Judge the state. Do not invent the score.
This page records how the open methods lab may use BeatAPI. It is not a product, a procurement, or a vendor endorsement. BeatAPI is a third-party API. No vendor sponsors this work. A catalogue count is not a FrontierOrg result.
How a call is made
Agents call REST at https://api.beatapi.io. The credential is a private bearer token on the agent host. It is not in this repository. This page does not have it. The page only fetches the static file data/beatapi-catalogue.json.
The browser never holds the key. It does not send an Authorization header. Scoring, if it is ever run, happens on the agent side and may write a result file. That file may hold answers and request ids. It must not hold the key. This page does not start a run. A paid run needs an explicit ask. See the maintainer note.
- beatapi.io/skill.md — agent recipe. Search and inspect before any run.
- docs.beatapi.io — vendor documentation. Not our method.
- Methods — why the repository is a lab.
- anorthernman-developer/frontierorg-diary — the evidence.
What the lab would use it for
Sample data stays in this repository. Generators and fixtures are the evidence. BeatAPI is not a data warehouse for FrontierOrg and it is not a measured outcome.
- Decisions. model:jev-1.13-free takes a written state and typed questions. It returns typed answers. It does not generate prose. A go, revise, or stop judgment can be posed as a choice. The answer is not a board minute, not a benefit, and not a cell on the scorecard.
- Web research. The catalogue includes a web search and read group. A fetched page is untrusted text. It does not become a diary fact unless a person writes it and the diary words are approved.
- Optional scoring. A score question type exists. We do not use it to invent an enablement index, a return on investment, or a percentage the fixture does not support. Actuals stay not measured.
Pause beats a fake win. If the state cannot be grounded in a fixture or an approved note, the question is not asked. Cost, risk, and engagement and enablement stay in the same view. A model reply does not close an issue.
Inspected capability
The snapshot inspects model:jev-1.13-free (JEV decisions, free tier). Required inputs are state and questions. Question types in that schema are noul, choice, and score. Execution is synchronous: the answer comes back on the response. The snapshot price is zero US dollars per million tokens. That is a snapshot of a free tier, not a live quote and not permission to run it from the browser.
- noul. A calibrated likelihood from 0 to 1. The name is the upstream term, not a misspelling. We do not treat a likelihood as a saving.
- choice. Pick one named option. Go, revise, and stop fit this type. Day 22 already recommends revise, in prose, with no vote claimed. A later model choice would be a working note, not that recommendation rewritten.
- score. A position on an ordered scale. FrontierOrg will not fill a scorecard from it. There is no enablement score.
FrontierOrg would pass the state as text or JSON the diary already stands behind, and name the questions in the same call. The schema rejects unknown fields. We would not add a hidden field to smuggle a conclusion. We would not copy an answer into a board pack and call it measured.
Snapshot catalogue
Tables below are rendered from data/beatapi-catalogue.json. If that file does not load, the tables stay empty. This page will not invent a catalogue.