AxiomCodex/axiom-poi-intelligence-us-pa

Axiom POI Intelligence — Pennsylvania (Overture Places, full)

Readiness Status

Public research release of upstream open data; see source licensing and attribution below. No commercial or training authorization is implied. Candidate-bound evidence, independent signoff, release-candidate validation, exact-target release audit, and owner approval are complete.

Summary

One full normalized extract of eligible Overture Maps Places records assigned to Pennsylvania from the pinned release 2026-09-23.1 (upstream schema v2.0.0). “Full” means all eligible rows in this candidate's captured state population after the exclusions described below, not every Overture place or a verified-complete directory of every local business.

State assignment uses point-in-polygon against Census TIGER/Line 2024 state boundaries (tl_2024_us_state.zip, sha256:ad00cbe66c7177091b668cee202e93d4a1ddcee271c28d1c9f9874af59c04b92), not address labels. No row quota or confidence cutoff is applied. Coverage varies across places, regions, and categories; this is a static observation of one release.

Quick Start

Install datasets (pip install datasets) and stream a record without loading the full state extract into memory:

from datasets import load_dataset

places = load_dataset(
    "AxiomCodex/axiom-poi-intelligence-us-pa",
    split="train",
    streaming=True,
)
for place in places:
    print(place["name"], place["primary_type"], place["location"])
    break

location is JSON-encoded point data; parse it before spatial calculations.

Schema and Observation Semantics

The flat Axiom Codex POI v1 projection is documented below. In the pinned state output, the Arrow type is string for every field except confidence (double). Date-like fields are strings, not Arrow timestamp values. location and types are serialized JSON strings. The staged manifest is authoritative for the exact schema and nullability of each rendered output.

Column	Arrow type	Meaning
acl_tier	string	Governed access-tier label; the field is not an access grant.
confidence	double	Unthresholded upstream confidence about the place, not operating status; no validated cutoff is supplied.
first_seen_at	string	Source first-seen observation in this snapshot; not an opening date.
h3_index	string	Resolution-8 cell for spatial joins; not a cell-level place record.
id	string	Overture place/GERS identifier.
ingested_at	string	Axiom Codex ingestion time.
last_seen_at	string	Source last-seen observation in this snapshot; not a closure date.
location	string	JSON-encoded point location.
modified_at	string	Package record modification time.
name	string	Source place name; contact-pattern names are excluded as described below.
normalization_version	string	Axiom Codex normalization version.
observed_at	string	Observation timestamp in the source snapshot; not a business lifecycle date.
ontology_type	string	Primary mapped Overture taxonomy type.
overture_id	string	Overture source object identifier bound to the snapshot.
primary_type	string	Primary category from taxonomy.primary, falling back to basic_category only when absent.
record_id	string	Stable normalized source-record identity.
schema_version	string	Axiom Codex POI projection schema version.
source	string	Upstream source label (overture).
source_feed	string	Governed source identifier (overture-maps-places).
source_system	string	Source system (overture).
source_uri	string	Upstream source record URI; not a substitute for omitted per-record license metadata.
status	string	Reported operating-status field; null means unreported, not open.
types	string	JSON-encoded, de-duplicated taxonomy hierarchy, alternates, primary type, and basic category.

confidence is preserved without thresholding and is not evidence that a place is open. When status is null, it means unreported, not open. first_seen_at, last_seen_at, and observed_at describe source observations, not verified opening or closure events. No authoritative lifecycle history, cross-source entity resolution, adjudicated outcomes, or leakage-safe ML split is supplied. The single train split is a Hugging Face transport convention, not an ML split.

Provenance & Source Licensing

Source: Overture Maps Foundation, Places release 2026-09-23.1 (upstream schema v2.0.0). The documented upstream Places guide is https://docs.overturemaps.org/guides/places/. State membership is computed from the source point and the pinned Census TIGER/Line 2024 state boundary file; the boundary digest is shown above.

The projection carries source_feed=overture-maps-places, but does not carry per-record sources[]. A row therefore does not expose its contributing-source license. The package includes the CDLA-Permissive-2.0 and Apache-2.0 texts and the Foursquare NOTICE; the upstream components also include AllThePlaces (CC0-1.0). Apply the included upstream terms and attribution; package metadata does not grant additional rights or replace record-level source attribution where available.

Modified by Axiom Codex on 2026-10-28: projected to the documented columns, normalized to the POI v1 envelope, contact-pattern names dropped, filtered to one US state. The state filter uses the pinned boundary geometry described above. Contact-pattern exclusions remove names containing email addresses or phone-like patterns. The candidate's manifest binds its exact staged files, source receipts, and release-specific snapshot evidence; consult it for this export's measured inventory.

This is a static snapshot, not an automatically refreshed feed. Later Overture releases or source corrections require a new pinned acquisition, updated evidence, and release-specific review.

The package ships the upstream terms and attribution files: CDLA-Permissive-2.0.txt, Apache-2.0.txt, FOURSQUARE-NOTICE.txt. The staged Overture records were modified by normalization to the Axiom Codex POI schema and filtered to the governed field projection. This package contains the full available source population (630,066 of 630,066 planned rows).

Privacy and Exclusions

Business listings can contain a sole trader's personal name, and some places may be home-based businesses. These are residual privacy risks; the candidate does not claim that any named person is identifiable or that a listed address is residential.

Rows whose business name contains an email address or a phone-like contact pattern are dropped. An Oregon screening of 233,655 rows found 33,342 possible_personal_name name-pattern matches (14.27%), 23,016 residential-category proxy matches (9.85%), and 51,725 rows matching either screen (22.14%; overlap counted once). These are screening signals, not verified personal identities or residential addresses, and the counts are specific to Oregon—not this state when different, or another release. The owner decision is disclosure only: no per-row flags are added and no rows are dropped based on either screen. The separately documented email/phone contact-pattern exclusion remains in effect. This screening summary does not represent a new state-specific privacy approval. Treat names as business listings; do not profile, track, or contact individuals. Google/Yelp enrichments and restricted payloads are excluded.

Quality Evidence

The generated package's manifest, quality report, and baseline report describe its exact inventory and candidate-specific checks. Review those artifacts for the rendered output; successful Hub loading alone is not a quality-pass claim. This card does not claim model performance or an ML split.

Citation

Overture Maps Foundation, Places release 2026-09-23.1, schema v2.0.0; Pennsylvania state extract. See the shipped licence texts and Foursquare NOTICE.

Static Update Policy

This card describes one pinned Overture Places release and one state extract. The upstream release cadence does not refresh these files automatically; a newer release requires a new collection, state-population binding, schema and quality review, and updated source evidence.

Contact and Support

When this dataset repository is available, use its Hugging Face Community discussions for questions, corrections, or removal requests, including a place listing. Avoid posting personal or sensitive information. You may also contact Axiomancer Labs at www.axiomcodex.io.

Governed Dataset Card Metadata

• Source / license posture: Overture Maps Places (overture-maps-places) is redistributed with package-level CDLA-Permissive-2.0 and Apache-2.0 texts and the Foursquare NOTICE covering all contributors (including AllThePlaces CC0-1.0); per-record sources[] and source licence metadata are not carried; no other source is included.
• Intended use: Research, exploration, and evaluation of static place observations, category mix, and H3-level business composition.
• Forbidden use: No profiling, tracking, or contacting of individuals named in place listings, and no treatment of status or seen-at dates as verified business lifecycle facts.
• Known missingness / bias: No authoritative lifecycle, cross-source entity resolution, evaluation, or leakage-safe ML splits; operating status may be unreported (null means unreported, not open), and coverage is uneven by region and category.
• Update cadence: Static full-population snapshot of Overture Places release 2026-09-23.1; later releases require a new collection and review.
• Geographic / temporal coverage: One US state: Pennsylvania, assigned by point-in-polygon on Census TIGER/Line 2024 boundaries; observation dates describe the pinned snapshot, not business opening or closure.
• Label semantics: Place fields are static source observations, not verified entity truth or ML labels.
• Confidence / provenance method: Overture source-policy receipt, per-record source URI, and content-addressed collector and staged manifest evidence.
• Training policy: Not authorized for training under the Axiom Codex governed training-use policy; reuse rights come only from the upstream terms, and the single train split is a Hub convention, not an ML split.

Release Scope & Quality Evidence

• Scope: source_records; 630066 staged rows from upstream collector Parquet inputs.
• Observed time window: 2026-10-27T18:58:33Z to 2026-10-27T18:58:33Z.
• Machine-readable evidence: manifest.json, quality.json, and baseline.json.
• Packaging validation command: python3 tools/hf_packaging/validate.py --staged <staged-dir> --dataset poi-intelligence --release-candidate --variant full-us-pa --tier research. This records the build's validation entry point and requires the Axiom Codex packaging environment, which is not included in this Hugging Face data repository. Use the quickstart above to load the data. This card does not claim a model baseline or an ML split unless a future task/evaluation configuration supplies one.
• Promotion: a release candidate requires the snapshot-bound quality report, baseline, current source-policy replay, and all applicable thresholds; blocked or unknown gates remain blocked.