[Bug]: Usage → Limits shows a Fable-only Claude reset credit on every Claude window #33509

Area

apps/server

Steps to reproduce

Version 0.1.7, commit b8c2cfa.

1. Use a Claude Pro or Team Standard login that got the Fable launch-week reset (grant launch-2610-fable-week).
2. Open Usage → Limits, or the composer usage popover.

Expected behavior

The reset credit badge appears only on the windows the grant clears. For this grant that's Session and Weekly · Fable. The all-models Weekly row has no badge.

A Fable reset should also look clearly different from a normal one, because it only clears Fable usage. A badge marked "Fable" would do it.

Hovering a badge should show a tooltip that says what the reset does, using the grant's label and the limits it clears. For example: "1 Fable reset. Clears Session and Weekly · Fable. Doesn't clear Weekly. Expires Nov 17." With mixed grants, the tooltip lists each one.

Actual behavior

Session, Weekly and Weekly · Fable all show the same banked-reset badge (ticket icon, "1"). That reads as "one reset clears everything", including the all-models weekly limit, and nothing in the app says what the Fable reset actually does. I know at least I read it that way publicly and was let down, so it could happen to others too!

Impact

Minor bug or occasional failure

Environment

Linux, desktop and web. Any Claude account that has a scoped grant.

Logs or stack traces

Raw cedar_ember grant from GET https://api.anthropic.com/api/oauth/usage?cedar_ember=1&skip_spend=1 (event props removed):

{
  "id": "launch-2610-fable-week",
  "label": "New model launch: one Fable reset for Pro and Team Standard, launch week only",
  "resets_total": 1,
  "resets_left": 1,
  "starts_at": "2026-11-09T18:00:00+00:00",
  "ends_at": "2026-11-17T07:59:00+00:00",
  "clears": ["five_hour", "seven_day_overage_included"],
  "paused": false,
  "usable_now": true,
  "use_requires_limit": false,
  "percent_used": { "five_hour": 59, "seven_day_overage_included": 3, "seven_day": 93 },
  "type": "full",
  "gives_back_weekly": true
}

The same response's limits array has the Fable bucket as {"kind": "weekly_scoped", "scope": {"model": {"display_name": "Fable"}}}, so seven_day_overage_included is the window T3 shows as seven_day_fable ("Weekly · Fable").

Supporting files

Cause

claudeResetCreditsToContract (apps/server/src/provider/claudeResetCredits.ts) decodes each grant down to id, resets_left, ends_at, paused and usable_now, so it drops clears. ServerProviderResetCredits has no field for it either. Web (UsageLimitsPooled.tsx) and mobile then read account.limits.resetCredits.availableCount for every window of the account.

Accounts holding several grants make this worse. availableCount adds up resets_left across every live grant, and redeem always claims next_grant_id. With one Fable grant and one normal grant, every row shows "2". If Anthropic picks the Fable grant as next_grant_id, clicking redeem while blocked on the all-models Weekly spends the Fable reset and leaves you blocked.

Possible fix: report credits per grant (id, label, count, expiry, cleared window ids) instead of one total. Map seven_day_overage_included to the scoped window id the server already derives from its display name. Each row then counts only the grants that clear it. Redeeming from a row claims a grant that clears that row, since the claim endpoint already takes a grant_id. A grant without clears keeps today's behavior, so Codex and older grants don't change.

One open question: gives_back_weekly: true sits next to a clears list that leaves out seven_day. I don't know whether redeeming also returns some of the all-models weekly. If it does, the Weekly row should show that amount, not a full reset.

Workaround

None in the UI. Hovering the badge doesn't say which limits the credit clears.