Usage/Models token chart skips days despite real recorded session activity, and shows a phantom spike on the earliest visible day

Status Open
Reported on v2.1.217
Maintainer reply None cached
Activity 0 comments · opened Jul 30, 2026

App: Claude desktop app, Claude Code panel (version 2.1.217/219)

Bug 1 — missing days: The Usage → Models chart (both "All" and "7d" views) shows no bar for several consecutive days despite confirmed local session activity.

  • "7d" view: y-axis rescales to ~400k max, yet shows only two bars — Jul 23 and Jul 29 — with Jul 24-28 rendered as empty space.
  • Local session logs (~/.claude/projects/*/*.jsonl) show real activity on every missing day:

| Date | Messages | Output tokens | Total incl. cache |
|---|---|---|---|
| Jul 24 | 952 | 1.42M | 606.9M |
| Jul 25 | 220 | 175k | 27.7M |
| Jul 26 | 140 | 344k | 20.5M |
| Jul 27 | 489 | 620k | 215.9M |
| Jul 28 | 74 | 98k | 52.5M |

Bug 2 — phantom spike on the leftmost day: The "All" view's earliest bar (Jul 2) shows ~10-12M tokens across Opus 4.8, Fable 5, Sonnet 5, and Opus 5, but there is zero corresponding local session activity on that date anywhere in local logs or the app's own session index — the earliest recorded session on this machine is Jul 3. Jul 2 is exactly 28 days before the date this was observed (Jul 30), suggesting "All" is a fixed 28-day window and the leftmost bucket is acting as a catch-all rather than showing real per-day data.

Bug 2a — proof the Jul 2 data is misattributed, not just missing locally: The Jul 2 breakdown attributes 630.1k tokens to the Fable 5 model. Searching all local session logs, the earliest any Fable-model message appears is Jul 5, 18:10 UTC — three days later. There is no local record of Fable being used before Jul 5, making it provably impossible for genuine Fable usage to have occurred on Jul 2. This confirms the leftmost bucket is aggregating/misattributing data from later dates rather than reflecting actual Jul 2 activity.

Additional oddity: The 7d view's Jul 29 bar also has zero matching local session log entries, suggesting the chart's data source isn't reconciled with the local transcript store.

Ruled out: An unrelated local git-config issue (accidental repo at the home-directory level) was found and fixed Jul 22, predating the gap, with no cwd/path mismatches in any session metadata — not the cause.

Expected: Each day should render a bar proportional to that day's actual usage; no usage should be silently dropped or misattributed to an unrelated day.

View original on GitHub ↗