/usage: attribute extra-credit draws to the limit that triggered them (per-gate overage meters)

Status Open
Maintainer reply None cached
Activity 1 comment · opened Aug 6, 2026

Problem

The one-sentence version: I was billed for crossing a threshold whose size is disclosed nowhere. Silent billing from an unannounced transition would be bad; over an undisclosed budget it becomes unverifiable in principle — no amount of user diligence could predict or audit the charge.

The mechanics: extra-usage credits can be drawn by any limit (the 5-hour window, weekly per-model, weekly all-models), but /usage shows only an aggregate credit draw with no attribution to which gate produced it.

Real sequence that prompted this: I discovered a $38.83 draw from a $75 balance while BOTH weekly bars sat near 60%. Nothing on the panel could explain which limit billed it (it turned out to be the 5-hour window tripping during agent bursts, billing silently until each window rolled). It took three rounds of guessing to reconstruct the cause.

Three specific gaps:

  1. No per-gate attribution. The aggregate credit line can't answer "which limit billed me, when, and how much."
  2. No transition notice. Before holding a credit balance, a hit limit is a hard wall ("come back after X"). Once credits exist, the same limit silently becomes a meter — work continues and bills, with no cue at the moment of crossing. Users who learned "limits stop work" have no reason to expect "limits now cost money."
  3. The "5-hour limit" name implies time, but the meter is tokens — and the token budget itself is undisclosed. A parallel-agent burst can exhaust the window's budget in under an hour, which reads as "exceeding 5 hours of work in 47 minutes." The label hides the unit, and the panel shows a percentage of a number users are never told (see comment below).

Proposal

Each limit row gets a linked overage line showing time-in-overage and cost:

Extra credits: ENABLED
5-hour window   ████████████ 100%
  ↳ exceeded: 17 min · $4.39
Weekly · Fable  ████████░░░░  63%
Weekly · all    ████████░░░░  60%

Plus: a one-time visible notice when a session first crosses from quota into credit draw; a unit-honest label for the 5-hour meter; and token counts (used/budget) on every limit row so the panel reconciles against the public price sheet.

Billing that can't be traced to its trigger — over thresholds that aren't disclosed — erodes trust in a product I otherwise rely on daily. Unambiguous cost attribution is cheap goodwill.

🤖 Filed with Claude Code on the user's request

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗