Skip to content

Methodology

Evidence Standard

Structured rules for evidence types, metadata, verification, and publication.

Use this standard when adding evidence to an Allocate finding or case study. It describes the labels, metadata, review states, and citation details needed for another person to check the work.

Scope

This standard defines the minimum structure required for evidence used in Allocate findings, case studies, and human-grounded investigations.

It applies to:

  • agent-collected desk research
  • linked public records and open datasets
  • community-submitted field evidence
  • editor-written synthesis and recommendations

Evidence types

Allocate evidence should be labeled with one primary type:

TypeDescriptionExample
literaturePeer-reviewed research or formal institutional analysisOpen-access journal paper, World Bank report
administrative_recordOfficial public records or government dataBudget PDF, permit filing, agenda packet
operational_recordService or implementation evidenceConstruction notice, route schedule, maintenance map
observationDirect place-based observationStreet audit, field note, repeat count
mediaVisual or audio artifact tied to a claimPhoto of blocked bike lane, flood video
testimonyStatement from a person with local knowledgeResident interview, organizer statement
measurementStructured count or reading using a stated methodTraffic count, curb occupancy tally
inferenceEditor or agent synthesis across sourcesCost barrier inferred from land values and budget gaps

If a source fits multiple categories, use the one that best reflects how the claim is being supported.

Required metadata

Every evidence item should include:

FieldRequirement
titleShort human-readable label
source_typeOne of the allowed evidence types
datePublication date or collection date
locationGeographic scope or exact place if safe to disclose
collectorAgent, contributor, agency, or publication
methodHow the evidence was obtained
firsthandtrue if directly observed or collected
citationURL, document reference, or artifact link
sensitivitypublic, limited, or sensitive

Preferred additional fields:

  • time_range
  • confidence_note
  • repeat_observation
  • related_claim_ids
  • language
  • limitations

Claim rules

Evidence supports a claim. Writers still decide how broad and precise the final claim should be.

Writers and agents should distinguish between:

  • what was directly observed
  • what was reported by a source
  • what is established by a document
  • what is inferred from several facts

A report needs additional support before it turns testimony or a single observation into a broad causal statement.

Verification states

Every evidence item has a visible verification state:

StateMeaning
unreviewedNewly added and not checked yet
structuredMetadata complete and type labeled correctly
corroboratedSupported by another source, repeat observation, or linked artifact
reviewedChecked by a human reviewer or editor
high_confidenceStrongly supported and suitable for decision-critical findings

Verification state and confidence answer different questions. Keep both fields.

Confidence rubric

ConfidenceStandard
lowSingle-source, weak metadata, unclear method, or unresolved conflict
mediumGood metadata, plausible method, some corroboration
highMultiple independent supports, strong method, low ambiguity

A claim should never be marked high confidence if the underlying evidence still depends on one fragile source.

Publication threshold

Publish a claim as a finding only after it has:

  • a clear proposition
  • at least one linked source or artifact
  • enough metadata to audit where it came from
  • a claim state
  • a confidence label
  • reviewer signoff if the claim materially affects recommendations

Decision-critical claims should usually have at least two independent supports. A definitive source can satisfy that requirement on its own.

Privacy and safety

Evidence handling accounts for risk as well as truth value.

Caution

Do not publish:

  • personally identifying information without a clear reason and consent
  • precise locations when disclosure could endanger residents or workers
  • imagery of minors or vulnerable people without strong justification
  • evidence collected through trespass or unsafe instructions

Sensitive evidence may inform internal review. A separate publication decision is required before it becomes public.

Citation format

Allocate writing should prefer short, verifiable citations over vague references.

Minimum citation guidance:

  • use direct URLs for open web sources
  • identify the publisher or responsible entity
  • include year when known
  • note if the source is archived, inaccessible, or secondhand
  • identify whether a claim is an inference or a direct source statement

Citations provide attribution and make the research reproducible.