Bay EvidenceInvestigative research · Bay Area public systemsMethodology is part of the product
Bay Evidence is designed to show how information was produced, what population it covers, what it can support, what it cannot support, and when comparison across time or place becomes unsafe.
Evidence hierarchy
Tiers describe evidentiary role, not moral worth. Administrative data, scholarship, journalism, and lived-experience evidence answer different questions and should be triangulated rather than flattened into one credibility score. Evidence tier and source type are separate fields: a source can be Tier D and investigative journalism, or Tier A and a court order, without collapsing those concepts into one badge.
The critical source landscape applies this rule to the federal, California, academic, advocacy, provider, legal, journalism, and international resources Bay Evidence monitors.
Never collapse stock, flow, inflow, and durable exits
Cross-sectional versus longitudinal evidence
Cross-sectional studies describe people observed at a point or period. They are powerful for prevalence, characteristics, and associations, but can overrepresent people with long-duration episodes. Longitudinal studies follow people through time and are stronger for trajectories, exits, returns, and temporal ordering. Bay Evidence will not describe CASPEH as longitudinal; its findings must be read with its cross-sectional design and duration-bias caution intact.
Conflicting sources
When credible sources disagree, Bay Evidence should display the disagreement, definitions, universes, dates, and methodological differences. It should not silently select the more convenient number. A conflict can be evidence about measurement architecture, not merely an editorial inconvenience.
Bibliographic verification
Every citation used in original Bay Evidence research is checked against a stable publisher, journal, court, agency, or institutional record before publication. Candidate citations that cannot be verified as written are excluded from the authoritative bibliography and are not used to support published claims.
Editorial verification notes belong to the research process, not automatically to the finished publication. If a citation is corrected, replaced, or removed after publication, the material change is dated in the public corrections ledger and the affected report version is incremented.
Evidence Desk inclusion rule
A development belongs in the Bay Evidence Desk only when it materially bears on housing loss, homelessness response, displacement, or the fiscal, legal, administrative, health, or service systems governing those issues in the nine-county Bay Area. Broader state or federal developments are included only when the Bay Area relevance is explicit and documentable. General political news does not enter the Desk solely because it concerns government power or public administration.
Data quality states
Verified and methodologically comparable for the intended use.
Recent and useful, but subject to revision.
Still useful historically, but should not be presented as current.
Comparison across periods requires adjustment or explicit qualification.
The data universe excludes part of the target population.
Credible sources materially disagree.
A claim exists but adequate documentation has not been located.
Discovered or ingested, but not cleared for authoritative use.
Link audit policy
The source ledger distinguishes live, platform-unverified, fetch-limited, and moved-or-unconfirmed destinations. “An automated agent could not retrieve this page” is not logically equivalent to “the source is dead.” Constrained sources remain visible with their status and are not silently removed.
Version control and corrections
Important metrics should expose source date, publication date, retrieval date, geography, population universe, methodology version, known exclusions, missingness, revision history, comparability warnings, review status, and verification status. Corrections should be public and dated. Changes in counting method, eligibility, geography, or administrative coverage must be treated as potential method breaks rather than invisible footnotes.