Loading…

Coverage

20,438 of 56,681 columns decided — 36.1% · 4,487 bound (7.9%) · 36,243 still unread

Bound: 4,487; 7.9%; a concept was found and asserted for the column.
Refused: 10,870; 19.2%; the column was read and no concept fits. A verdict, not a gap.
Insufficient: 5,081; 9.0%; read, but the evidence did not settle it.
Unanalysed: 36,243; 63.9%; nobody has read this column yet.

56,681 columns in the coverage ledger · read from /api/field-coverage on citytap-prod · source last observed Oct 10, 2026, 10:53 AM

Where the unread columns are

3,026 datasets, 56,681 catalogued fields, 36,243 of them still unread. Every band below is a share of that, worst first. Open one to see its datasets.

Gaps in the data behind this page

36,243 of the 56,681 columns in the coverage ledger have never been analysed.
603 of the 3,026 datasets in the catalog list no columns at all.

Source last observed Oct 10, 2026, 10:53 AM.

Where each note comes from

36,243 of the 56,681 columns in the coverage ledger have never been analysed.
36,243 / 56,681
Source: /api/field-coverage
Check it yourself: /api/field-coverage#/coverage/unanalysed
603 of the 3,026 datasets in the catalog list no columns at all.
603 / 3,026
Source: /api/field-coverage
Check it yourself: /api/field-coverage#/catalog/datasets_no_inventory

These pages state their own gaps: /datasets, /freshness, /scorecard, /scorecard/fields, /analyses, /ontology, /graph, /alerts, any topic page under /domains.

What we have understood

Freshness above is about the data being current. This is about the data being understood: how many of the catalogue's fields carry a meaning we can act on, which datasets are still waiting, and how far the joinable web reaches.

← Citywide scorecard

Datasets by how much of them we have read

Filtered to datasets carrying month. Clear

Datasets with the most fields left to read

Showing 1–11 of 11
Dataset	Fields	Decided	Understood	Nobody looked	Shared terms
Weigh in Motion (from 2025) 4fwc-j3vn	48	1 2%	1 1 matched by name, unread	47	1	See every field
Weigh in Motion (Historical) n5c8-x7x2	48	1 2%	1 1 matched by name, unread	47	1	See every field
Summary of Syringe Data in NYC Parks t8xi-d5wb	23	6 26%	2	17	1	See every field
Local Law 159 of 2019: Vacate Relocation Charges By Building 3bkg-usya	23	13 57%	10	10	7	See every field
FHV Base Aggregate Report 2v9c-2k7f	9	1 11%	1 1 matched by name, unread	8	1	See every field
311 Resolution Satisfaction Survey 5ijn-vbdv	11	4 36%	4 1 matched by name, unread	7	2	See every field
Credited Job Placement Report (Historical) afsf-hz68	7	1 14%	1 1 matched by name, unread	6	1	See every field
DOHMH Call Center Summary wymh-dvmd	6	1 17%	1 1 matched by name, unread	5	1	See every field
Local Law 33 - Security Indicators Report 2wuc-x56b	6	2 33%	1	4	1	See every field
Cash Assistance Recipients in NYC (Historical) qtrj-g3nm	5	1 20%	1 1 matched by name, unread	4	1	See every field
Home Care Caseload (Historical) xjur-zbxw	5	1 20%	1 1 matched by name, unread	4	1	See every field

The web of datasets and fields

A line means a dataset carries a field tagged with a shared vocabulary term. Two datasets on the same term speak the same language about it, which is what makes a cross-dataset question possible. Sharing a term is not the same as a working join — the dataset page says whether a specific pair compiles.

Bipartite. A dimension node is a conformed-dimension family; a dataset node is a catalog dataset; an edge means that dataset carries at least one column tagged with that dimension, and names the columns that carry it.

No field list yet: We hold no columns for these at all.
Nothing read yet: We hold their columns and have decided none of them.
Under half read: Fewer than half their fields carry a verdict.
Half read: At least half their fields carry a verdict.
Most read: At least three quarters of their fields carry a verdict.
Every field decided: Every field carries a verdict — which does not mean every field is understood.

How many of an edge's tagging columns carry a method that actually read something. Zero means the edge rests entirely on carried lexical tags.

Two datasets on the same dimension node share a vocabulary, which is weaker than a join. GET /api/datasets/{fourFour}/bindings/{partner} decides whether a specific pair compiles; executing it additionally needs both marts present.