Token count regression after v1.4.3 update: 5B → 3.6B

[xodn348]

Summary

After updating tokscale to v1.4.3, tokscale submit reported significantly fewer tokens than before the update. My profile went from ~5B tokens to 3.6B tokens, and my leaderboard ranking dropped from ~#200 to #431.

Environment

• macOS (Darwin 25.3.0, arm64)
• tokscale v1.4.3 (updated from an earlier version via npm update -g tokscale)
• Sources: Claude Code (30,531 messages), Codex CLI/omx (449 messages)

Steps to reproduce

1. Had ~5B tokens on tokscale profile before update
2. Ran npm update -g tokscale to v1.4.3
3. Ran tokscale submit
4. Profile now shows 3.6B tokens instead of ~5B

Evidence

Local token breakdown from tokscale monthly --json:

Month	Input	Output	Cache Read	Cache Write	Cost
2025-12	152K	23K	588K	0	$0.50
2026-01	4K	228	979K	127K	$0.62
2026-02	9K	896	128K	0	$0.05
2026-03	5.1M	611K	159.7M	36.9M	$295.86
2026-04	17.9M	1.7M	838.5M	173.0M	$1,214.56
Total	23.1M	2.4M	1.0B	210.0M	$1,511.58

All combined (input + output + cacheRead + cacheWrite) = 1.24B locally, but server shows 3.6B (accumulated from previous submits). Before the update, the profile showed ~5B.

It seems like the v1.4.3 update changed how local sessions are parsed, causing some older sessions to no longer be counted. When submit ran, it overwrote the server-side total with the lower number.

Expected behavior

Token count should not decrease after an update + resubmit. Either:

• The new version should parse all the same sessions the old version did
• Or submit should not overwrite/reduce previously submitted data on the server

Profile

https://tokscale.ai/u/xodn348