[BUG] Activity overview shows blank days, wrong tokens, reset streak — usage-event uploads silently failing (1p_failed_events accumulating)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

The activity overview on the home page (highlighted days, daily token counts, streak) badly undercounts real usage:

• Streak shows 2 days, but I used Claude Code on this Mac every day for 10 days (Jul 14-23).
• Heavy-usage days (Jul 16, 18, 20, 21) render as blank squares.
• Jul 17 shows ~22K tokens; local transcripts for that day total ~1.28M output tokens.

Ground truth recomputed from ~/.claude/projects/<project>/*.jsonl transcripts (UTC timestamps converted to local IST):

| Date (IST) | User msgs | Assistant msgs | Output tokens | Active hours |
|---|---|---|---|---|
| Jul 14 | 13 | 587 | 636,675 | 4 |
| Jul 15 | 24 | 207 | 176,790 | 4 |
| Jul 16 | 59 | 1,105 | 1,407,406 | 12 |
| Jul 17 | 32 | 1,501 | 1,280,387 | 13 |
| Jul 18 | 101 | 2,701 | 2,665,769 | 17 |
| Jul 19 | 70 | 2,153 | 1,521,640 | 10 |
| Jul 20 | 99 | 2,783 | 1,739,366 | 13 |
| Jul 21 | 86 | 2,407 | 1,880,328 | 16 |
| Jul 22 | 76 | 1,658 | 1,254,403 | 11 |
| Jul 23 (partial) | 13 | 442 | 384,404 | 3 |

~13M output tokens across 10 consecutive days with zero gaps, while the overview shows mostly blank days and a 2-day streak.

Likely cause: the overview is built from server-side usage events, and uploads from this machine fail silently — five 1p_failed_events.<sessionId>.<batchId>.json batches are stranded in ~/.claude/telemetry/ (Jul 18 x2, Jul 22 x2, Jul 23 x1 — the newest appeared today during normal use, so it is ongoing). Days whose batches fail render blank, tokens vanish, streak resets.

What Should Happen?

Every day with Claude Code activity highlighted in the overview, daily token counts in the right order of magnitude, and a 10-day streak.

Also: failed usage-event uploads should retry/flush eventually (or surface an error), rather than failing silently and accumulating on disk. If feasible, backfill the affected days — local transcripts hold complete ground truth.

Error Messages/Logs

$ ls -l ~/.claude/telemetry/
1p_failed_events.1d67eecd-....json   (Jul 22 15:21 IST)
1p_failed_events.5419f666-....json   (Jul 18 16:23 IST)
1p_failed_events.5419f666-....json   (Jul 22 14:14 IST)
1p_failed_events.5419f666-....json   (Jul 23 12:56 IST)
1p_failed_events.f4a95cb5-....json   (Jul 18 16:23 IST)

No visible error in the app; uploads fail silently.

Steps to Reproduce

I cannot force the upload failure, but the mismatch is directly observable on this machine:

1. Use Claude Code daily (sessions exist in ~/.claude/projects/ with timestamped transcripts every day, Jul 14-23).
2. Open the activity overview on the home page.
3. Overview shows blank squares on active days, ~22K tokens on a ~1.28M-token day, and a 2-day streak instead of 10.
4. Check ~/.claude/telemetry/ — 1p_failed_events.*.json batches from Jul 18/22/23 are stranded and never re-uploaded.

Happy to share the failed-event files or transcript-derived stats with the team.

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

Claude Code desktop app for Mac, 1.24012.1

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

Cosmetic but genuinely misleading — the overview suggests near-zero usage during a 10-17 hours/day stretch. Nothing else appears affected (sessions, transcripts, and plan limits all fine).