How models and budgets work

The two rules

RIGOR never raises what you authorised. If your configuration costs more than the limit you set, RIGOR refuses to start and tells you the minimum. It does not round you up, and it does not begin a run it cannot finish.

Page count is an output of coverage, never a target. A larger budget does not buy a longer report. It buys more independent verification of what the report claims. A short audit of a short paper that closes cleanly is the correct outcome.

How RIGOR chooses a model

You pick a provider, billed to your own account, and a model from that provider. RIGOR then prices a real audit of this manuscript on every model before spending a cent, and marks each one viable or not against your limit.

If the model you chose cannot start on this paper — a long manuscript on an expensive model is the usual reason — the Start button stays disabled and you are shown what that model needs and which cheaper models of the same provider would fit. RIGOR never quietly downgrades your choice.

Providers never mix: an Anthropic key only ever runs Claude models, an OpenAI key only ever runs GPT models. Because the minimum is measured against your actual manuscript, it is shown on the audit screen and not here — a figure printed on this page would be a figure about some other paper.

The providers

Within each provider the ladder runs from most expensive to most affordable. Price is not evidence of quality here: every published sample and every canary was produced on Opus 5, which is what RIGOR recommends. Fable 5 is offered and priced and has never completed a paid audit on this system.

What a spend limit buys

The limit is the most RIGOR may spend. It is not a price, and an audit that finishes early costs less. What more budget buys: more roles working on the paper, more searches behind each claim, more findings carried to a conclusion, and more independent verification of each one. Every admitted budget still produces a complete report; what a bigger one removes is the number of things the audit had to leave unresolved.

Two limits, not one. A detailed audit needs at least $20, because it owes two independent checks on every major finding. A general audit is not offered above $20: it stops when it has an answer, so the extra money would buy nothing visible. RIGOR refuses both cases before starting rather than taking the money.

Every audit is bounded in time. An audit is stopped at 30 minutes of wall clock, whatever the budget. It usually finishes well inside that; if it does not, RIGOR reports what it reached and never bills a limit it could not use.