since Friday. I’m quietly pleased.'}
         </p>
       </Card>

apps/desktop/src/renderer/src/components/screens/runway-screen.tsx

@@ -8,15 +8,22 @@ import { RUNWAY, CAPACITY } from '../../data/fixtures.js'
 export function RunwayScreen({
   onOpenCalibration,
   onOpenMilestone,
+  calibration,
 }: {
   onOpenCalibration: () => void
   onOpenMilestone: () => void
+  calibration?: { n: number; coldStart: boolean }
 }) {
+  const calibNote = calibration
+    ? calibration.coldStart
+      ? `cold-start priors · ${calibration.n}/20 closed issues estimated`
+      : `calibrated on ${calibration.n} closed ${calibration.n === 1 ? 'issue' : 'issues'}`
+    : 'calibrated on 27 closed issues'
   return (
     <div style={{ display: 'flex', flexDirection: 'column', gap: 16 }}>
       <header style={{ borderBottom: 'var(--rule-double)', paddingBottom: 14 }}>
         <h1 style={{ font: 'var(--text-display)', color: 'var(--ink-1)', margin: 0 }}>Runway</h1>
-        <p style={{ font: 'var(--text-data)', color: 'var(--ink-3)', margin: '6px 0 0' }}>capacity vs milestone dates · calibrated on 27 closed issues</p>
+        <p style={{ font: 'var(--text-data)', color: 'var(--ink-3)', margin: '6px 0 0' }}>capacity vs milestone dates · {calibNote}</p>
       </header>
 
       <Card overline="Milestones" flush>

apps/desktop/src/renderer/src/components/shell/app-shell.tsx

@@ -2,7 +2,7 @@ import React, { useEffect, useState } from 'react'
 
 import logoIcon from '../../design/assets/logo-icon.png'
 import type { IssueRef } from '../../data/fixtures.js'
-import { forecastBacklog, issuesToBoardColumns, scheduleFocus } from '../../lib/backlog.js'
+import { backlogCalibration, forecastBacklog, issuesToBoardColumns, scheduleFocus } from '../../lib/backlog.js'
 import { useBacklog } from '../../lib/use-backlog.js'
 import { PrimitivesGallery } from '../gallery.js'
 import { BoardScreen } from '../screens/board-screen.js'
@@ -90,8 +90,12 @@ export function AppShell() {
     backlog.status === 'ready' ? issuesToBoardColumns(backlog.issues, backlog.timelines) : undefined
   const focus =
     backlog.status === 'ready' ? scheduleFocus(backlog.issues, backlog.deps, backlog.timelines) : undefined
+  const calibration =
+    backlog.status === 'ready' ? backlogCalibration(backlog.issues, backlog.timelines) : undefined
   const forecast =
-    backlog.status === 'ready' ? (forecastBacklog(backlog.issues, backlog.deps) ?? undefined) : undefined
+    backlog.status === 'ready'
+      ? (forecastBacklog(backlog.issues, backlog.deps, new Date(), calibration?.model) ?? undefined)
+      : undefined
 
   useEffect(() => {
     document.documentElement.setAttribute('data-theme', dark ? 'dark' : 'light')
@@ -178,10 +182,11 @@ export function AppShell() {
           <RunwayScreen
             onOpenCalibration={() => setView('calibration')}
             onOpenMilestone={() => setView('milestone')}
+            calibration={calibration ? { n: calibration.model.n, coldStart: calibration.model.coldStart } : undefined}
           />
         )
       case 'calibration':
-        return <CalibrationScreen onBack={() => setView('runway')} />
+        return <CalibrationScreen onBack={() => setView('runway')} data={calibration?.data} />
       case 'milestone':
         return <MilestoneScreen onBack={() => setView('runway')} onOpenIssue={openIssue} />
       case 'inbox':

apps/desktop/src/renderer/src/lib/backlog.ts

@@ -1,17 +1,24 @@
 import {
+  type CalibrationModel,
+  type CalibrationSample,
+  calibrationSamples,
+  COLD_START_THRESHOLD,
   type DependencyEdge,
+  fitCalibration,
   forecast,
   type GiteaIssue,
   inferLifecycle,
   type LifecycleColumn,
   type LifecycleEvent,
   type LifecycleInference,
+  PRIOR_BUCKETS,
   schedule,
   type ScheduledItem,
   selectFocus,
+  toDurationModel,
 } from '@commitea/core'
 
-import { type BoardColumn, type BoardIssue, type FocusIssue } from '../data/fixtures.js'
+import { type BoardColumn, type BoardIssue, type CalibrationData, type FocusIssue } from '../data/fixtures.js'
 import { type BurnUpData, buildBurnUpData } from './dates.js'
 
 type Timelines = Record<number, LifecycleEvent[]>
@@ -100,22 +107,109 @@ export interface ForecastView {
   cone: BurnUpData
   p80Label: string
   rangeLabel: string
+  /** true while the forecast still runs on code priors (calibration not yet trusted). */
+  coldStart: boolean
+  /** Closed-with-estimate issues feeding calibration so far. */
+  calibratedN: number
 }
 
 /**
  * Monte Carlo forecast over the open backlog, mapped onto a calendar-anchored
- * burn-up cone. Returns null when there's nothing to forecast (no open scope) —
- * the UI then falls back to the demo cone. `today` is injectable for tests.
+ * burn-up cone. When a calibration model is supplied and past cold-start, its
+ * fitted params drive the sim. Returns null when there's nothing to forecast.
+ * `today` is injectable for tests.
  */
 export function forecastBacklog(
   issues: GiteaIssue[],
   deps: DependencyEdge[],
   today: Date = new Date(),
+  calibration?: CalibrationModel,
 ): ForecastView | null {
-  const f = forecast(toSchedulable(issues), deps)
+  const model = calibration ? toDurationModel(calibration) : undefined
+  const f = forecast(toSchedulable(issues), deps, model ? { model } : {})
   const cone = buildBurnUpData(f, today)
   if (!cone) return null
-  return { scope: f.scope, cone, p80Label: cone.p80Label, rangeLabel: cone.rangeLabel }
+  return {
+    scope: f.scope,
+    cone,
+    p80Label: cone.p80Label,
+    rangeLabel: cone.rangeLabel,
+    coldStart: f.coldStart,
+    calibratedN: calibration?.n ?? 0,
+  }
+}
+
+/** Fit the calibration model from the closed backlog's inferred actuals (#1). */
+export function calibrateBacklog(
+  issues: GiteaIssue[],
+  timelines: Timelines = {},
+  asOf: Date = new Date(),
+): CalibrationModel {
+  return fitCalibration(calibrationSamples(issues, timelines, asOf))
+}
+
+/** Calibration model + its screen view in one pass over the closed backlog. */
+export function backlogCalibration(
+  issues: GiteaIssue[],
+  timelines: Timelines = {},
+  asOf: Date = new Date(),
+): { model: CalibrationModel; data: CalibrationData } {
+  const samples = calibrationSamples(issues, timelines, asOf)
+  const model = fitCalibration(samples)
+  return { model, data: calibrationData(model, samples, issues) }
+}
+
+const pctFromMu = (mu: number) => Math.round((Math.exp(mu) - 1) * 100)
+
+/**
+ * Shape the calibration model + its samples into the screen's view. Buckets and
+ * people only earn a bias once their sample clears the fit floor; everything
+ * degrades honestly on a thin (cold-start) dataset.
+ */
+export function calibrationData(
+  model: CalibrationModel,
+  samples: CalibrationSample[],
+  openIssues: GiteaIssue[],
+): CalibrationData {
+  const labels = PRIOR_BUCKETS.map((b) => {
+    const inBucket = samples.filter((s) => s.bucket === b)
+    const fit = model.byBucket[b]
+    const mu = fit ? fit.mu : model.global.mu
+    return {
+      label: `est/${b}d`,
+      n: fit ? fit.n : inBucket.length,
+      median: inBucket.length ? `${(b * Math.exp(mu)).toFixed(1)}d` : '—',
+      bias: fit ? pctFromMu(fit.mu) : null,
+    }
+  })
+
+  const people = Object.entries(model.byPerson).map(([who, pb]) => ({
+    who,
+    n: pb.n,
+    bias: pctFromMu(model.global.mu + pb.biasMu),
+    note: '',
+  }))
+
+  const openEst = openIssues
+    .filter((i) => i.state === 'open')
+    .reduce((sum, i) => sum + (i.facts.estimateDays ?? 2), 0)
+  const effect = model.coldStart
+    ? { raw: `${model.n}/${COLD_START_THRESHOLD} estimated closes`, banded: 'cold-start priors', p50: '—' }
+    : {
+        raw: `${openEst}d estimated`,
+        banded: `×${Math.exp(model.global.mu).toFixed(2)} median drift`,
+        p50: `≈${Math.round(openEst * Math.exp(model.global.mu))}d`,
+      }
+
+  return {
+    n: model.n,
+    active: !model.coldStart,
+    labels,
+    people,
+    scatter: samples.map((s) => [s.estimateDays, s.actualWorkingDays]),
+    fit: Number(Math.exp(model.global.mu).toFixed(2)),
+    effect,
+  }
 }
 
 /**