# Examples

Worked examples live in the repository-level
[examples](https://github.com/sashakolpakov/bayesilisk/tree/main/examples)
folder.

Coding agents should ingest
[examples/connector-agent-contract.json](https://github.com/sashakolpakov/bayesilisk/blob/main/examples/connector-agent-contract.json)
before writing a connector. It lists the required source-context fields,
observed-evidence fields, allowed agent steps, and non-negotiable boundaries.

## ABAG Connector Context

The [ABAG example](https://github.com/sashakolpakov/bayesilisk/blob/main/examples/abag-action-graph-context.json)
shows a source-only Abstract Bayesilisk Action Graph. It maps concrete connector
actions such as `create-invite`, `revoke-invite`, and `accept-invite-route` onto
universal typed tokens such as `identifier.invitation_token` and
`state.revoked`. Bayesilisk can compose a reusable revoked-token replay motif
from that graph while the connector remains responsible for executing concrete
app fixtures and returning observed evidence.

## Cal.com Connector Evidence

The [Cal.com example](https://github.com/sashakolpakov/bayesilisk/tree/main/examples/calcom)
shows Bayesilisk used with its general core and an app-specific connector. The
connector follows the documented contract: it supplies source facts, explicit
proposal rules, connector actions, expected statuses, and observed local
evidence. Bayesilisk expands only those supplied rules, then verifies
expected-versus-observed facts.

Captured evidence is included for Cal.com commit
`180ede28f0bddf2738933a6e60a8e80f6116d7da` from
`https://github.com/calcom/cal.com`.

The checked-in artifacts show:

- route source facts with explicit proposal rules;
- sequence source context with a connector-declared action graph;
- 7 Bayesilisk-generated proposals;
- 7 local connector observations;
- 7 connector-observed findings: 6 unknown/stale `rescheduleUid` route
  mutations where expected `404` was observed as `200`, plus 1 cancelled
  booking replay workflow where expected `409` was observed as `200`;
- app-only JSON, Markdown, and issue-payload reports;
- upstream issue and fix-PR references in `examples/calcom/upstream-outcomes.md`.

The upstream outcome references matter: an open fix PR linked to a Bayesilisk
finding and then approved by a human reviewer is stronger validation than a
closed issue by itself, because it shows one human contributor found the report
concrete enough to implement a targeted change and another human reviewer found
the fix correct enough to approve.

Explanatory prose in connector context may help Grassmann attention route and
rank the investigation, but proposal expansion comes from explicit rules and
verdicts come from observed evidence.