Evidence Standard and Source Hierarchy

Version 1.0. Issued 21 August 2026. Status: adopted. This document is the load-bearing standard of the programme. Where anything else conflicts with it, it governs.

1. Sourced claims only

Every factual claim in every Index deliverable rests on a named, retrievable source. There are no exceptions, no background knowledge, and no claims that are true enough to state without checking. A claim that cannot be sourced is not softened, hedged, or attributed to general understanding. It is removed, or it is restated as a gap.

2. Never estimate, model, or infer a figure

The Index does not estimate. It does not interpolate between published years, does not scale a national rate to a state population, does not derive a count from a percentage and a denominator drawn from different instruments, and does not carry forward a prior year's figure as if it were current. Where a figure does not exist, the Index states that it does not exist and states what was searched.

This rule is more restrictive than ordinary research practice and it is deliberate. The field this Index documents is full of numbers whose provenance dissolved somewhere between a source and a citation. The Index's usefulness depends on a reader being able to trust that every number in it came from somewhere nameable.

3. Explicit gaps

A gap is a finding and is written as one. The formulation is: what was sought, what was searched, what was found instead, and what would close the gap. A gap recorded as "data not available" without the search behind it is not acceptable, because it is indistinguishable from not having looked.

Gaps carry re-run information wherever the retrieval is likely to succeed later or with different tooling, including exact report identifiers, element identifiers, dashboard parameters, and file paths.

4. Record conflicts rather than resolving them silently

Where two sources disagree, both are recorded, the disagreement is described, and the file states which source it prefers and why. Silent resolution is prohibited. The preference is usually justified on one of three grounds: the source is the originating instrument rather than a reproduction; the source reconciles across levels of aggregation; or the source is more recent and the older figure is superseded rather than contradicted.

Instruments measuring different populations are never reconciled at all. A state's IDEA deafblindness category count and its national deafblind child count measure different things under different criteria, and presenting them as a discrepancy to be resolved would be an error of category rather than of arithmetic.

5. Separate fact from interpretation

Statements of interpretation are marked as interpretation and kept separate from statements of fact. A reader must be able to see where the record ends and the analysis begins. Where the Index draws a structural conclusion from a pattern across sources, the pattern is stated first with its sources, and the conclusion follows, marked.

6. Quote the instrument

Statutory and regulatory bases are cited to section with the operative sentence quoted. Published philosophies, admission criteria, and eligibility language are quoted verbatim rather than paraphrased. Paraphrase in this field loses exactly the content that matters, because the questions the Index asks turn on whether a document says a thing or merely implies it.

7. Where you exercise a preference, say why

Any methodological choice that a reasonable reader might have made differently is stated with its reasoning. This applies to frame decisions, boundary decisions, source preferences, coding judgements at the margin, and the treatment of any entity that does not fit the taxonomy cleanly.

8. Silence is not evidence of absence, and absence of publication is not absence of service

Two formulations are mandatory in every workbook README. NONE FOUND is a statement about publication, not about service delivery. And obligations under IDEA are not evidence of a programme. Both exist to prevent the Index's coverage data being read as a service audit, which it is not and does not claim to be.

9. No em dashes

No em dashes appear anywhere in any Index deliverable. This is a formatting standard rather than an evidentiary one and it is checked programmatically before release under STD-003.

Version 1.0. Issued 21 August 2026. Status: adopted.

1. The hierarchy

Sources are ranked by proximity to the instrument that created the fact. Where sources at different tiers conflict, the higher tier is preferred and the preference is stated under SRC-001 section 4.

Tier 1. The constituting instrument. State statute, session law, administrative code, state board rule, appropriations act, and the text of a federal regulation. Read the text itself, not a summary of it, and record which version was read, including whether it was the codified text or a session law as enacted.

Tier 2. The state agency's own primary data publication. Certified enrollment files, special education child count files, directory data, and official reports issued by the state education agency, the state health department, or the state auditor. Downloadable data files are preferred over dashboards in every case.

Tier 3. The entity's own published statements. A school's admission criteria, philosophy, handbook, board minutes, budget documents, and job classifications. These are authoritative about the entity's own claims and are quoted as claims rather than as verified facts about practice.

Tier 4. Federal data collections. IDEA Section 618 files, the national deafblind child count, and Centers for Disease Control hearing screening survey data. Authoritative for what they measure, frequently mismatched to state category boundaries, and often inaccessible in practice.

Tier 5. Legislative and oversight records. Bill text, fiscal notes, committee records, audit reports, and litigation filings. Authoritative for what was proposed, found, or alleged, and never for what is currently the case unless a Tier 1 or Tier 2 source confirms it.

Tier 6. Contemporary journalism and organisational reporting. Usable for establishing that an event occurred and for locating a primary source. Never the sole basis for a figure.

Not a source. Encyclopaedic summaries, aggregator sites, advocacy compilations, and prior versions of this Index. Each of these may be used to find a primary source and none may be cited in place of one.

2. Organisational self-description is not a governance basis

A body's account of its own statutory standing is a Tier 3 claim and is checked against Tier 1. Florida's believed statewide Deaf service centre network described itself in terms the code did not support, and the discrepancy was found only by reading the code text. This check is routine rather than exceptional.

3. Directory accuracy is itself a finding

State-published directories of local education agency websites are the standard starting point and they are not reliable. Florida published a superintendent contact directory with websites, which saved considerable time, and two of its sixty-seven listed hosts were wrong, one serving an invalid certificate and one returning a nonexistent domain. Both were found only in a rendered browser pass. Hosts are verified rather than trusted, and the accuracy of the state's directory is recorded in the file in both directions.

4. Access dates

Every citation carries an access date. Public sector web content in this field changes without notice and without versioning, and several findings in completed state files concern content that had already gone stale at the point of retrieval.

5. Version discipline on legal texts

Where a statute or rule is available in more than one form, the file records which was read. Live administrative code sites are frequently JavaScript shells, and rule text is frequently served as a document download behind a gateway. Delaware and Florida both presented this obstacle. Where a rule could not be read in any form, that is recorded as a NOT VERIFIABLE finding at the instrument level with the reason named.