Compare commits

5 Commits

Author SHA1 Message Date
Croissant Le Doux
6fdbc302d2 SQLite cache bootstrap + single-issue mirror upsert (#3)
A rebuildable local mirror of the reconciled backlog — an index over gitea's
durable truth, never the source of truth (D4). This lands the core seam:

- `cache/cache-v0.ts`: the 5-table schema (issues/labels/milestones/comments/
  issue_events), a `CacheDriver` injected-IO interface (exec/run/get/all), and
  pure row<->domain mappers. `upsertIssue` mirrors one issue (+ its milestone),
  upsert-by-number so a re-reconcile updates in place; `readIssue` re-derives
  `facts` from the stored label set via extractLabelFacts, so the mirror can't
  drift from the label semantics. Facts are never stored.
- Resolves #3's open scope: pure mappers + SQL in core (driver-agnostic), native
  driver in main. Tests bind node:sqlite (present in the Node 24 test runtime) to
  the same CacheDriver seam better-sqlite3 will fill in main.

Acceptance met: upsert one reconciled issue, read back, assert
estimateDays/priority/hardDeadline (+ upsert-in-place, no-milestone, and miss
cases). Core suite green; typecheck clean.

Follow-up (noted in the plan): the main-process better-sqlite3 adapter +
snapshot-store migration — packaging-sensitive (native module), kept out of this
slice so the shippable .dmg stays verified.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-09 15:30:22 -04:00
2a6413821a Merge pull request 'calibration: count same-day closes honestly (#34)' (#54) from feat/calibration-honesty into main
Reviewed-on: #54
2026-07-09 19:23:59 +00:00
008435f1c2 Merge branch 'main' into feat/calibration-honesty 2026-07-09 19:23:55 +00:00
354ba9227e Merge pull request 'apply_changes: unified mutation tool — estimate/priority/assign/milestone, in-app + agent (#24)' (#53) from feat/apply-changes-assign-milestone into main
Reviewed-on: #53
2026-07-09 19:23:49 +00:00
Croissant Le Doux
89c873b368 calibration: count same-day closes honestly (#34)
The cold-start surface showed "N/20 closed issues estimated", implying you're
just (20−N) closes away. But calibrationSamples silently drops closed+estimated
issues that closed in 0 working days (same-day closes) — real closes that
structurally can't calibrate. On this repo that's 10 of 24 closes hidden: the
note read 14/20 as if 6 away, when a third of the history will never count.

- core: `calibrationCoverage(issues, timelines, asOf)` → { candidates, usable,
  excludedSameDay }, counting the silently-excluded same-day closes. Pure, tested.
- surface it: CalibrationData gains `excludedSameDay`; backlogCalibration returns
  the coverage; the Runway note and the Calibration screen now say "… · N same-day
  closes can't calibrate" so the thin sample is explained, not just reported.

Verified on christian/commitea: closed=24, usable=14, excludedSameDay=10.
131 core green (incl. new coverage test); core + desktop typecheck; 14 fixture e2e.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-09 15:18:16 -04:00
10 changed files with 341 additions and 5 deletions

View File

@@ -178,6 +178,9 @@ export function CalibrationScreen({ onBack, data }: { onBack: () => void; data?:
{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 — Im forecasting from cold-start priors and widening the cone to stay honest. The curve takes over at 20.'}
{!c.active && c.excludedSameDay > 0
? ` And ${c.excludedSameDay} closed ${c.excludedSameDay === 1 ? 'issue' : 'issues'} closed the same day they were started — 0 working days cant calibrate, so they dont count toward the 20.`
: ''}
</p>
</Card>
</div>

View File

@@ -17,7 +17,7 @@ export function RunwayScreen({
}: {
onOpenCalibration: () => void
onOpenMilestone: (id?: number) => void
calibration?: { n: number; coldStart: boolean }
calibration?: { n: number; coldStart: boolean; excludedSameDay?: number }
milestones?: RunwayMilestone[]
capacity?: CapacityMember[]
}) {
@@ -30,9 +30,11 @@ export function RunwayScreen({
hours: `${capacityPerWorkday(m).toFixed(2)} pd/day`,
}))
: CAPACITY
const excluded = calibration?.excludedSameDay ?? 0
const calibNote = calibration
? calibration.coldStart
? `cold-start priors · ${calibration.n}/20 closed issues estimated`
? `cold-start priors · ${calibration.n}/20 closed issues estimated` +
(excluded > 0 ? ` · ${excluded} same-day close${excluded === 1 ? '' : 's'} cant calibrate` : '')
: `calibrated on ${calibration.n} closed ${calibration.n === 1 ? 'issue' : 'issues'}`
: 'calibrated on 27 closed issues'
return (

View File

@@ -306,7 +306,15 @@ export function AppShell() {
setMilestoneId(id ?? null)
setView('milestone')
}}
calibration={calibration ? { n: calibration.model.n, coldStart: calibration.model.coldStart } : undefined}
calibration={
calibration
? {
n: calibration.model.n,
coldStart: calibration.model.coldStart,
excludedSameDay: calibration.coverage.excludedSameDay,
}
: undefined
}
milestones={runwayMilestones}
capacity={capacityMembers}
/>

View File

@@ -361,11 +361,14 @@ export interface CalibrationData {
scatter: number[][]
fit: number
effect: { raw: string; banded: string; p50: string }
/** Closed+estimated issues that can't calibrate (same-day / 0-day closes). */
excludedSameDay: number
}
export const CALIBRATION: CalibrationData = {
n: 27,
active: true,
excludedSameDay: 0,
labels: [
{ label: 'est/1d', n: 8, median: '1.1d', bias: 8 },
{ label: 'est/2d', n: 9, median: '2.4d', bias: 18 },

View File

@@ -1,6 +1,8 @@
import {
type CalibrationCoverage,
type CalibrationModel,
type CalibrationSample,
calibrationCoverage,
calibrationSamples,
type CapacityMember,
capacityPerWorkday,
@@ -171,10 +173,11 @@ export function backlogCalibration(
issues: GiteaIssue[],
timelines: Timelines = {},
asOf: Date = new Date(),
): { model: CalibrationModel; data: CalibrationData } {
): { model: CalibrationModel; data: CalibrationData; coverage: CalibrationCoverage } {
const samples = calibrationSamples(issues, timelines, asOf)
const model = fitCalibration(samples)
return { model, data: calibrationData(model, samples, issues) }
const coverage = calibrationCoverage(issues, timelines, asOf)
return { model, coverage, data: calibrationData(model, samples, issues, coverage.excludedSameDay) }
}
const pctFromMu = (mu: number) => Math.round((Math.exp(mu) - 1) * 100)
@@ -188,6 +191,7 @@ export function calibrationData(
model: CalibrationModel,
samples: CalibrationSample[],
openIssues: GiteaIssue[],
excludedSameDay = 0,
): CalibrationData {
const labels = PRIOR_BUCKETS.map((b) => {
const inBucket = samples.filter((s) => s.bucket === b)
@@ -227,6 +231,7 @@ export function calibrationData(
scatter: samples.map((s) => [s.estimateDays, s.actualWorkingDays]),
fit: Number(Math.exp(model.global.mu).toFixed(2)),
effect,
excludedSameDay,
}
}

View File

@@ -0,0 +1,92 @@
import { DatabaseSync } from 'node:sqlite'
import { describe, expect, it } from 'vitest'
import type { GiteaIssue } from '../gitea/types.js'
import { extractLabelFacts } from '../labels/label-schema.js'
import { type CacheDriver, initCache, readIssue, upsertIssue } from './cache-v0.js'
/** Adapt node:sqlite's DatabaseSync to the CacheDriver seam (main uses better-sqlite3). */
function memoryDriver(): CacheDriver {
const db = new DatabaseSync(':memory:')
return {
exec: (sql) => db.exec(sql),
run: (sql, params = []) => {
db.prepare(sql).run(...(params as never[]))
},
get: (sql, params = []) => db.prepare(sql).get(...(params as never[])) as Record<string, unknown> | undefined,
all: (sql, params = []) => db.prepare(sql).all(...(params as never[])) as Record<string, unknown>[],
}
}
function issue(over: Partial<GiteaIssue> = {}): GiteaIssue {
const labels = over.labels ?? ['est/5d', 'p/1', 'deadline/hard']
return {
number: 42,
title: 'Monte Carlo engine',
body: 'percentile bands',
state: 'open',
labels,
facts: extractLabelFacts(labels),
milestone: { id: 7, title: 'P2 — Scheduler', dueOn: '2026-09-01T00:00:00Z' },
assignee: 'christian',
assignees: ['christian'],
createdAt: '2026-07-08T00:00:00Z',
updatedAt: '2026-07-08T01:00:00Z',
closedAt: null,
url: 'https://gitea/christian/commitea/issues/42',
...over,
}
}
describe('cache-v0', () => {
it('mirrors one issue and reads its facts back through extractLabelFacts (acceptance)', () => {
const d = memoryDriver()
initCache(d)
upsertIssue(d, issue({ labels: ['est/5d', 'p/1', 'deadline/hard'] }))
const back = readIssue(d, 42)!
expect(back.labels).toEqual(['est/5d', 'p/1', 'deadline/hard'])
// facts are re-derived on read, not stored
expect(back.facts.estimateDays).toBe(5)
expect(back.facts.priority).toBe(1)
expect(back.facts.hardDeadline).toBe(true)
// the rest of the domain shape round-trips
expect(back.milestone).toEqual({ id: 7, title: 'P2 — Scheduler', dueOn: '2026-09-01T00:00:00Z' })
expect(back.assignee).toBe('christian')
expect(back.state).toBe('open')
})
it('re-derives facts from the current labels after a re-reconcile (upsert in place, no dup)', () => {
const d = memoryDriver()
initCache(d)
upsertIssue(d, issue({ labels: ['est/2d', 'p/3'] }))
// reconcile again with changed labels + closed
upsertIssue(d, issue({ labels: ['est/8d', 'p/1'], state: 'closed', closedAt: '2026-07-09T00:00:00Z' }))
expect(d.all('SELECT number FROM issues')).toHaveLength(1) // upsert by number, not a second row
const back = readIssue(d, 42)!
expect(back.facts.estimateDays).toBe(8)
expect(back.facts.priority).toBe(1)
expect(back.facts.hardDeadline).toBe(false) // deadline/hard dropped
expect(back.state).toBe('closed')
expect(back.closedAt).toBe('2026-07-09T00:00:00Z')
})
it('reads an issue with no milestone / empty labels', () => {
const d = memoryDriver()
initCache(d)
upsertIssue(d, issue({ number: 9, labels: [], milestone: null, assignee: null, assignees: [] }))
const back = readIssue(d, 9)!
expect(back.milestone).toBeNull()
expect(back.labels).toEqual([])
expect(back.facts.estimateDays).toBeNull()
expect(back.assignee).toBeNull()
})
it('returns null for an uncached issue', () => {
const d = memoryDriver()
initCache(d)
expect(readIssue(d, 999)).toBeNull()
})
})

159
packages/core/src/cache/cache-v0.ts vendored Normal file
View File

@@ -0,0 +1,159 @@
/**
* SQLite cache, v0 (#3) — a rebuildable local mirror of the reconciled backlog.
* It is an index over the durable truth in gitea, never the source of truth (D4):
* delete it, resync, lose nothing. This module owns the schema + the pure
* row<->domain mappers; the actual SQLite handle is injected as a `CacheDriver`,
* so core stays free of any native driver (better-sqlite3 lives in main; tests
* use node:sqlite). Facts are never stored — they are re-derived from the label
* set on read via `extractLabelFacts`, so the mirror can't drift from the label
* semantics.
*/
import type { GiteaIssue, GiteaMilestoneRef } from '../gitea/types.js'
import { extractLabelFacts } from '../labels/label-schema.js'
/**
* The injected IO boundary: a thin synchronous SQL executor. Core writes the SQL;
* the host binds a real driver (better-sqlite3 in the desktop main process,
* node:sqlite in tests). Kept minimal on purpose — no ORM, no query builder.
*/
export interface CacheDriver {
/** Run one or more DDL/utility statements (no params, no result). */
exec(sql: string): void
/** Execute a single parameterized write. */
run(sql: string, params?: readonly unknown[]): void
/** First row of a parameterized query, or undefined. */
get(sql: string, params?: readonly unknown[]): Record<string, unknown> | undefined
/** All rows of a parameterized query. */
all(sql: string, params?: readonly unknown[]): Record<string, unknown>[]
}
/** The cache schema — five tables mirroring gitea's shape. Regenerable; drop and rebuild freely. */
export const CACHE_SCHEMA = `
CREATE TABLE IF NOT EXISTS milestones (
id INTEGER PRIMARY KEY,
title TEXT NOT NULL,
state TEXT,
due_on TEXT
);
CREATE TABLE IF NOT EXISTS issues (
number INTEGER PRIMARY KEY,
title TEXT NOT NULL,
body TEXT NOT NULL DEFAULT '',
state TEXT NOT NULL,
labels TEXT NOT NULL DEFAULT '[]', -- JSON array of label names; facts re-derived on read
milestone_id INTEGER,
assignee TEXT,
assignees TEXT NOT NULL DEFAULT '[]', -- JSON array of logins
created_at TEXT,
updated_at TEXT,
closed_at TEXT,
url TEXT,
FOREIGN KEY (milestone_id) REFERENCES milestones(id)
);
CREATE TABLE IF NOT EXISTS labels (
id INTEGER PRIMARY KEY,
name TEXT NOT NULL
);
CREATE TABLE IF NOT EXISTS comments (
id INTEGER PRIMARY KEY,
issue_number INTEGER NOT NULL,
author TEXT,
body TEXT NOT NULL DEFAULT '',
created_at TEXT
);
CREATE TABLE IF NOT EXISTS issue_events (
id INTEGER PRIMARY KEY AUTOINCREMENT,
issue_number INTEGER NOT NULL,
type TEXT NOT NULL,
at TEXT NOT NULL
);
CREATE INDEX IF NOT EXISTS idx_issue_events_number ON issue_events(issue_number);
CREATE INDEX IF NOT EXISTS idx_comments_number ON comments(issue_number);
`
/** Create the schema if absent. Idempotent. */
export function initCache(driver: CacheDriver): void {
driver.exec(CACHE_SCHEMA)
}
const UPSERT_ISSUE = `
INSERT INTO issues (number, title, body, state, labels, milestone_id, assignee, assignees, created_at, updated_at, closed_at, url)
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
ON CONFLICT(number) DO UPDATE SET
title = excluded.title, body = excluded.body, state = excluded.state, labels = excluded.labels,
milestone_id = excluded.milestone_id, assignee = excluded.assignee, assignees = excluded.assignees,
created_at = excluded.created_at, updated_at = excluded.updated_at, closed_at = excluded.closed_at, url = excluded.url
`
const UPSERT_MILESTONE = `
INSERT INTO milestones (id, title, state, due_on) VALUES (?, ?, ?, ?)
ON CONFLICT(id) DO UPDATE SET title = excluded.title, state = excluded.state, due_on = excluded.due_on
`
/**
* Mirror one reconciled issue into the cache (and its milestone, if any). Upsert
* by `number`, so re-reconciling the same issue updates in place — never duplicates.
*/
export function upsertIssue(driver: CacheDriver, issue: GiteaIssue): void {
if (issue.milestone) {
driver.run(UPSERT_MILESTONE, [issue.milestone.id, issue.milestone.title, null, issue.milestone.dueOn])
}
driver.run(UPSERT_ISSUE, [
issue.number,
issue.title,
issue.body,
issue.state,
JSON.stringify(issue.labels),
issue.milestone?.id ?? null,
issue.assignee,
JSON.stringify(issue.assignees),
issue.createdAt,
issue.updatedAt,
issue.closedAt,
issue.url,
])
}
const READ_ISSUE = `
SELECT i.*, m.title AS m_title, m.due_on AS m_due
FROM issues i LEFT JOIN milestones m ON m.id = i.milestone_id
WHERE i.number = ?
`
function str(v: unknown): string {
return typeof v === 'string' ? v : ''
}
/**
* Read one mirrored issue back as a domain object, re-deriving `facts` from the
* stored label set (so the mirror can't disagree with the label semantics).
* Returns null when the issue isn't cached.
*/
export function readIssue(driver: CacheDriver, number: number): GiteaIssue | null {
const row = driver.get(READ_ISSUE, [number])
if (!row) return null
const labels = (JSON.parse(str(row.labels) || '[]') as string[]) ?? []
const assignees = (JSON.parse(str(row.assignees) || '[]') as string[]) ?? []
const milestone: GiteaMilestoneRef | null =
row.milestone_id != null
? { id: Number(row.milestone_id), title: str(row.m_title), dueOn: (row.m_due as string | null) ?? null }
: null
return {
number: Number(row.number),
title: str(row.title),
body: str(row.body),
state: row.state === 'closed' ? 'closed' : 'open',
labels,
facts: extractLabelFacts(labels),
milestone,
assignee: (row.assignee as string | null) ?? null,
assignees,
createdAt: str(row.created_at),
updatedAt: str(row.updated_at),
closedAt: (row.closed_at as string | null) ?? null,
url: str(row.url),
}
}

View File

@@ -5,6 +5,7 @@ import type { LifecycleEvent } from '../lifecycle/lifecycle-v0.js'
import type { GiteaIssue } from '../gitea/types.js'
import {
CALIBRATION_BUCKET_FLOOR,
calibrationCoverage,
calibrationSamples,
type CalibrationSample,
COLD_START_THRESHOLD,
@@ -118,4 +119,26 @@ describe('calibrationSamples', () => {
const noEst = issue({ number: 9, labels: [] })
expect(calibrationSamples([open, noEst], { ...events(8), ...events(9) }, asOf)).toEqual([])
})
it('coverage counts same-day closes as excluded candidates, not as "more closes needed"', () => {
// usable: commit Wed 01-07 → close Mon 01-12 = 3 working days
const usable = issue({ number: 7, labels: ['est/2d'] })
// same-day close: commit and close on the same day = 0 working days → excluded
const sameDay = issue({ number: 10, labels: ['est/2d'], createdAt: '2026-01-12T08:00:00Z' })
const sameDayEvents = {
10: [
{ type: 'commit', at: '2026-01-12T09:00:00Z' } as LifecycleEvent,
{ type: 'close', at: '2026-01-12T17:00:00Z' } as LifecycleEvent,
],
}
const open = issue({ number: 8, state: 'open', labels: ['est/2d'], closedAt: null })
const noEst = issue({ number: 9, labels: [] })
const cov = calibrationCoverage([usable, sameDay, open, noEst], { ...events(7), ...sameDayEvents }, asOf)
expect(cov.candidates).toBe(2) // closed + estimated only (usable + sameDay)
expect(cov.usable).toBe(1)
expect(cov.excludedSameDay).toBe(1)
// the honest denominator: usable matches the model's n
expect(cov.usable).toBe(calibrationSamples([usable, sameDay, open, noEst], { ...events(7), ...sameDayEvents }, asOf).length)
})
})

View File

@@ -125,3 +125,40 @@ export function calibrationSamples(
}
return out
}
/** How the closed+estimated backlog splits into usable samples vs. what can't calibrate. */
export interface CalibrationCoverage {
/** Closed issues carrying an estimate — the calibration candidates. */
candidates: number
/** Candidates that yielded a usable actual (> 0 working days) → become samples. */
usable: number
/**
* Candidates excluded because the issue closed with 0 working days (same-day
* close) or no resolvable actual — real closes that structurally can't
* calibrate. Counting them keeps `usable/threshold` honest: it's not "N more
* closes away" if some of your closes will never count.
*/
excludedSameDay: number
}
/**
* Coverage of the calibration candidates — how many closed+estimated issues are
* usable vs. silently unusable (same-day / 0-day closes). {@link calibrationSamples}
* drops the latter; this counts them so the UI can say *why* the sample is thin.
*/
export function calibrationCoverage(
issues: GiteaIssue[],
timelines: Record<number, LifecycleEvent[]>,
asOf: Date,
): CalibrationCoverage {
let candidates = 0
let usable = 0
for (const issue of issues) {
if (issue.state !== 'closed') continue
if (issue.facts.estimateDays == null) continue
candidates++
const inf = inferLifecycle(issue, timelines[issue.number] ?? [], asOf)
if (inf.actualWorkingDays != null && inf.actualWorkingDays > 0) usable++
}
return { candidates, usable, excludedSameDay: candidates - usable }
}

View File

@@ -23,6 +23,8 @@ export type {
GiteaRequestInit,
} from './gitea/types.js'
export { CACHE_SCHEMA, initCache, readIssue, upsertIssue } from './cache/cache-v0.js'
export type { CacheDriver } from './cache/cache-v0.js'
export { describeChange, isLabelChange, planIssueChange, proposalsFor, summarizeChange } from './changes/apply-changes-v0.js'
export type {
ChangeProposal,
@@ -80,6 +82,7 @@ export type {
export {
CALIBRATION_BUCKET_FLOOR,
calibrationCoverage,
calibrationSamples,
COLD_START_THRESHOLD,
fitCalibration,
@@ -87,6 +90,7 @@ export {
} from './calibration/calibration-v0.js'
export type {
BucketFit,
CalibrationCoverage,
CalibrationModel,
CalibrationSample,
PersonBias,