> ## Documentation Index
> Fetch the complete documentation index at: https://haiprotocol.com/llms.txt
> Use this file to discover all available pages before exploring further.

# One bounded counter effect

> Exercise execution refusal, launch fencing and outcome retries with mock state.

`examples/http/counter.ts` exports `runCounter`, an executor fixture for exactly one `counter.increment` action with amount one. It accepts only the fixed action, context and policy commitments used by the tests. It is not a general-purpose runner.

After ordinary human authorisation, it obtains an exclusive claim and fresh admission, verifies the original signed records and independent checkpoint acceptance, and creates a durable exclusive fence in a private per-request directory. Only then can it update the counter once. Initial reads count towards the same execution window.

Reinvocation uses a saved result to retry outcome reporting without repeating the effect. A fence without a result is uncertain and must not be relaunched, even if an external operation appears to have failed. Do not delete or share the directory to bypass that rule.

The local regression tests run normal HTTP confirmation with isolated PostgreSQL, OIDC and explicitly non-production storage fixtures:

```sh theme={null}
node --import tsx --test tests/http.test.ts tests/counter-refusals.test.ts tests/counter-timing.test.ts
```

They cover the successful fixed effect, unconfirmed/review-only authority, altered bindings, pending anchoring, revocation, expiry and delays during the initial counter read. They do not validate real provider storage, credentials, Plasm dispatch or production recovery.

Executors must supply a genuine independent acceptance verifier; replacing it with a no-op invalidates the demonstration's safety checks. Read the [execution contract](/protocol/execution) and [SDK verification responsibilities](/sdk/execution) before adapting any part of this fixture.