Trust and data handling
What solvemax stores, refunds guarantees, and how the engine stays server-side.
This page exists because solvemax's whole premise is that a verdict slip is a record you can trust enough to show other people. That only means something if the platform itself is honest about its own limits. Every commitment below points to the mechanism that enforces it.
We never claim a result was audited when it wasn't
If the independent verification pass can't finish inside a run's time budget, the slip ships
flagged UNVERIFIED (shown as UNAUDITED on the slip itself) — never silently rendered as
CONFIRMED. This is enforced in the same component that draws the confirmation stamp: an
unverified result takes a structurally different code path to a visually different label. See
Confidence & verification.
A spent credit is never silently lost
Every run failure — a timeout, a provider outage, a stopped run, even an infrastructure blip during the refund itself — either refunds the credit immediately or leaves the run in a state the platform's automated recovery sweep will catch and finish refunding. This exact failure mode (a database hiccup during a refund silently destroying a paid credit) was found in production, fixed, and is now guarded by a dedicated regression test specifically checking that a credit is never dropped. See Refunds.
Idempotent billing — never double-charged or double-granted
Every Stripe event is claimed (via a database compare-and-swap) before it's processed, and released if processing fails so Stripe's own retry can safely re-claim it — a duplicate webhook delivery (which Stripe's own docs say can happen) can never grant credits twice for the same event.
We operate margin-safe, on purpose
solvemax refuses to run a decision whose worst-case model cost would eat too much of what that run actually earned — enforced by a cost guard checked against real, current per-model pricing before a run is priced into a plan or product. This exists so pricing stays honest rather than becoming a bet against usage patterns nobody can predict.
What we never do
- Never train on your decisions. Your decision text and the resulting slip are not used to train or fine-tune any model.
- Never make a slip public without your explicit action. Every slip is private by default; visibility only changes when you flip it. See Sharing & revisiting.
- Never expose internal error detail. Provider names, model identifiers, and stack traces never reach a client-facing error — every failure is mapped to a small, closed set of public codes. See Errors.
- Never fabricate proof. solvemax doesn't show inflated usage numbers or invented testimonials; social-proof claims on the marketing site are gated to real counts.
What "honest by default" means in practice
Every claim above is enforced by a specific, checkable code path — not a promise sitting separately from the product. If you find a place where the product's behavior doesn't match this page, that's a bug in the product, and this page is what it's being held to.