Answer Radar
Usage, Credits and Limits
Every audit spends credits, and the usage view is where you check the account's position before starting work. It shows what has been reserved and consumed today, what is left, the caps that bound a run, and the size limits a run may ask for. Everything is in credits; one credit is one measured sample, one question asked of one engine.
Quick Answer
answer_usage_get returns reserved and consumed credits today, the remaining daily allowance, the per-run cost cap, whether the operator kill switch is engaged, and the per-run size limits. Pass a project id to also get that project's lifetime consumption. A null cap means not configured, not unlimited and not zero. can_start_run allowed true means nothing visible refuses a run, not that one is guaranteed.
- Configured per-run limits on the day: 50 questions, 5 samples per question, 200 samples.
- Charged credits are a true count of measured samples, never a conversion from spend.
- A platform-wide daily cap that is not visible to you also applies.
On this page
What the usage view reports
Reserved today: credits held for runs in progress. Consumed today: credits measured and charged. Remaining daily allowance: what the account may still spend before the daily cap refuses a reservation. Per-run cost cap: the largest estimate a single run may carry. Kill switch: whether the operator has paused all measurement. Per-run size limits: the maximum questions per run, samples per question and samples per run a request may ask for. With a project id, the project's lifetime consumption is added.
How credits are counted
One credit is one measured sample. Reservations convert exactly, a flat rate times the sample count, so an estimate divides back to the count it came from. Charges do not convert and are not made to: real cost varies by engine, so charged credits are read as the count of measured samples from the usage ledger, which is what keeps the unit's one promise. Where a ledger row lacks that count, the total is reported with its own completeness rather than silently summed short.
Customers see credits and engine names. Nothing in the product shows a currency figure or a model id for a measurement.
Reading the caps honestly
A cap that reads null is not configured. That does not mean unlimited and it does not mean zero; it means that particular ceiling is not set, and the others still apply. can_start_run allowed true means nothing visible in this view refuses a run. It is not a guarantee: the per-run cap is checked against the estimate at start, and a platform-wide daily cap you cannot see also applies. When a run is refused, the run row reads budget_exceeded and the activity list records the denied reservation with its reason.
Planning a month of measurement
Multiply approved questions by enabled engines by samples per question for one audit, then by the number of audits in the month. Twenty questions on four engines with one sample each is 80 credits an audit; weekly is about 350 a month. Keep the prompt set stable so the series is comparable, and raise samples per question rather than adding prompts if you want tighter resolution on the rates you already track.
- Check usage before a scheduled slot if the account has been busy.
- A partial run charges only the samples it measured.
- Archiving a project stops its scheduled spend without losing its history.
FAQ
Where do I buy more credits?
Credits are part of the Compete plan's allowance. The pricing page describes the plan; usage shows what is left today.
Why was my run refused when usage said I could start one?
Either the estimate exceeded the per-run cap, or the platform-wide daily cap, which the view cannot show, refused the reservation. The activity list records the denial and its reason.
Do failed samples cost credits?
No. A failed sample is not a measured sample. The run row shows credits reserved and credits measured separately.
Related help pages
Answer Radar
The Audit Run and What It Costs in Credits
What Run audit does: the estimate, the reservation, the fan-out per question per engine per sample, the run states, and why one credit is one measured sample.
Answer Radar
Scheduling Recurring Audits
How to set a weekly or monthly audit in your timezone, why enabling does not run one now, what pausing keeps, and how to find out why a scheduled audit did.
Answer Radar
Creating an Answer Radar Project
What each field on the new project form does, why only the domain is required, how idempotent creation protects a retry, and how archiving reverses safely.
Answer Radar
What Answer Radar Does Not Do
The honest list: no prompt volume, no traffic attribution, no causal claims, no consumer-app reproduction, no written briefs, no publishing, no model ids,.