Evidence operations

Where every reading came from, and what is still missing.

Environmental intelligence cannot scale if nobody can see the provenance. Every reading below was refused by the engine — either because a model proposed it and no person has confirmed it, or because nothing was ever gathered. Neither reaches a score, and neither is treated as a zero.

Unconfirmed proposals8

No evidence at all7

Evidence grades

Grade decides the ceiling on confidence. No intake can claim more certainty than its class of source allows.

Evidence grade distribution
Grade	Readings	Confidence ceiling	What it means
Partner confirmed	33	100	The partner asserted it, on the record.
Public source	27	90	Read on a named public source of record, with the citation retained.
Derived	45	85	Computed from a verified fact about the location, never from the property's own marketing.
AI proposal	8	60	Proposed by a model with retained evidence and no human verification. Does not reach the score.

›Why the ceiling is applied on ingest

A reading arrives with a claimed confidence. The grade’s ceiling is applied when it is projected, not when it is rendered, so nothing downstream has to remember to apply it and no surface can accidentally show the claimed number instead of the allowed one. Where a claim was capped, the profile shows both figures.

The queue

Filter by property, dimension or class. Every row expands into everything a reviewer needs to act.

PropertyDimensionOrigin

Impact

15 of 15 readings. Ages are measured against 2026-08-13, the newest verification date in this portfolio — never against today’s clock, so the same data always reads the same way.

How the classes differ

An unknown is not a zero, and a proposal is not evidence.

Property: The property is answerable for this. It reflects what the property itself provides.
Location: Inherited from where the property sits. The property is not credited with it.
Computed: Computed per guest at score time from the property's location. Never stored.
Inferred: Proposed with retained evidence but not confirmed by a human. Excluded from every score.
Unknown: No reading exists. Excluded from both sides of the arithmetic — never counted as zero.

Verification write-back is not built

The verification queue is computed and reviewable. Confirming a reading from the browser would change what the engine scores, so it is not wired up. Partner-scoped write path with an audit trail, reviewer identity, and a re-assessment pipeline triggered on evidence change. Until that exists, this centre is a read surface: it shows what must happen, and the change itself is made in the portfolio record and re-assessed.

See which gaps repeat across the portfolio See what the gaps are costing

Produced underscoring 1 · weights 4 · ontology 1 · climate model 1 · engagement 1 · min coverage 0.5

Grade

Readings

Confidence ceiling

What it means

Partner confirmed

100

The partner asserted it, on the record.

Public source

Read on a named public source of record, with the citation retained.

Derived

Computed from a verified fact about the location, never from the property's own marketing.

AI proposal

Proposed by a model with retained evidence and no human verification. Does not reach the score.