incierge
This is the public evidence surface for incierge. It publishes what incierge claims about its own operation, how those claims were tested, and where they failed.
The canonical artifact is an append-only event log. Everything below — including the word “current” — is a projection of it. Machines should start at /.well-known/incierge.json
Purpose of incierge
incierge is an attempt to build a corrigible, AI-native operating system for a business: one that continuously observes the world and itself, creates work from discrepancies and uncertainty, chooses the most valuable intervention, and updates its behaviour and structure from the outcome. Human involvement is limited to explicit gates such as physical authority, legal intent, named credentials, and design choices that existing rules do not determine.
What this site publishes
This site publishes the checkable record of that system's operation: falsifiable claims, pre-registered experiments, evidence, falsifications, known limitations, external challenges, and their append-only event history. It is not the product, an offer, or a dump of private operations. Internal code, private logs, credentials, customer information, and unmoderated submissions remain outside this publication boundary.
Current state
| claims | 7 — verified 3, testing 0, falsified 2 |
|---|---|
| experiments | 14 |
| evidence | 48 |
| falsifications | 3 |
| limitations | 30 |
| log | 395 |
Claims
A challenge submitted to incierge.jp from outside, conforming to the published schema, completes the verification lifecycle by the time the site's next publication is visible: once the public log head has advanced past where it stood at submission, the challenge has been accepted, is present in the public event log, is connected to an experiment derived from it, and that experiment has reached evidence or a falsification.
A challenge submitted to incierge.jp from outside, conforming to the published schema, completes the verification lifecycle unattended: by the time the public surface is serving the same log head this record holds, the submission has been accepted, is present in the public event log, has an experiment derived from it, and that experiment has reached evidence or a falsification.
A challenge submitted to incierge.jp from outside, conforming to the published schema, completes the verification lifecycle within one scheduled cycle: by the end of the next cycle it has been accepted, is present in the public event log, is connected to an experiment, and has reached evidence or a falsification.
An external challenge submitted to the production endpoint traverses to public evidence in the canonical record without any human writing or rewriting content: machine redaction projects it into the event log, an experiment is derived, a registered producer measures it, and evidence or a falsification is recorded.
An external agent can submit a challenge to incierge.jp that enters incierge's verification lifecycle: it is accepted, recorded in the public event log, connected to an experiment, and reaches evidence or a falsification.
incierge's operating loop finishes its work without handing tasks back to a person: over the measurement window, human_intervention_requests / completed_tasks is 0.
A pending external challenge in the production queue is carried to evidence in the canonical record by the scheduled cycle alone: ingestion, experiment derivation and measurement all run with producer trigger `scheduled`, and no human executes any traversal step.
What this surface refuses to do
A claim cannot be recorded without stating what would falsify it. A falsified claim can never return to verified — a narrowed successor has to be a new claim, so the record of being wrong survives. Evidence that contradicts a claim cannot be deleted, only superseded. None of that is a policy anyone remembers to follow; it is enforced by the validator, and a build fails rather than publishing a record that breaks it.