anthropic sometimes zeroes pro/max usage counters mid-week. often it is quiet compensation after an incident or a metering bug. sometimes it gets a post on x. often it does not. either way there is no in-app notice, no email, nothing in the status-page incident text.
this page watches one heavy reference account (the canary), whose official meter is polled every ~30 minutes, and answers from that. a mid-week reset looks like this: used% plunges to zero while the week keeps elapsing and the reset date stays put.
the canary cannot see your account, and resets are sometimes cohort-scoped. check yours from a terminal. the command reads your own local claude code token and asks anthropic directly.
TOKEN=$(security find-generic-password -s "Claude Code-credentials" -w | python3 -c "import sys,json;print(json.load(sys.stdin)['claudeAiOauth']['accessToken'])") curl -s https://api.anthropic.com/api/oauth/usage \ -H "Authorization: Bearer $TOKEN" -H "anthropic-beta: oauth-2025-04-20" \ | python3 -m json.tool
TOKEN=$(python3 -c "import json,os;print(json.load(open(os.path.expanduser( '~/.claude/.credentials.json')))['claudeAiOauth']['accessToken'])") curl -s https://api.anthropic.com/api/oauth/usage \ -H "Authorization: Bearer $TOKEN" -H "anthropic-beta: oauth-2025-04-20" \ | python3 -m json.tool
read seven_day.utilization. if it sits far below where your week's work should have it, and seven_day.resets_at has not moved, your counter was reset too.
never paste your token into a website.
people file these as bugs (claude-code #52497, #49616). it is not a bug.
a compensation reset drops used% while elapsed time and resets_at stay put. the normal weekly roll zeroes both together. a re-anchor moves the window itself. a single weird reading between two normal ones is metering flicker, and does not count.
incident context comes from the claude status page. the resets themselves do not appear in its incident text.