> ## Documentation Index
> Fetch the complete documentation index at: https://docs.beaconrevenue.io/llms.txt
> Use this file to discover all available pages before exploring further.
# How confident a number is
> The three confidence measures Beacon publishes — per record, per connected system, and per forecast — what each governs, and the floors below which a figure is withheld.
Beacon measures its own confidence at three levels, and they answer different questions. A figure is never published as though every input behind it were equally sound.
This page describes a designed model, not shipped behaviour. The measures and thresholds below are fixed in Beacon's model; their published names and door availability are being authored into the field catalogue and may change before they are callable.
## Definition
| Level | Measure | Answers |
| -------------------- | ------------------------- | ------------------------------------------------------------------------------ |
| One record | `entity_confidence_score` | How sound is the data behind this account or this deal |
| One connected system | `system_confidence_score` | How sound is everything coming from your CRM, or billing, or usage, or finance |
| One forecast period | `fci_score` | How much weight this forecast carries |
The three compound. A record's confidence rolls into its system's; the system's is the largest single input to the forecast's.
## How it behaves
**Low confidence is disclosed, never hidden.** A figure computed from thin data is labelled, and the label travels with the figure on every door. It is not quietly downgraded and it is not silently withheld.
**Below a floor, a figure is withheld rather than published.** A record under 60 is treated as data-absent — not estimated, not carried forward, and the floor is not overridable. A connected system under 60 suspends every prompt and alert Beacon would otherwise send, company-wide, while the underlying figures keep computing.
**Confidence governs what Beacon may do on its own.** Forecast confidence sets the autonomy ceiling directly: at 80 and above Beacon's agents may act within their grants; between 60 and 79 they draft and wait; between 40 and 59 they recommend only; below 40 they observe and do nothing. That last step is unconditional and cannot be overridden by any instruction or urgency.
**A missing input lowers coverage, not the score.** A dimension whose source is not connected contributes a neutral score rather than a zero, and the forecast declares that its coverage is partial. A gap is stated, never scored as though it were bad news.
**A sharp drop is treated as an event.** Five points inside a day opens an anomaly review rather than passing unremarked.
## Constraints
* **Confidence is not accuracy.** It measures how much of the required data is present, current, internally consistent and reconciled across systems — not whether the answer turns out to be right.
* **Two of the three are computed by fixed formula.** Record and system confidence run in the calculation layer, from your data, with no model involved. Forecast confidence is the one number Beacon reasons rather than calculates — and it governs how much weight a forecast carries; it never overwrites a figure, and it cannot change a sealed one.
* **It is not a setting.** No configuration raises a confidence score, and no plan tier changes a floor.
* **It is not a data-quality report.** Beacon publishes these to govern its own behaviour. Fixing what they surface happens in your systems.
## Related
The five dimensions behind a record's confidence, and the thresholds.
The six dimensions behind a forecast's confidence, and the autonomy ladder.
The two timestamps on every figure.