Skip to content

Citations and baked artifacts

Export citation records and inspect or replay the package's immutable scientific artifacts. Artifact verification and literature novelty are separate questions.

geometric_function_atlas.citation

Deterministic citation export for registry and literature records.

The website emits four browser-side formats. This module keeps the same small metadata contract available to installed-package users without a browser. It never fetches or validates bibliographic records; callers remain responsible for the provenance and correctness of supplied metadata.

format_citation = citation_formats module-attribute

__all__ = ('CitationBundle', 'citation_export', 'citation_formats', 'format_citation') module-attribute

CitationBundle dataclass

All citation representations for one caller-supplied metadata record.

key: str instance-attribute

formats: Mapping[str, str] instance-attribute

__init__(key: str, formats: Mapping[str, str]) -> None

__post_init__() -> None

to_dict() -> dict[str, Any]

__getitem__(format_name: str) -> str

citation_formats(metadata: Mapping[str, Any] | None = None, *, accessed: str | None = None, **fields: Any) -> dict[str, str]

Return BibTeX, RIS, plain-text, and LaTeX citation strings.

metadata is deliberately a mapping rather than a paper model so this operation also covers generated registry entries. accessed is explicit to keep default output deterministic; no current date is read implicitly.

citation_export(metadata: Mapping[str, Any] | None = None, *, accessed: str | None = None, **fields: Any) -> CitationBundle

Return a typed citation bundle for one metadata record.

geometric_function_atlas.artifacts

Versioned baked scientific artifacts and bounded certificate replay.

The Geometric Function Atlas website publishes immutable scientific artifacts: Schur-parameter coefficient expansions, a coefficient-bound catalog, machine proof certificates, open-problem brackets and conjectures, literature reconciliation rows, and a method reference document. This module exposes those artifacts through the standalone package as a versioned, checksummed, read-only snapshot (src/geometric_function_atlas/data/), regenerated by scripts/bake_artifacts.py from the website repository.

Evidence boundaries (the website's honesty model, preserved here):

  • Exact expansion and bound strings are exact transcriptions of the published snapshot; sharpness, enclosure, attainment, and provenance are carried by the record fields and are never inferred by this package.
  • verify_certificate replays the certificate's declared extremal through the package's exact Schur machinery. A passing replay certifies that the stored extremal attains the stored candidate under the declared class assumptions; it is not a new proof and not peer review.
  • Every lookup is fail-closed: unknown keys, missing artifacts, checksum or schema mismatches, unsupported functionals, and resource exhaustion raise explicit errors; nothing is guessed.

All payloads are deterministic JSON under the closed snapshot result envelope documented in docs/RESULT_CONTRACT.md.

SNAPSHOT_SCHEMA_VERSION = 1 module-attribute

MAX_SNAPSHOT_JSON_NODES = 10000 module-attribute

MAX_REPLAY_ORDER = 3 module-attribute

snapshot_info() -> dict[str, Any]

Return the snapshot identity record (version, date, source, files).

snapshot_verify() -> dict[str, Any]

Verify every baked artifact against the shipped manifest checksums.

Any missing file, checksum mismatch, or schema-version mismatch raises :class:CorruptArtifactError; a passing call returns the integrity record.

list_classes() -> list[dict[str, Any]]

Return the 39-class Ma–Minda catalog with exact phi coefficients.

class_info(class_key: str) -> dict[str, Any]

Return one class record; unknown classes fail closed as unsupported.

expansion(class_key: str) -> dict[str, Any]

Look up the website's Schur-parameter coefficient expansion for a class.

coefficient_bound(class_key: str, functional_key: str | None = None) -> dict[str, Any]

Look up the baked coefficient-bound catalog for one class.

The published record preserves the sharp-versus-enclosure distinction in its sharp field: a sharp entry carries value_exact equal to the bound; an enclosure carries only the certified upper bound and slack.

coefficient_supplement_version() -> str

Return the independently checksummed supplement identity.

is_supplement_certificate(name: str) -> bool

coefficient_table(functional_key: str) -> dict[str, Any]

Return exact attained H3 witnesses or analytically sharp a3/a2a3 values.

H3 sharpness requires a cited direct-class theorem; all remaining rows are lower witnesses, regardless of numerical optimization reports.

list_proofs(class_key: str | None = None, functional_key: str | None = None, status: str | None = None, search: str | None = None) -> dict[str, Any]

Return proof-gallery rows, optionally filtered.

Unknown class or functional filters fail closed as unsupported; an empty match set is a valid result (the corpus simply has no such row).

get_proof(name: str, *, raw: bool = False) -> dict[str, Any]

Return one proof certificate's detail record, or the raw record.

open_problems(kind: str = 'all') -> dict[str, Any]

Return open-problem rows, preserving the two published categories.

enclosure rows are certified brackets awaiting sharp closure; numerical rows are uncertified numerical conjectures. The two are never merged, and neither is presented as a proved sharp value.

reconciliation(class_key: str | None = None, functional_key: str | None = None, category: str | None = None) -> dict[str, Any]

Return literature-reconciliation counts and rows.

counts are the published A–D summary buckets; rows may carry compound buckets (for example "A+C*"). category filters rows whose bucket includes the given letter. No-match rows are literature findings, never novelty certificates.

references() -> dict[str, Any]

Return the versioned method-reference companion document.

verify_certificate(name: str) -> dict[str, Any]

Replay one certificate's declared extremal through the exact Schur machinery.

The replay recomputes the coefficient functional at the certificate's stored extremal Schur parameters using the class phi coefficients from the versioned snapshot, then compares the exact value with the certificate's declared candidate. Fail-closed outcomes:

  • unknown certificate or unsupported functional -> UnsupportedError;
  • extremal not exactly representable or absent -> UnresolvedError;
  • resource bounds (Schur depth / coefficient order) -> ResourceLimitError;
  • recomputed value inconsistent with the declared candidate or upper bound -> CorruptArtifactError.

A passing replay certifies extremal attainment of the stored candidate under the declared class assumptions; it is not a new proof and not peer review, and it does not depend on the stripped leaf archive.

snapshot_payload(*, result_type: str, canonical_inputs: Mapping[str, Any], record: Mapping[str, Any], evidence_status: str, assumptions: tuple[str, ...] | list[str], source_references: tuple[str, ...] | list[str], verification: VerificationReport, exact_expressions: Mapping[str, Any] | None = None, record_count: int | None = None, supplement_version: str | None = None) -> dict[str, Any]

Assemble the closed snapshot result envelope for one operation.

validate_snapshot_payload(payload: Mapping[str, Any]) -> None

Validate a snapshot result against the closed envelope.

This dependency-free validator mirrors validate_error_payload: it rejects unknown top-level keys, requires the exact envelope, and enforces the fail-closed status coherence rules. The JSON schema shipped at schema/result.schema.json accepts the same payloads.