pekth · Sep 27, 2026 · Sep 27, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -4,6 +4,18 @@ All notable changes to this project are documented here. The format is based on
 [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and this project
 follows [Semantic Versioning](https://semver.org/).
 
+## [Unreleased]
+
+### Fixed
+
+- **Pace displays stop hiding a deficit once the burn cools.** The quota bar,
+  side-notch banner and window rows, and the menu-bar ETA chip now report the
+  window's own pace, so a window ahead of the even-burn line keeps its
+  projected run-out — a weekly at 43% with 6 days left — instead of flipping to
+  "on pace" 30 minutes after the last local session. The present-tense
+  "burning fast" failover nudge, pace alerts, and `meterusage json` pacing stay
+  gated on a current burn, so the stale-nudge fix is unchanged.
+
 ## [0.2.38] - 2026-09-26
 
 ### Added

diff --git a/README.md b/README.md
@@ -99,14 +99,14 @@ Toggle providers on or off, choose refresh cadence, and switch themes:
 
 ## 🚀 Key Features
 
-* **Ambient Time-to-Empty** — Dynamic ETA calculations (`~47m left at current pace` / `Paced to last until reset`) directly in the menu bar and side notch, warning you of rapid burn cliffs before limits are reached. Pacing claims are present tense: a window whose burn has gone quiet reports the reset countdown instead of a burn alarm.
+* **Ambient Time-to-Empty** — Dynamic ETA calculations (`~47m left at current pace` / `Paced to last until reset`) directly in the menu bar and side notch, warning you of rapid burn cliffs before limits are reached. The bar, banner, and menu-bar chip report the window's own pace, so a deficit keeps its projected run-out after the burst has cooled; only the present-tense "burning fast" nudge and pace alerts require a current burn.
 * **Window Burn Attribution** — Explains *"Where did my tokens go?"* by decomposing the last 7 days of token consumption by project, model, and message turns across all token-bearing providers.
 * **Context Waste & Cache Hints** — Diagnostic metadata highlighting cache hit rate %, average tokens per turn, and long-chat flags ($\ge 10$ turns or $\ge 100\text{k}$ tokens) to curb silent context waste.
 * **Durable Daily History Store** — Preserves daily token tallies, estimated spend, and peak window utilization in a durable local database (`~/Library/Application Support/MeterUsage/durable-daily-history.json`) that survives CLI transcript pruning.
 * **Unified AI Coding Strip** — High-level summary card in the popover showing all AI coding today (tokens, weekly volume, and estimated USD spend) across all active providers.
 * **Floating Side Notch HUD** — Fixed-black, hardware-like collapsible strip. Hovering a ring expands a docked detail card with smooth spring animations. Features auto-flip positioning (switches left/right depending on screen position). The strip stays anchored to its parked corner while cards change, so provider switches never shift it.
 * **Cross-Provider Headroom Failover** — Instant suggestions when a provider is burning fast or near exhaustion, identifying which alternative model has headroom available. "Burning fast" requires a current burn; a window that is merely near its limit says so in its own words.
-* **Agent Budget API** — Run `meterusage json` to export machine-readable quota telemetry, burn velocity, pacing status, and seconds-to-empty for autonomous AI agents. Pacing is gated on burn recency the same way the UI is.
+* **Agent Budget API** — Run `meterusage json` to export machine-readable quota telemetry, burn velocity, pacing status, and seconds-to-empty for autonomous AI agents. Pacing in the report is gated on burn recency, matching the failover nudge and pace alerts.
 * **26-Week Activity Heatmaps** — GitHub-style activity matrix inside Codex and Claude cards with Day, Week, or Cumulative views, accompanied by 7-day sparklines.
 * **Opt-In Pacing Alerts** — Native macOS notifications when an active window crosses critical burn velocity or drops below 30 minutes to empty. Pace alerts fire only on a current burn; threshold alerts (80%/95%) remain state-based.
 * **Share screenshot**: The share button on each provider's usage card (side notch panel detail card) shares a sharp 2x image of the panel through macOS share services, or saves it for X and other apps.

diff --git a/Sources/MeterUsage/Views/MenuBarLabel.swift b/Sources/MeterUsage/Views/MenuBarLabel.swift
@@ -159,14 +159,10 @@ struct MenuBarLabel: View {
             }
 
             if let window {
-                // The effective pace demotes a deficit whose burn has gone
-                // quiet, so the ambient chip shows a burn ETA only while the
-                // provider is actually burning; a stale window falls through
-                // to the honest reset countdown (or no chip at all).
-                let pace = window.pace(now: coordinator.clock)?.effective(
-                    lastBurn: BurnRecency.lastBurn(
-                        of: coordinator.activities[slot]?.value?.sessions ?? []),
-                    now: coordinator.clock)
+                // The chip reports the window's own pace: a deficit shows the
+                // projected exhaustion while the window is ahead of the
+                // even-burn line, however long ago the burst that caused it.
+                let pace = window.pace(now: coordinator.clock)
                 let showAmbient = pace?.shouldShowAmbientETA(resetsAt: window.resetsAt, now: coordinator.clock) ?? false
                 let eta = showAmbient ? pace?.etaText(resetsAt: window.resetsAt, now: coordinator.clock, short: true) : nil
                 let isDeficit = pace?.status.isDeficit ?? false

diff --git a/Sources/MeterUsage/Views/PopoverRoot.swift b/Sources/MeterUsage/Views/PopoverRoot.swift
@@ -364,9 +364,7 @@ struct PopoverRoot: View {
                         // pass nothing and render unchanged. Each account
                         // slot shades its own activity.
                         heatmapDaily: heatmapDaily(for: slot),
-                        heatmapIntensity: slot.provider == .codex ? .sessions : .tokens,
-                        lastBurn: BurnRecency.lastBurn(
-                            of: coordinator.activities[slot]?.value?.sessions ?? [])
+                        heatmapIntensity: slot.provider == .codex ? .sessions : .tokens
                     )
                 }
             }

diff --git a/Sources/MeterUsage/Views/QuotaSection.swift b/Sources/MeterUsage/Views/QuotaSection.swift
@@ -35,10 +35,6 @@ struct QuotaSection: View {
     /// Codex has no token ledger and shades by sessions; token-bearing
     /// sources (Claude) shade by tokens.
     var heatmapIntensity: HeatmapView.Intensity = .tokens
-    /// The provider's most recent observable burn. Gates the pacing label:
-    /// a pace deficit without a current burn is the window's past, not a
-    /// present-tense "burning fast" (see `BurnRecency`).
-    var lastBurn: Date? = nil
 
     @State private var resetPrompt: ResetPrompt?
     @State private var consumingResetID: String?
@@ -54,8 +50,7 @@ struct QuotaSection: View {
         now: Date,
         onUseReset: ((String) async throws -> Void)? = nil,
         heatmapDaily: [DailyActivity] = [],
-        heatmapIntensity: HeatmapView.Intensity = .tokens,
-        lastBurn: Date? = nil
+        heatmapIntensity: HeatmapView.Intensity = .tokens
     ) {
         self.slot = slot
         self.state = state
@@ -64,7 +59,6 @@ struct QuotaSection: View {
         self.onUseReset = onUseReset
         self.heatmapDaily = heatmapDaily
         self.heatmapIntensity = heatmapIntensity
-        self.lastBurn = lastBurn
     }
 
     var body: some View {
@@ -222,7 +216,7 @@ struct QuotaSection: View {
                                     .foregroundColor(MU.text)
                             }
                             ForEach(Array(group.windows.enumerated()), id: \.offset) { _, window in
-                                WindowRow(provider: provider, window: window, now: now, lastBurn: lastBurn)
+                                WindowRow(provider: provider, window: window, now: now)
                             }
                         }
                     }
@@ -352,8 +346,6 @@ private struct WindowRow: View {
     let provider: Provider
     let window: QuotaWindow
     let now: Date
-    /// Gates the pacing label on burn recency (see `QuotaSection.lastBurn`).
-    let lastBurn: Date?
 
     @AppStorage(PrefKey.showPacingBurnRate) private var showPacingBurnRate: Bool = true
 
@@ -391,10 +383,12 @@ private struct WindowRow: View {
                         .foregroundColor(MU.textTertiary)
                         .lineLimit(1)
                         .minimumScaleFactor(0.7)
-                    // Effective pace: a deficit whose burn went quiet reads
-                    // as on-pace, so the label never calls an idle day
-                    // "burning fast".
-                    if showPacingBurnRate, let pace = window.pace(now: now)?.effective(lastBurn: lastBurn, now: now) {
+                    // The bar reports the window's pace as its shape: a
+                    // deficit means the window is ahead of where even burn
+                    // would put it, whatever the last burn recency. The
+                    // present-tense "burning fast" claim (and its nudge) is
+                    // what recency gates, not this figure.
+                    if showPacingBurnRate, let pace = window.pace(now: now) {
                         Text("·")
                             .font(.muCaption)
                             .foregroundColor(MU.textTertiary)

diff --git a/Sources/MeterUsage/Views/SideNotchPanelView.swift b/Sources/MeterUsage/Views/SideNotchPanelView.swift
@@ -552,14 +552,12 @@ struct SideNotchPanelView: View {
                 StatusBadge(severity: status.severity)
             }
 
-            // Ambient Time-To-Empty banner. The effective pace demotes a
-            // deficit whose burn has gone quiet, so a stale window shows the
-            // reset countdown in the calm tint instead of a burn alarm.
+            // Ambient Time-To-Empty banner. It reports the window's own pace:
+            // a deficit empties before reset at that shape even when the burn
+            // has since gone quiet. Only the present-tense "burning fast"
+            // nudge and alerts are gated on burn recency.
             if let headline = slot.provider.headlineWindow(from: Self.effectiveWindows(for: slot.provider, quota: quota)),
-               let pace = headline.pace(now: coordinator.clock)?.effective(
-                    lastBurn: BurnRecency.lastBurn(
-                        of: coordinator.activities[slot]?.value?.sessions ?? []),
-                    now: coordinator.clock),
+               let pace = headline.pace(now: coordinator.clock),
                let etaText = pace.etaText(resetsAt: headline.resetsAt, now: coordinator.clock) {
                 HStack(spacing: 8) {
                     VStack(alignment: .leading, spacing: 2) {
@@ -780,14 +778,11 @@ struct SideNotchPanelView: View {
 
     @ViewBuilder
     private func windowRow(window: QuotaWindow, quota: ProviderQuota?, provider: Provider, slot: ProviderSlot) -> some View {
-        // Effective pace: a deficit without a current burn reads as on-pace,
-        // never as a "burning fast" alarm for a burst that already cooled.
-        // Burn evidence is per account slot.
+        // The row reports the window's own pace: a deficit means the window
+        // is ahead of the even-burn line, and it carries the projected
+        // exhaustion even when the burst that caused it has gone quiet.
         let pace = showPacingBurnRate
-            ? window.pace(now: coordinator.clock)?.effective(
-                lastBurn: BurnRecency.lastBurn(
-                    of: coordinator.activities[slot]?.value?.sessions ?? []),
-                now: coordinator.clock)
+            ? window.pace(now: coordinator.clock)
             : nil
         let figure = Self.windowFigure(window: window, quota: quota, provider: provider)
         VStack(alignment: .leading, spacing: 4) {
@@ -1356,7 +1351,6 @@ struct SideNotchPanelView: View {
         quotas: [ProviderSlot: Loaded<ProviderQuota>],
         statuses: [Provider: Loaded<ServiceStatus>],
         archivedQuotas: [ProviderSlot: ProviderQuota] = [:],
-        lastBurn: [ProviderSlot: Date] = [:],
         now: Date = Date()
     ) -> [Entry] {
         // Additional accounts number from 2 within their tool, in stable
@@ -1378,12 +1372,9 @@ struct SideNotchPanelView: View {
                 // otherwise.
                 markTint = providerColor(slot.provider)
             }
-            // Effective pace: a deficit without a current burn is demoted to
-            // on-pace, so the strip never reports "burning fast" from a
-            // stale window and the ETA chip carries the honest reset
-            // countdown instead of a projected exhaustion that already ended.
-            // Burn evidence is per account slot.
-            let pace = window.pace(now: now)?.effective(lastBurn: lastBurn[slot], now: now)
+            // The strip reports the window's own pace: a deficit carries the
+            // projected exhaustion chip whatever the last burn recency.
+            let pace = window.pace(now: now)
             let showAmbient = pace?.shouldShowAmbientETA(resetsAt: window.resetsAt, now: now) ?? false
             let eta = showAmbient ? pace?.etaText(resetsAt: window.resetsAt, now: now, short: true) : nil
             let isDeficit = pace?.status.isDeficit ?? false
@@ -1414,7 +1405,6 @@ struct SideNotchPanelView: View {
             quotas: coordinator.quotas,
             statuses: coordinator.statuses,
             archivedQuotas: coordinator.archivedQuotas,
-            lastBurn: coordinator.lastBurnBySlot,
             now: coordinator.clock
         )
     }

diff --git a/Tests/MeterUsageTests/SideNotchPanelTests.swift b/Tests/MeterUsageTests/SideNotchPanelTests.swift
@@ -249,6 +249,27 @@ final class SideNotchPanelTests: XCTestCase {
 
     // MARK: - Entries
 
+    @MainActor
+    func testEntriesKeepWeeklyDeficitWithoutBurnEvidence() async throws {
+        // A fresh weekly window at 43% with 6d 13h left is far ahead of pace.
+        // The strip reports the deficit and its projected run-out even with no
+        // local session to prove a current burn — only the present-tense
+        // nudge and pace alerts gate on burn recency.
+        let coordinator = try await Self.coordinator(quotas: [
+            (Provider.codex, [("Weekly", 43.0, 6.0 * 86_400 + 13.0 * 3_600)]),
+        ])
+
+        let entries = SideNotchPanelView.entries(
+            menuBarSlots: coordinator.menuBarSlots,
+            quotas: coordinator.quotas,
+            statuses: coordinator.statuses
+        )
+
+        XCTAssertEqual(entries.count, 1)
+        XCTAssertTrue(entries[0].isDeficit)
+        XCTAssertNotNil(entries[0].etaText)
+    }
+
     @MainActor
     func testEntriesTakeHeadlineWindowInMenuBarOrder() async throws {
         let coordinator = try await Self.coordinator(quotas: [

diff --git a/docs/KB.md b/docs/KB.md
@@ -18,7 +18,7 @@ Last verified: 2026-09-26
 - `CHANGELOG.md` records version 0.2.38 as the latest repository release entry, with multi-account meters for Codex and Claude (ADR 0005–0006) dated 2026-09-26, following the 0.2.37 incident.io outage-vocabulary fix and identity-tinted provider marks. This is repository release-note state, not proof of a published release.
 - `CONTRIBUTING.md` requires focused changes, synthetic fixtures, and `swift build`, `swift test`, and `Scripts/make-app.sh` before a code pull request.
 - The side notch panel is anchored by the strip's top-right corner and uses whole-point frames, so switching providers never moves the strip. `docs/SIDE-NOTCH.md` describes the states, geometry invariants, and required evidence. ADR 0004 records the decision.
-- Pace claims are gated on burn recency (`BurnRecency` in `Sources/MeterUsage/Models/UsageModels.swift`): a quota window's shape can hold a pace deficit long after the burst that caused it, so "burning fast" requires the provider to have burned within a 30-minute quiet period. Surfaces consume `QuotaPace.effective(lastBurn:now:)`, which demotes a burn-quiet deficit to on-pace. A headline window at 80%+ without a current burn renders as "near its limit", not "burning fast". Providers without a local session store have no burn evidence: they keep the 80%/95% threshold alerts but never raise pace alerts.
+- Window-shape displays report the raw pace (`QuotaPace.pace(now:)`): a quota bar, side-notch banner/row, or menu-bar chip keeps a deficit's projected run-out even when the burn that caused it has gone quiet. Present-tense claims are gated on burn recency (`BurnRecency` in `Sources/MeterUsage/Models/UsageModels.swift`): the failover nudge, pace alerts, and the machine report consume `QuotaPace.effective(lastBurn:now:)`, which demotes a burn-quiet deficit to on-pace, and require the provider to have burned within a 30-minute quiet period. A headline window at 80%+ without a current burn renders as "near its limit", not "burning fast". Providers without a local session store have no burn evidence: they keep the 80%/95% threshold alerts but never raise pace alerts.
 - Burn attribution renders only measured burn (`BurnAttributionCalculator.calculate` in `Sources/MeterUsage/Services/BurnAttributionCalculator.swift`): sessions without a token ledger (for example Codex realtime sessions whose rollout records no `token_count` event) are excluded, and a scope with no token-bearing sessions yields no breakdown, so the burn card hides instead of showing all-zero rows.
 - GitHub Actions runs `swift build` and `swift test` on macOS 15 for pushes to `main` and pull requests. This repository fact does not prove that a workflow run has passed.
 - The app preserves unreadable durable history and suspends history writes until relaunch after the file is repaired. Sanitized history load and save failures are visible in Copy diagnostics.