Most AI systems are designed to tell you what is known. HNX-1 is being built to estimate what happens next — narrowly, with a record that reality can later check.
For a supported forecast, the number belongs to the registered predictor. The explanation only makes the stored probability inspectable; it cannot replace it.
A future-facing question with a defined horizon.
A registered event with an objective resolution rule.
Only observations available before the prediction cutoff.
For the first route: a registered Bundesliga Poisson model, not chat prose.
The numeric vector, feature snapshot and source lineage are stored before explanation.
The source-confirmed final outcome later scores the record. Football calibration is still experimental.
One identifiable fixture and an objective resolution rule.
Completed pre-cutoff results and league baselines.
A status earned by clean evaluation, not a marketing adjective.
Provenance, availability and immutable feature snapshots.
The public system documents only the inputs it actually uses. Project 2049 is a living technical record, not a promise that every listed layer is already live.
Every grounded forecast is timestamped, frozen, resolved, and scored. When the outcome is known, the call becomes a record rather than a promise.
The prior ECB calibration lineage is invalidated. The first football route is explicitly uncalibrated experimental until a clean, prospective evaluation earns a different status.
HNX-1 stores the complete probability vector, model versions and evidence snapshot. If a forecast is revised, it becomes a new immutable revision; the original remains visible.
Market Intelligence and the Intelligence Mesh may collect or evaluate signals in research and shadow modes. They do not influence consumer football probabilities.
Selected upcoming Bundesliga fixtures can use pre-cutoff completed results, league baselines and a Poisson predictor. Market odds do not enter the predictor. The rollout is controlled and experimental.
New domains need their own point-in-time data, specialist predictor, canonical resolution, scoring and evaluation gates. They are not activated by a model rename.
Oracle Pig does not make the forecasts. HNX-1 does. He just takes all the credit when the record resolves well.