Evidence Weighting Model

Assumptions

Everything the model believes, stated before the math. If you do not buy one of these, do not buy the output.

1. The scoring function is fixed and published, not fitted. Weights are ordinal policy, documented and versioned - judgment made explicit, never learned from the data it judges.
2. Weight multiplies four factors: source tier, milestone maturity, recency, corroboration.
3. Independent primary sources compound; echoes of the same source do not.
4. Every other model consumes evidence only through these weights - no model reads a raw claim.

Derivation

The weight of an evidence item is the product of its four factors:

Published tier weights (policy v1.0.0): tier 1 = 1.0, tier 2 = 0.6, tier 3 = 0.35, tier 4 = 0.1. Maturity weights are the proposed land / water / air = 1, deposit = 2, ESA = 3, slot = 1 (awaiting the owner's ruling, tracked in Decisions for Eni). Recency scales with the field's freshness class; staleness cuts it. Corroboration compounds only across independent primary sources.

The weight behind a claim aggregates to the primary-side share - the fraction of total weight carried by primary evidence - which is what the Project State Model reads when it caps a secondary-only state below 0.5:

Identification strategy

Why the estimate supports the claimed meaning - and where the model refuses to claim more than the data carries.

• Because the function is published policy rather than fitted, every weight is auditable to a rule. A fitted scorer would smuggle the training data's biases into every downstream model; a fixed one can be argued with, versioned, and changed in the open.
• The non-compounding rule for echo sources is the identification guard against circularity: five articles re-reporting one press release carry one source's weight, not five.
• Validation is adversarial review, not backtesting: could a sponsor who understood the weights game them into a different answer? The review exists because the weights are the attack surface.

Inputs and outputs

Calibration and gate evidence

Validation design

• None by design - the weights are policy, ruled on, not estimated. Validation is adversarial review against gaming.

Evidence on record

• The maturity weights are marked PROPOSED in source, awaiting the owner's ruling (Decisions for Eni). The page reflects them as proposed, not settled.

Failure modes

The ways this model declines, degrades or goes UNKNOWN - and what it prints when it does.

• Evidence with missing tier metadata defaults to the weakest defensible tier, never the strongest.
• Stale evidence keeps its tier but loses recency; a stale tier-1 document does not bind like a fresh one.

Sensitivity

• Tier assignments dominate: moving a source class one tier moves every claim it touches by a factor of up to ~3x.
• The maturity-weight ruling is the largest open policy sensitivity; the values on this page are the proposed ones, flagged as such in source.