BLS CPI source profile

Status: accepted bounded operational profile under DEC-0028 after independent review. Profile openucem.bls-cpi, version 0.1.0, is separate from CBP and BEA. Its JSON Schema identifier names this contract; it is not a remotely resolved dependency or canonical UCEM syntax. All references are local. The accepted SC/RE/DE obligations remain unchanged.

Scope and source material

The finite universe is 24 retained CPI-U monthly records: exact series CUSR0000SA0 and CUUR0000SA0, September 2025 through August 2026. Twenty-two supply numeric tokens; both October 2025 rows supply literal padded - tokens and observation footnote X. A source record can exist while its numeric level is unavailable. A retrieval outside this finite universe is a coverage outcome for the consumer, not another publisher-unavailable observation.

The immutable ALE-158 handoff supplies eleven direct BLS bodies and an independently reviewed capture/replay history. Its retained manifest (repository reference) contains exact body hashes, clocks and generation-pinned storage references. The owner capsule SHA-256 is 3e135f736e70ad56e43ed65b00010c90a2ac7a62c0733c8275a0fa528e67d95a. This work rechecks existing local bytes without source requests, cloud reads or new acquisition. Source ownership remains with the acquisition owner; Alethic owns its authenticated workflow.

The source manifest (repository reference) contains reviewed native selections, sources and capture headers. It is a projection of the frozen bodies, not the complete BLS database. The projection receipt (repository reference) pins its input and counts. The complete owner capsule, raw bodies and larger histories remain separately retained at the acquisition and generation-pinned storage locations. Native decoded entities are retained; encoded HTTP response bodies are not recreated. No capture-time or current-value claim replaces a source edition. Sequential captures do not establish an atomic multi-file publisher release.

CPI01: exact native source and context

A bundle must preserve the exact source, capture and native records admitted by the finite manifest. Native field order, position, original header names, padded tokens, empty strings, CRLF text, physical locators, body hashes and acquisition clocks are significant. Trimmed series IDs and reference months in bindings are separate mapping interpretations.

The two actual cu.series rows have thirteen fields. The older cu.txt illustration omits series_title and series footnote_codes; it must not replace the actual captured header. Area and item context includes the native display level, selectable and sort fields. HTTP Date and Last-Modified, capture start/finish and observation reference time remain separate. The source artifact shape retains explicit unknown edition/publication claims.

The full fixture includes 150 supporting records: nine TSV headers, 26 dictionary rows, two complete document bodies and 113 NSA annual averages. M13 is an annual average, not a monthly observation. It cannot become a monthly binding or replace an unavailable month. The 2025 annual average at L04 line 3903 is source material only. Optional extra context does not increase the monthly universe or establish full source coverage.

CPI02: interpreted index meaning and representation

Each monthly row has exactly one binding to its exact native row and thirteen-field series record. The supported meaning is CPI for all urban consumers, all items, U.S. city average, area code 0000, item SA0, with the retained S/U adjustment distinction and monthly period. Adjustment belongs to the definition; shared labels do not authorize SA/NSA substitution. These are index levels with unit index, literal scale 1 and reference base 1982-84=100. Currency and percentage/rate expression are inapplicable. No inflation derivation is included.

Preserve conflicting or differently scoped publisher wording without silently reconciling it. The base dictionary calls S Current, while the format document calls it Standard reference base. The periodicity dictionary calls R Monthly; the document says Regular. The concrete series base period and monthly data/dictionary source material support the bounded interpretation. Current is never a latest-selection policy. The document's published three-decimal-place statement does not establish economic accuracy or uncertainty.

Geographic coverage details, complete classification/weight vintages and historical method context remain explicit gaps. The population label does not recover its complete historical boundary. Original observation publication dates and editions are unknown. Capture times and endpoint fields do not resolve them. The retained format document supplies scoped explanations, not assurance that every description reflects the capture's method vintage.

CPI03: unavailable values and footnote attachment

L04 lines 947 and 3900 retain exact value token - and footnote X. Their bindings must say publisher_unavailable and retain the L12 line 3 explanation: data unavailable due to the 2025 lapse in appropriations. All other admitted rows say supplied. No numeric value is parsed, imputed, rounded, reformatted or substituted by this profile.

Observation and series footnotes are separate arrays with separate source material. Both series footnote fields are empty. Only the two unavailable observations carry X. Empty arrays mean no code in the retained field, not proof that every possible caveat is absent. The supporting R dictionary entry alone does not attach a revision footnote to any selected observation. All 24 records remain eligible for retrieval; numeric availability is not record existence.

CPI04: authority, identity and operational use

Operational bindings start proposed. A reviewed, withdrawn or superseded mapping requires a nonempty scoped receipt; a proposed mapping has no receipt. The scope is exactly one native row with its captured series/context and no numeric operations. Receipt shape does not authenticate a reviewer. Consumers must enforce their own authorization, artifact verification and state transitions before publication.

Every binding retains identity_assessment: not_established and permits_numeric_operations: false. Profile acceptance does not establish cross-source identity, economic truth, complete method identity, causality, arbitrary CPI coverage, original release reconstruction, statistical inference or arithmetic permissions.

CPI05: validation and consumer fidelity

Consumers must use both JSON Schema and the finite validator:

uv run --locked python -m scripts.cpi_profile docs/specification/bls-cpi-v0.1/fixtures/monthly-window.json

Schema validation checks closed shapes and types. The validator additionally checks exact native membership, capture/source correspondence, context resolution, one-to-one binding, and reviewed finite interpretation expectations in admission.json (repository reference). That file is an authored admission oracle, not publisher-native data or a general semantic classifier. In this release, interpretation IDs, values, knowledge states, reasons, source material and version are exact; only a properly shaped operational review receipt may change. Changing interpretations requires a separately reviewed profile update.

Nonempty monthly subsets are allowed. Each retained binding needs its exact series record and every cited native context record, with source and capture records for everything included. Unused admitted context may remain; arbitrary extra records cannot. All supplied monthly observations need bindings. A bundle cannot replace its observation list with M13 records. An empty query result belongs to the consumer query envelope, not a valid bundle.

Finite validation does not reread source bodies, verify cloud availability, authenticate actors or establish source truth. Its output reports the supplied and unavailable counts and finite universe, with these authority limits explicit. Consumer tests must preserve all native tokens and contexts, including an unavailable-only query, through generated models and human/agent public projections. Zero, absence and - are distinct. Structural validity alone does not establish this round trip or deployed behavior.

All exported definitions use Cpi names. Five source material shapes copy the accepted BEA shapes with renamed references, without mutating either BEA or CBP. Alethic owns generation, formatting, public projection, import, review, publication and live acceptance. Consumer readiness review precedes UCEM adoption; its source material belongs in the review record.