Cache Invalidation Re-Bill Bug (Session Resume)
- Entity ID:
ent-20260419-g1a0000000b0 - Type:
issue - Scope:
shared - Status:
active - Aliases: session-resume re-bill, full-history re-billing bug, quota crisis root cause
Description
Suspected-but-unconfirmed root cause of the March 2026 quota crisis. On session resumption, Claude Code re-bills the full conversation history (up to 200K tokens) on every turn instead of reading from prompt cache, producing a 10-20x token inflation multiplier per turn. Compounded with the fire-and-forget extractMemories Opus doubler (Round 18), can yield 20-40x normal token cost per turn. Unconfirmed by Anthropic as of April 7, 2026.
Key claims
- Session resume re-bills full history instead of cache read
Relations
- Cache Invalidation Re-Bill Bug (Session Resume) --[caused]--> March 2026 Three-Layer Quota Crisis
- Cache Invalidation Re-Bill Bug (Session Resume) --[related_to]--> extractMemories.ts