Publish Max 5x/20x weekly allowances and add a usage history: measured 20x week ≈ 1.6× the 5x week, not 4×

Environment: Claude Code 2.1.286, desktop app (Code tab), Windows 11. Max 20x in late September 2026, Max 5x since 2026-10-01. Claude Code is 95%+ of my usage.

Method: I priced every call in my local Claude Code transcripts at API list prices ("API-equivalent $"). I then matched that against the usage page and against the your session/weekly limit resets … messages, each of which marks a window at exactly 100%.

Findings

1. Max 20x, one full 5-hour session (2026-09-24): ≈ $508. The usage page then showed the week at 27%.
2. Max 20x, weekly limit hit on 2026-09-30 after ≈ $1,827 since the weekly reset. That is about 3.6 full sessions per week.
3. Max 5x, one full 5-hour session (2026-10-03): ≈ $158. The week stood at 18% after ≈ $211, so the week is ≈ $1,170 (extrapolated). That is about 7.4 full sessions per week.
4. Ratios 20x / 5x:

| | 20x / 5x |
|---|---|
| 5-hour session | ≈ 3.2× (advertised 4×) |
| Week | ≈ 1.6× (2× at most) |
| Price | 2× |

1. The yardstick tracks the meter. It predicted the Fable share of the week (43.5% vs ≈ 44% shown) and a live 5-hour reading (4.5% vs 4% shown) to within a point.

So "20x" describes the 5-hour window only. Per dollar, 20x buys less weekly usage than 5x.

Requests

1. Publish the weekly allowance of each plan in any unit, or at least the 20x : 5x weekly ratio, next to the session multiplier on the plans page. Today a buyer reasonably reads "20x" as applying to the week, and it does not.
2. Add a usage history to the usage page, per week and per 5-hour window. Only the current bars exist now. The only way to learn what a plan gave you last week is to reverse-engineer your own transcripts, as above.

Related: #57699 and #72872 (weekly cap not scaling with the session multiplier; closed and locked), #52135, #26271. This is not the upgrade-not-applied bug in #79773 / #76093: here the 20x session cap clearly applied.