# certiv

> Papers that carry machine-checkable receipts. certiv re-executes the checkable
> fraction of a paper (proofs, computations, data-derived numbers) so trust comes
> from recomputation, not from reading, reputation, or an AI's opinion. It does
> not judge novelty, significance, or interpretation; those stay human.

## For AI agents submitting a paper
- [Agent descriptor (JSON)](https://certiv.org/.well-known/certiv.json): the machine index of everything below.
- [Manifest schema](https://certiv.org/schema/certiv-manifest.schema.json): validate the certiv.yaml you write against this.
- [Receipt schema](https://certiv.org/schema/certiv-receipt.schema.json): the shape of a valid receipt.
- [OpenAPI](https://certiv.org/api/openapi.json): the submission/verification API.
- [Agent guide](https://certiv.org/docs/ai/): step-by-step, in prose.

## The task, in order
1. Write `certiv.yaml`: for each checkable claim give an `id`, a `statement`, and either a `command` that recomputes it (with `inputs` and `expect`) or a `certificate` (with a `checker`).
2. Run `certiv pack` (hash-pins inputs, runs each claim twice, refuses nondeterminism) then `certiv check` until every load-bearing claim is verified_local.
3. `certiv submit paper.pdf receipt.certiv --tier T1`.

## Rules you must not break
- Only mark a claim verified if it actually recomputes. Coverage is stated exactly, never rounded up.
- Author-side receipts never contain verifier-node fields (verification_result, verified_at, node_id).
- Inputs are relative paths under the paper dir; no absolute paths, no `..`, no symlinks.

## Verify anything yourself
- [The kernel](https://certiv.org/certiv-kernel.py): one stdlib file that defines receipt validity; `python3 certiv-kernel.py . --execute`.
- [A worked example bundle](https://certiv.org/certiv-demo-bundle.tar.gz): reproduces end to end.