post-0.9.18 polish: compare flags in the TUI, context id prefix, active-days label

src/compare-stats.ts

@@ -2,6 +2,7 @@ import { readdir, readFile } from 'fs/promises'
 import { join } from 'path'
 
 import type { ProjectSummary } from './types.js'
+import { getShortModelName } from './models.js'
 
 const PLANNING_TOOLS = new Set(['TaskCreate', 'TaskUpdate', 'TodoWrite', 'EnterPlanMode', 'ExitPlanMode'])
 
@@ -73,6 +74,16 @@ export function aggregateModelStats(projects: ProjectSummary[]): ModelStats[] {
   return [...byModel.values()].sort((a, b) => b.cost - a.cost)
 }
 
+/// Look up a model by the exact canonical id (what --model-a/--model-b has
+/// always accepted) or, failing that, by its display name (what the compare
+/// picker actually shows the user, e.g. "Opus 4.8") - case-insensitively.
+/// Reuses getShortModelName, the existing canonical -> display mapping,
+/// rather than a new alias table.
+export function findModelStat(models: ModelStats[], input: string): ModelStats | undefined {
+  return models.find(m => m.model === input)
+    ?? models.find(m => getShortModelName(m.model).toLowerCase() === input.toLowerCase())
+}
+
 export type ComparisonRow = {
   section: string
   label: string

src/compare.tsx

@@ -2,7 +2,7 @@ import React, { useState, useEffect, useRef } from 'react'
 import { render, Box, Text, useInput, useApp, useStdout } from 'ink'
 
 import type { ModelStats, ComparisonRow, CategoryComparison, WorkingStyleRow } from './compare-stats.js'
-import { aggregateModelStats, computeComparison, computeCategoryComparison, computeWorkingStyle, scanSelfCorrections } from './compare-stats.js'
+import { aggregateModelStats, computeComparison, computeCategoryComparison, computeWorkingStyle, findModelStat, scanSelfCorrections } from './compare-stats.js'
 import { formatCost } from './format.js'
 import { parseAllSessions, setInteractiveScanUI } from './parser.js'
 import { getAllProviders } from './providers/index.js'
@@ -333,9 +333,13 @@ function ComparisonResults({ modelA, modelB, rows, categories, workingStyle, onB
 type CompareViewProps = {
   projects: ProjectSummary[]
   onBack: () => void
+  // Pre-resolved canonical model ids from --model-a/--model-b (already
+  // validated to exist in `projects`' aggregated stats by the caller). When
+  // set, comparison results load immediately instead of showing the picker.
+  presetModels?: [string, string]
 }
 
-export function CompareView({ projects, onBack }: CompareViewProps) {
+export function CompareView({ projects, onBack, presetModels }: CompareViewProps) {
   const { exit } = useApp()
   const [phase, setPhase] = useState<'select' | 'loading' | 'results'>('select')
   const [models, setModels] = useState<ModelStats[]>(() => aggregateModelStats(projects))
@@ -347,13 +351,13 @@ export function CompareView({ projects, onBack }: CompareViewProps) {
     }
     return recs
   })
-  const [pickedNames, setPickedNames] = useState<[string, string] | null>(null)
+  const [pickedNames, setPickedNames] = useState<[string, string] | null>(presetModels ?? null)
   const [selectedA, setSelectedA] = useState<ModelStats | null>(null)
   const [selectedB, setSelectedB] = useState<ModelStats | null>(null)
   const [rows, setRows] = useState<ComparisonRow[]>([])
   const [categories, setCategories] = useState<CategoryComparison[]>([])
   const [style, setStyle] = useState<WorkingStyleRow[]>([])
-  const [loadTrigger, setLoadTrigger] = useState(0)
+  const [loadTrigger, setLoadTrigger] = useState(presetModels ? 1 : 0)
   const projectsRef = useRef(projects)
   projectsRef.current = projects
 
@@ -504,7 +508,7 @@ export function CompareView({ projects, onBack }: CompareViewProps) {
   )
 }
 
-export async function renderCompare(range: DateRange, provider: string): Promise<void> {
+export async function renderCompare(range: DateRange, provider: string, modelA?: string, modelB?: string): Promise<void> {
   // Interactive Ink UI: suppress the CLI scan-progress line for the whole
   // lifetime so it can't print over the rendered comparison. Plain CLI
   // commands still show progress.
@@ -517,8 +521,28 @@ export async function renderCompare(range: DateRange, provider: string): Promise
 
   patchStdoutForWindows()
   const projects = await parseAllSessions(range, provider)
+
+  // --model-a/--model-b: resolve up front (by canonical id or display name,
+  // same lookup the JSON path uses) so the TUI jumps straight to results
+  // instead of ignoring the flags and showing the picker.
+  let presetModels: [string, string] | undefined
+  if (modelA && modelB) {
+    const models = aggregateModelStats(projects)
+    const a = findModelStat(models, modelA)
+    const b = findModelStat(models, modelB)
+    if (!a) {
+      process.stderr.write(`codeburn compare: model not found: "${modelA}".\n`)
+      process.exit(1)
+    }
+    if (!b) {
+      process.stderr.write(`codeburn compare: model not found: "${modelB}".\n`)
+      process.exit(1)
+    }
+    presetModels = [a.model, b.model]
+  }
+
   const { waitUntilExit } = render(
-    <CompareView projects={projects} onBack={() => process.exit(0)} />
+    <CompareView projects={projects} onBack={() => process.exit(0)} presetModels={presetModels} />
   )
   await waitUntilExit()
 }

src/dashboard.tsx

@@ -3,7 +3,7 @@ import { homedir } from 'os'
 import React, { useState, useCallback, useEffect, useRef } from 'react'
 import { render, Box, Text, useInput, useApp, useWindowSize } from 'ink'
 import { CATEGORY_LABELS, type DateRange, type ProjectSummary, type TaskCategory } from './types.js'
-import { formatCost, formatTokens, markEstimated } from './format.js'
+import { formatCost, formatTokens, markEstimated, carriedCostNote } from './format.js'
 import { aggregateModelEfficiency } from './model-efficiency.js'
 import { parseAllSessions, filterProjectsByDateRange, filterProjectsByName, setInteractiveScanUI } from './parser.js'
 import { findUnpricedModels, loadPricing } from './models.js'
@@ -37,6 +37,16 @@ export function scrollHistoryCursor(cursor: number, direction: -1 | 1, pageSize:
   return Math.max(0, Math.min(cursor + direction, maxCursor))
 }
 
+// The Daily Activity panel's row count comes from a bounded live scan (see
+// getDashboardScanRange), which can undercount vs. the durable-cache-backed
+// Overview headline for the same period (expired session files aren't in the
+// live scan but are still in the durable cache). "days scanned" names that
+// population so the two counts read as different questions, not a
+// contradiction.
+export function dailyActivityFooter(cursor: number, days: number, rowCount: number): string {
+  return `Showing ${cursor + 1}–${Math.min(cursor + days, rowCount)} of ${rowCount} days scanned · newest first`
+}
+
 // Scrollable mode keeps the dashboard up when only the selected period is
 // empty (full history still renders), but a truly-new user with no history at
 // all should still get the clean empty state instead of a zeroed shell.
@@ -143,6 +153,11 @@ export type DurableOverview = {
   outputTokens: number
   cacheReadTokens: number
   cacheWriteTokens: number
+  // Cost from days whose session logs have since expired, carried forward
+  // from the daily cache. Surfaced so the Overview headline can explain why
+  // its total may exceed what the (live-scan-bounded) Daily Activity panel
+  // below can show. See carriedCostNote in format.ts.
+  carriedCostUSD: number
 }
 
 async function computeDurableOverview(
@@ -154,7 +169,7 @@ async function computeDurableOverview(
   day: string | null,
 ): Promise<DurableOverview> {
   const range = day ? getDayRange(day) : customRange ?? getPeriodRange(period)
-  const { data } = await buildDurablePeriod(
+  const { data, carriedCostUSD } = await buildDurablePeriod(
     { range, label: PERIOD_LABELS[period] },
     { provider, project: projectFilter ?? [], exclude: excludeFilter ?? [] },
   )
@@ -167,6 +182,7 @@ async function computeDurableOverview(
     outputTokens: data.outputTokens,
     cacheReadTokens: data.cacheReadTokens,
     cacheWriteTokens: data.cacheWriteTokens,
+    carriedCostUSD,
   }
 }
 
@@ -315,6 +331,9 @@ function Overview({ projects, label, width, planUsages, durable }: { projects: P
           <Text dimColor> saved by local models</Text>
         </Text>
       )}
+      {durable && carriedCostNote(durable.carriedCostUSD) && (
+        <Text dimColor wrap="truncate-end">  {carriedCostNote(durable.carriedCostUSD)}</Text>
+      )}
       {activePlanUsages.length > 0 && (
         <>
           {activePlanUsages.map(planUsage => {
@@ -379,7 +398,7 @@ function DailyActivity({ projects, days = 14, pw, bw, scrollable = false, cursor
               </Text>
             ))}
             {scrollable && orderedRows.length > 0 && (
-              <Text dimColor wrap="truncate-end">Showing {cursor + 1}–{Math.min(cursor + days, orderedRows.length)} of {orderedRows.length} · newest first</Text>
+              <Text dimColor wrap="truncate-end">{dailyActivityFooter(cursor, days, orderedRows.length)}</Text>
             )}
           </>}
     </Panel>

src/format.ts

@@ -15,6 +15,16 @@ export function markEstimated(costStr: string, isEstimated: boolean): string {
   return isEstimated ? `~${costStr}` : costStr
 }
 
+/// Shared wording for the durable-cache carry-forward footnote: some of a
+/// period's total came from days whose session logs have since expired, but
+/// the figure is real (preserved in the durable daily cache). overview.ts and
+/// dashboard.tsx both show this so a headline that includes carried days
+/// doesn't read as inconsistent with detail views that can only see
+/// surviving session files.
+export function carriedCostNote(carriedCostUSD: number): string | null {
+  return carriedCostUSD > 0 ? `includes ${formatCost(carriedCostUSD)} preserved from expired session logs` : null
+}
+
 export function formatTokens(n: number): string {
   // Guard against Infinity / NaN / negatives that would otherwise leak into
   // the UI as "Infinity" or "NaN" strings when an upstream calculation glitches.

src/main.ts

@@ -1857,7 +1857,7 @@ program
     await loadPricing()
     const { range, label } = getDateRange(opts.period)
     if (opts.format === 'json') {
-      const { aggregateModelStats, buildCompareJson, renderCompareJson, scanSelfCorrections } = await import('./compare-stats.js')
+      const { aggregateModelStats, buildCompareJson, findModelStat, renderCompareJson, scanSelfCorrections } = await import('./compare-stats.js')
       const projects = await parseAllSessions(range, opts.provider)
       const models = aggregateModelStats(projects)
 
@@ -1880,8 +1880,8 @@ program
         process.stderr.write('codeburn compare: --model-a and --model-b must be provided together.\n')
         process.exit(1)
       }
-      const modelA = models.find(model => model.model === opts.modelA)
-      const modelB = models.find(model => model.model === opts.modelB)
+      const modelA = findModelStat(models, opts.modelA)
+      const modelB = findModelStat(models, opts.modelB)
       if (!modelA) {
         process.stderr.write(`codeburn compare: model not found: "${opts.modelA}".\n`)
         process.exit(1)
@@ -1893,7 +1893,13 @@ program
       process.stdout.write(renderCompareJson(buildCompareJson(projects, modelA, modelB, label, opts.provider)) + '\n')
       return
     }
-    await renderCompare(range, opts.provider)
+    if (opts.modelA || opts.modelB) {
+      if (!opts.modelA || !opts.modelB) {
+        process.stderr.write('codeburn compare: --model-a and --model-b must be provided together.\n')
+        process.exit(1)
+      }
+    }
+    await renderCompare(range, opts.provider, opts.modelA, opts.modelB)
   })
 
 program

src/web-dashboard.ts

@@ -322,8 +322,12 @@ export async function runWebDashboard(opts: {
           writeJsonError(res, 400, 'provider (claude|codex) and id are required')
           return
         }
+        // findClaudeSession/findCodexSession resolve an id prefix the same way
+        // `codeburn context <session>` does (see context-tree.ts's resolveSession) -
+        // trust that match instead of re-requiring the full id here, so a prefix
+        // that works on the CLI works through the API too.
         const ref = provider === 'claude' ? await findClaudeSession(id) : await findCodexSession(id)
-        if (!ref || ref.sessionId !== id) {
+        if (!ref) {
           writeJsonError(res, 404, `no ${provider} session ${id}`)
           return
         }

tests/cli-emitters.test.ts

@@ -79,6 +79,40 @@ describe('CLI JSON emitters', () => {
     }
   })
 
+  // Issue #767 item 1: --model-a/--model-b used to require the canonical id
+  // verbatim, with no way to pass the display name the compare picker itself
+  // shows (e.g. "Sonnet 4.5" for claude-sonnet-4-5).
+  it('resolves --model-a/--model-b by display name', async () => {
+    const home = await makeHome()
+    try {
+      const result = runCli([
+        'compare', '--format', 'json', '--period', 'all', '--provider', 'claude',
+        '--model-a', 'Sonnet 4.5', '--model-b', 'Opus 4.5',
+      ], home)
+      expect(result.status, result.stderr).toBe(0)
+      const report = JSON.parse(result.stdout)
+      expect(report.modelA.model).toBe('claude-sonnet-4-5')
+      expect(report.modelB.model).toBe('claude-opus-4-5')
+    } finally {
+      await rm(home, { recursive: true, force: true })
+    }
+  })
+
+  // Issue #767 item 1: the TUI path (default --format) used to silently
+  // ignore --model-a/--model-b entirely. It should at least apply the same
+  // "both or neither" validation the JSON path already had, rather than
+  // dropping a lone --model-a on the floor.
+  it('rejects a lone --model-a for the TUI (default) format same as JSON', async () => {
+    const home = await makeHome()
+    try {
+      const result = runCli(['compare', '--period', 'all', '--provider', 'claude', '--model-a', 'claude-sonnet-4-5'], home)
+      expect(result.status).not.toBe(0)
+      expect(result.stderr).toMatch(/--model-a and --model-b must be provided together/)
+    } finally {
+      await rm(home, { recursive: true, force: true })
+    }
+  })
+
   it('emits flat session rows as JSON', async () => {
     const home = await makeHome()
     try {

tests/compare-stats.test.ts

@@ -2,7 +2,7 @@ import { mkdtemp, mkdir, rm, writeFile } from 'fs/promises'
 import { join } from 'path'
 import { tmpdir } from 'os'
 import { describe, it, expect, beforeEach, afterEach } from 'vitest'
-import { aggregateModelStats, buildCompareJson, computeComparison, computeCategoryComparison, computeWorkingStyle, renderCompareJson, scanSelfCorrections, type ModelStats } from '../src/compare-stats.js'
+import { aggregateModelStats, buildCompareJson, computeComparison, computeCategoryComparison, computeWorkingStyle, findModelStat, renderCompareJson, scanSelfCorrections, type ModelStats } from '../src/compare-stats.js'
 import type { ProjectSummary, SessionSummary, ClassifiedTurn } from '../src/types.js'
 
 function makeTurn(model: string, cost: number, opts: { hasEdits?: boolean; retries?: number; outputTokens?: number; inputTokens?: number; cacheRead?: number; cacheWrite?: number; timestamp?: string; category?: string; hasAgentSpawn?: boolean; hasPlanMode?: boolean; speed?: 'standard' | 'fast'; tools?: string[] } = {}): ClassifiedTurn {
@@ -596,3 +596,30 @@ describe('computeWorkingStyle', () => {
     expect(planning.valueA).toBeCloseTo(50)
   })
 })
+
+// Issue #767 item 1: `codeburn compare --model-a/--model-b` (both the JSON
+// path and the TUI picker) required the canonical model id (e.g.
+// claude-opus-4-8) with no way to pass the display name shown in the picker
+// itself (e.g. "Opus 4.8"). findModelStat is the shared lookup both surfaces
+// use, reusing getShortModelName (the existing canonical -> display mapping)
+// instead of a new alias table.
+describe('findModelStat', () => {
+  it('matches the exact canonical model id', () => {
+    const project = makeProject([makeTurn('claude-opus-4-6', 0.10)])
+    const stats = aggregateModelStats([project])
+    expect(findModelStat(stats, 'claude-opus-4-6')?.model).toBe('claude-opus-4-6')
+  })
+
+  it('matches the display name, case-insensitively', () => {
+    const project = makeProject([makeTurn('claude-opus-4-6', 0.10)])
+    const stats = aggregateModelStats([project])
+    expect(findModelStat(stats, 'opus 4.6')?.model).toBe('claude-opus-4-6')
+    expect(findModelStat(stats, 'Opus 4.6')?.model).toBe('claude-opus-4-6')
+  })
+
+  it('returns undefined for an unknown model', () => {
+    const project = makeProject([makeTurn('claude-opus-4-6', 0.10)])
+    const stats = aggregateModelStats([project])
+    expect(findModelStat(stats, 'claude-opus-9-9')).toBeUndefined()
+  })
+})