[BUG] Code tab Overview: "All" totals shrink over time — today's all-time count is lower than 3 months ago

Status Open

Reported on v2.1.259

Maintainer reply None cached

Activity 1 comment · opened Sep 3, 2026

Preflight Checklist

[x] I have searched existing issues and this hasn't been reported yet
[x] This is a single bug report
[x] I am using the latest version of Claude Code

What's Wrong?

This report was written by Claude (Opus 5, via Claude Code) at my request. The observations, screenshots and diagnostics are from my machine.

Desktop app → Code tab → Overview, "All" range: the summary stats do not accumulate. They rise, then fall. Same account, same machine, nothing deleted, heavy continuous usage throughout.

| Metric | 2026-06-05 | ~mid-Aug 2026 | 2026-09-03 |
|---|---|---|---|
| Sessions | 71 | 125 | 73 |
| Messages | 31,333 | 45,117 | 24,221 |
| Total tokens | 19.8M | 40.3M | 18M |
| Active days | 27 | 40 | 33 |
| Longest streak | 8d | 15d | 12d |
| Favorite model | Opus 4.7 | Fable 5 | Fable 5 |

Today's "all-time" totals are lower than they were three months ago. 18M tokens now vs 19.8M on 2026-06-05, and 24,221 messages vs 31,333 — despite my having used Claude Code heavily and near-daily for that entire period. A lifetime counter that ends a heavy quarter below where it started is not a rounding or attribution problem.

"Longest streak" independently confirms data loss. It's a maximum over all history, monotonically non-decreasing by definition. It went 8d → 15d → 12d. The 15-day streak was real and is now unreachable, so the days behind it are no longer visible to whatever computes this pane.

Not a single-tile rendering glitch — the footer line tracks the totals independently in all three ("~319× The Great Gatsby" → "~69× The Lord of the Rings" → "~290× The Great Gatsby", ≈ 19.8M → 44M → 18M), and the losses are non-proportional (sessions −42%, messages −46%, tokens −55% over the second interval).

The heatmap looks like the cause. In all three screenshots, taken three months apart, the populated region spans only the rightmost ~7–8 columns of a ~26-column grid; the left ~18 columns are always empty. The filled region never grows — it slides. That is the fingerprint of a rolling window of roughly eight weeks, presented as "All", with the summary stats computed over only the surviving subset. It also explains the shape of the numbers above: the mid-August peak captured an unusually heavy June–July stretch, which had aged out by September.

7d and 30d look stable, consistent with those windows sitting inside whatever is retained.

Long-standing: this is not a recent regression. The 2026-06-05 reading shows data that has since been lost, so the affected span is at least 2026-06-05 → 2026-09-03. I've noticed the behaviour for months and am only now reporting it with numbers. #77956 and #80537, both describing related Overview data loss, were filed mid-July 2026 — inside the same window.

The middle screenshot was pasted into a chat and not saved, so its date is approximate ("1–2 weeks ago"); the June and September readings are firmly dated. I didn't try to date the middle one from its heatmap, since the heatmap is rendered from the same aggregation this bug affects.

What Should Happen?

Under "All", Sessions / Messages / Total tokens / Active days must be non-decreasing between reads, and Longest streak must never decrease.

If there is a genuine retention limit on the underlying usage events, the range shouldn't be labelled "All" — name the actual window rather than presenting a shrinking number as a lifetime total.

Error Messages/Logs

Nothing surfaces in the UI. This machine has two never-retried usage-event batches in ~/.claude/telemetry/, the same signature as #80537:

185888  Aug 31 15:39  1p_failed_events.32086160-....json
166411  Aug 31 15:39  1p_failed_events.95847e2c-....json

That would explain undercounting, but not a previously-reported total falling. Noting it in case they're related.

Steps to Reproduce

Desktop app → Code tab → Overview → All. Record the stats and screenshot them.
Use Claude Code heavily for several weeks.
Return to the same pane and select All again.
Figures that should only ever rise have fallen, including Longest streak. Over a long enough gap the totals end up below where they started.

Local transcripts in ~/.claude/projects/ are pruned at cleanupPeriodDays (default 30), so they can't serve as independent ground truth for "All" — on this machine they reach back only to 2026-08-03.

Claude Model

Not sure / Multiple models

Is this a regression?

I don't know — I can't name a working version. Data from 2026-06-05 has since been lost, so any good version predates that.

Claude Code Version

2.1.259 (Claude Code) — Claude desktop app 1.44121.4

Platform

Anthropic API

Operating System

macOS 26.6.2

Terminal/Shell

Other (Claude desktop app, Code tab — not a terminal)

Additional Information

The three readings, in order. Note that the filled region of the heatmap covers roughly the same
rightmost ~7–8 columns in all three, three months apart, while the left ~18 columns stay empty.

1 — 2026-06-05 · 71 sessions · 31,333 messages · 19.8M tokens · 27 active days · longest streak 8d

2 — ~mid-Aug 2026 (date approximate) · 125 sessions · 45,117 messages · 40.3M tokens · 40 active days · longest streak 15d

3 — 2026-09-03 · 73 sessions · 24,221 messages · 18M tokens · 33 active days · longest streak 12d

Between 1 and 3: three months of heavy near-daily use, and the all-time totals went down.

Possibly related, none describing the shrinking "All" totals: #77956 (heatmap gaps under "All" that shift between loads; Windows; stale) · #80537 (blank days, wrong tokens, reset streak from failed telemetry uploads) · #80805 (heatmap misses active days) · #87934 (cloud session tokens excluded from Total tokens).

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗