You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Buyer integrations can read GET /v1/usage/admission for cumulative accepted token
counters and the current SHELL quota without rebuilding historical reports. Exact
decimal strings preserve large totals; complete distinguishes confirmed totals
from incomplete history. The snapshot does not reserve capacity or authorize a request.
Fixes
dexdo seller retries a SELL rejected for insufficient contract SHELL reserve
after that contract is topped up. It verifies the original rejection and the
current contract and order state before retrying, preserves the rejected
attempt, and keeps waiting when the previous submission outcome is unknown.
Inference accepts additional provider cache metadata instead of failing with
HTTP 502. Supported cache read and write counts are still excluded from billing
exactly once; unrecognized fields stay in the response and do not change billing.
Cache reported only in an unsupported form can remain in the billable total.
dexdo note deploy reports confirmed insufficient wallet funds instead of an
unknown outcome, including when the wallet converted SHELL to its own gas before
the voucher send failed. If the receipt proves no voucher was sent, fund the
wallet and repeat the same command with the same pool path.
Temporary capacity refusals now include retry timing: busy deals suggest one second, while
blockchain-limited deals report the expected wait for the next claim and confirmation.
The onchain_deal_limit error code and dexdo-limit header identify that limit, with a
matching buyer Grafana panel. Retry hints are minimum waits, not admission guarantees.
Provider retry headers remain unchanged. A temporary provider HTTP 429 without a retry hint
now suggests one second; a milliseconds-only hint also supplies rounded-up seconds.
Explicit billing and quota refusals do not receive an invented delay.
Buyer HTTP and seller RPC duration histograms use finer buckets for long requests,
improving percentile estimates in Grafana. Historical samples retain their original buckets.