If you ran /audit-assumptions in your coding agent, you now have a ux-assumptions.md file listing the product and UX assumptions your codebase is already betting on. This page explains what that file is worth on its own, and what you gain by bringing it somewhere that can settle the questions in it.
What you have
The audit reads your repo — its docs, its routes, its recent changes — and writes down the beliefs the product depends on but never stated. Each entry has a belief, a falsifiable question, a signal that could answer it, and a threshold. The file is yours: it runs locally, sends nothing anywhere, needs no account, and is MIT-licensed.
What the file cannot do
A written assumption is not a tested one. The file cannot watch your product, cannot tell you when a change invalidates something you assumed six weeks ago, and cannot distinguish an assumption that is holding from one that nobody has ever checked. It is a snapshot of what you believe, not evidence about whether you are right.
What changes when you import it
- Assumptions get watched. A committed entry's signal runs against your connected analytics on a schedule, and the verdict updates itself.
- Honesty is enforced. Below the minimum sample the entry reads "still gathering — n of min_n" and is never upgraded to a verdict; an event your product never emitted is reported as unmeasurable, not as a zero.
- Shipping re-arms them. When a commit touches a file an assumption depends on, that assumption reverts to untested — yesterday's answer does not survive a materially different product.
- You get a read before you have traffic. Where there is not enough data to measure anything, the agent gives a qualitative read and the hard calls escalate to a senior human. That is evidence, and it is labelled as evidence.
- Settled verdicts come back to your coding agent. Ruled verdicts export as a CLAUDE.md block, so the agent writing your code knows what has been disproven.
How to import it
Create a free workspace, open the Assumptions page, and choose "Import drafts from repo". Sightspool reads the same sources the audit did and drafts entries in the identical ux-assumptions v1 format — the file format is a shared, frozen contract, so nothing is lost in translation. Everything arrives as a draft. Committing an entry is always your decision, and only committed entries can ever be ruled.
What it costs to try
Nothing, and no card. A new workspace starts free with sample data, and exploring sample data does not start any clock. You can import your register and read it without connecting a single real source.
Where to go next
- The assumption register — how an entry becomes falsifiable, and how verdicts are ruled.
- Connecting sources — what Sightspool reads, and how your credentials are stored.
- Honesty guarantees — the rules that stop a thin read becoming a confident answer.
- Coding-agent access — asking your register questions from inside your editor.