Skip to content

Remove unreliable model detection from usage stats - #5

Merged
stevensouza merged 1 commit into
mainfrom
stevensouza/usage-stats-rework
Feb 28, 2026
Merged

stevensouza merged 1 commit into
mainfrom
stevensouza/usage-stats-rework

Conversation

@stevensouza

Copy link
Copy Markdown
Owner

Summary

The app attempted to detect which Claude model was actively used via API per-model 7-day breakdowns and velocity heuristics. This detection was unreliable — it incorrectly showed "H" (Haiku) in the menu bar even when the user primarily used Sonnet/Opus across other sessions. Since the 5-hour session limit is aggregate across all models, the model detection added complexity without accuracy.

Changes

  • Menu bar: CC·21%·H → CC·21% (removed misleading model letter)
  • Usage Pace section: Removed "Active model" display, per-model time estimates table, and budget advisor tip
  • Kept: Aggregate session/weekly/all-time velocity and time remaining estimates (inherently accurate)
  • Code removal: ~300 lines deleted including ModelTimeEstimate struct, model detection logic, per-model estimates, budget advisor, and ModelPricing enum

Testing

✅ make lint passes with no violations
✅ make bundle builds successfully
✅ Backward compatibility preserved: existing usage_samples.json with old model data still loads correctly

The app attempted to detect which Claude model was actively used via API per-model 7-day breakdowns and velocity heuristics. This detection was unreliable — it incorrectly showed "H" (Haiku) in the menu bar even when the user primarily used Sonnet/Opus across other sessions. Since the 5-hour session limit is aggregate across all models, the model detection added complexity without accuracy.

Changes:
- Menu bar: CC·21%·H → CC·21% (removed misleading model letter)
- Usage Pace section: removed "Active model", per-model time estimates, budget advisor
- Kept: aggregate session/weekly/all-time velocity and time remaining (inherently accurate)
- Removed ~300 lines: ModelTimeEstimate struct, model detection logic, per-model estimates, budget advisor, ModelPricing enum

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
@stevensouza
stevensouza merged commit 9ce3cc1 into main Feb 28, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant