@urbankitstudio/atlas 0.6.43 → 0.6.45

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md ADDED
@@ -0,0 +1,2260 @@
1
+ # @urbankitstudio/atlas Changelog
2
+
3
+ ## 0.6.45 — 2026-10-25
4
+
5
+ Two counties join, each with a searchable owner-name field: Stafford County, Virginia (`va-stafford`, FIPS
6
+ 51179) and Carroll County, Ohio (`oh-carroll`, FIPS 39019). `totals` move to 503 counties, 498 endpoints and
7
+ 471 counties with an endpoint. No type changes.
8
+
9
+ - **Stafford VA**: the County GIS Office's own ArcGIS Server, `WebApp/WebApp_Parcels` MapServer layer 0, 61,280
10
+ parcels, 61,096 with an owner (`Owner1`), owner mailing address in `Own_Street`, `Own_City`, `Own_State` and
11
+ `Own_Zip`, parcel number `PARCEL_ID_NSP`. The coordinate system is Virginia North feet, so a point query
12
+ sets `inSR=4326`. The county's open-data copy has no owner columns and is not registered. The licence is
13
+ `unknown`: no term is published on the layer, the service or the county pages.
14
+ - **Carroll OH**: the County Auditor's own ArcGIS Online organisation (account `CCAuditorGIS`),
15
+ `Carroll_Parcels` FeatureServer layer 0, 28,680 parcels, 28,267 with an owner (`Listed_Name`), owner mailing
16
+ address in `Owner_Contact_*`, parcel number `Parcel_Number`. Last edited 2026-10-24. The licence is
17
+ `unknown`: the layer carries no terms, and the county's parcel viewer item says only that the data is
18
+ maintained for the county's own purposes, with no warranty.
19
+
20
+ ## 0.6.44 — 2026-10-25
21
+
22
+ The three confidential-flagged Texas records' `sampleQuery` URLs exclude flagged owners, and the advised
23
+ `where` clause casts the flag column to text. No type changes; `totals` are unchanged.
24
+
25
+ ### The flagged records' sample queries no longer return flagged owners
26
+
27
+ In 0.6.43 the `sampleQuery` of `tx-harris`, `tx-collin` and `tx-ector` searched owner names
28
+ (`UPPER(owner) LIKE UPPER('%SMITH%')`) without the flag column in `outFields` and without an exclusion, so
29
+ running a record's own sample could return an owner the county marks confidential under Texas Tax Code
30
+ §25.025. Each now requests the flag column and ANDs the exclusion:
31
+
32
+ - **Harris** (`tx-harris`): `... AND CAST(confidential_flag AS VARCHAR(255)) NOT LIKE '%Y%'`, outFields gain
33
+ `confidential_flag`.
34
+ - **Collin** (`tx-collin`): `... AND CAST(GIS_DBO_AD_Entity_confidential_ AS VARCHAR(255)) NOT LIKE '%T%'`,
35
+ outFields gain `GIS_DBO_AD_Entity_confidential_`.
36
+ - **Ector** (`tx-ector`): `... AND CAST(conf_flg AS VARCHAR(255)) NOT LIKE '%1%'`, outFields gain `conf_flg`.
37
+
38
+ Measured live 2026-10-25, whole layer (`1=1`, `returnCountOnly`): Harris 1,551,736 rows, 4,830 flagged and
39
+ 5,235 with a NULL flag cell, 1,541,671 kept by the clause; Collin 452,384 rows, 1,598 flagged and 15,003
40
+ NULL, 435,783 kept; Ector 77,042 rows, none flagged and none NULL, 77,042 kept. Both sums are exact, so the
41
+ clause drops every flagged row and every NULL cell. Each new sample answers HTTP 200 with 10 rows, the flag
42
+ column present and no flagged row.
43
+
44
+ `validate-atlas` now fails a record whose endpoint carries `confidentialFlag` and whose `sampleQuery` lacks
45
+ the flag column in `outFields` or the CAST exclusion for each value in its `where`, or whose `where` puts an
46
+ OR before the exclusion at the top level (an OR glued to a number or a bracket, as in `1=1OR 1=0`, counts:
47
+ the server reads it as an OR, and that shape returned all 4,830 flagged Harris rows on 2026-10-25).
48
+
49
+ ### The advised `where` clause casts the column
50
+
51
+ The 0.6.43 README, type comments and changelog advised `<field> NOT LIKE '%<value>%'`. On Ector, whose
52
+ `conf_flg` is an integer, that answers with an ArcGIS error (code 400, inside an HTTP 200 response, so a
53
+ client that checks only the HTTP status misses it). `CAST(<field> AS VARCHAR(255)) NOT LIKE '%<value>%'` runs
54
+ on all three layers with the same counts. A NULL cell still fails it and the row drops out, and a missing
55
+ column makes the query fail, so the SQL route stays fail-closed. The 0.6.43 entry below also gave Collin `<> 'T'`, which contradicted its own one-character
56
+ rule; Collin's `"T"` takes `NOT LIKE '%T%'` like any one-character value. A longer value still uses
57
+ `CAST(<field> AS VARCHAR(255)) <> '<value>'`.
58
+
59
+ The SQL route compares the untrimmed text: `TRIM(<field>)` answers with the same ArcGIS error on all three
60
+ layers, so the "trimmed and upper-cased" comparison belongs in a row filter in code.
61
+ `UPPER(CAST(<field> AS VARCHAR(255)))` runs on all three. Each layer reported its column type on
62
+ 2026-10-25: Harris `confidential_flag` is a string of length 10, Collin `GIS_DBO_AD_Entity_confidential_` a
63
+ string of length 1, Ector `conf_flg` a small integer. Harris's 470,972 blank cells hold spaces (they match
64
+ `LIKE ' %'`) and the clause keeps them. Separately, Harris has 5,235 NULL cells and Collin 15,003, which
65
+ 0.6.43 did not record; the clause drops them.
66
+
67
+ ### The site's owner-search examples for these counties carry the exclusion
68
+
69
+ On the Texas state page, the Harris owner-search examples (the hand-written FAQ and the state notes) now
70
+ request `confidential_flag` and carry the clause. The generated state FAQ does the same for any state whose
71
+ example county is flagged.
72
+
73
+ ### The tarball carries this changelog
74
+
75
+ `CHANGELOG.md` joins the package's published `files`; the 0.6.43 tarball had none.
76
+
77
+ ## 0.6.43 — 2026-10-24
78
+
79
+ One optional endpoint key and three Texas records that carry it (two measured, one a forward guess). No breaking change; `totals` are unchanged.
80
+
81
+ ### `confidentialFlag` on an endpoint record
82
+
83
+ `EndpointRecord.confidentialFlag?: { field: string; values: string[] }` (type `ConfidentialFlag`, exported)
84
+ names the county's own confidential-owner column on a layer and the values that mean "confidential". Texas
85
+ Tax Code §25.025 lets peace officers, judges, family-violence victims and others have their home address
86
+ kept confidential in appraisal records; Harris and Collin publish those accounts with the owner and mark them
87
+ in a flag column beside it. A row whose flag is set must not be shown or passed on: request the column
88
+ whenever you request the owner, drop a row whose cell, trimmed and upper-cased, equals a listed value or
89
+ (for a one-character value) contains it, and treat a row that lacks the column as flagged. The hosted REST
90
+ API, the hosted MCP and the site's free tools withhold such a row as `withheld: "county-redacted"`, exactly
91
+ like a row the county redacted itself, and never serve the flag column.
92
+
93
+ - **Harris County, TX** (`tx-harris`): `{ "field": "confidential_flag", "values": ["Y"] }`. The column is a
94
+ ten-character string with one Y/N per ownership slot. Measured 2026-10-24 over 1,551,736 rows:
95
+ `'Y '` 1,658 and `'YNNNNNNNNN'` 3,172 (4,830 flagged, every Y in slot 1), `'NNNNNNNNNN'` 1,066,977,
96
+ `'N '` 3,721, blank 470,972 (the county's "no flag recorded", not confidential). In a `where` clause:
97
+ `AND confidential_flag NOT LIKE '%Y%'` (superseded in 0.6.44 by the CAST form).
98
+ - **Collin County, TX** (`tx-collin`): `{ "field": "GIS_DBO_AD_Entity_confidential_", "values": ["T"] }`. The
99
+ layer's alias for the column is "Confidential Flag (t/f)". Measured 2026-10-24: `'T'` on 1,598 of 437,381
100
+ rows, `'F'` on the rest. In a `where` clause: `AND GIS_DBO_AD_Entity_confidential_ <> 'T'` (corrected in 0.6.44:
101
+ `AND CAST(GIS_DBO_AD_Entity_confidential_ AS VARCHAR(255)) NOT LIKE '%T%'`).
102
+ - **Ector County, TX** (`tx-ector`): `{ "field": "conf_flg", "values": ["1"] }`, a forward GUESS, not a
103
+ measurement: its `conf_flg` column (a small integer) was 0 on all 77,042 rows (2026-10-24), so no value
104
+ that means "confidential" has been observed; the marker reads a future non-zero cell as flagged. The
105
+ record's `notes` say so.
106
+
107
+ `validate-atlas` requires the field to read as a confidentiality flag (`conf`), not as an owner, mailing,
108
+ situs or id column, and the values to be in their trimmed upper-case form; every registration of one layer
109
+ URL must carry the same marker.
110
+
111
+ ## 0.6.42 — 2026-10-24
112
+
113
+ One county joins, one county is recorded as unreachable, two county records are retired, the county office's
114
+ email is on both answer types, the county record gains a `credit` field that every surface shows, the endpoint
115
+ record gains a `sharedService` marker on four statewide services, and the San Francisco city datasets move
116
+ host. No type is removed or renamed (`credit`, `county_office.email` and `sharedService` are additions), but
117
+ two records lose their endpoint: `wa-thurston` and `sc-charleston` now carry `hasPublicRest: false` and an empty
118
+ `endpoints` array (`coverage.reason: "licence-restricted"`), so a consumer that read either county's
119
+ `endpoints[0]` finds none and a `coverage` answer instead. `totals` move from 499 counties / 497 endpoints / 470
120
+ with an endpoint (0.6.41 moved none) to 501 / 496 / 469: San Francisco and Laurel join (two counties, one
121
+ endpoint), Thurston and Charleston lose theirs (two endpoints, two counties with one).
122
+
123
+ ### New: San Francisco, California
124
+
125
+ `ca-san-francisco` (06075), from SFGIS's daily ArcGIS Online copy of the DataSF dataset "Parcels - Active and
126
+ Retired" (org Zs2aNLFN00jrS4gG, item f81b5f0adb8c4d6a95282bec1ce10378): `Active_Parcels_from_DataSF_pulled_daily_`
127
+ FeatureServer layer 0, 227,695 active parcels on 2026-10-20, refreshed daily, CORS open, pages
128
+ (maxRecordCount 2000, resultOffset accepted). THE LAYER PUBLISHES NO OWNER NAME AND NO OWNER MAILING ADDRESS:
129
+ the Assessor-Recorder says state law bars it from posting ownership online, so the record carries
130
+ `capabilityOverrides.owner_name: not_published` and an `ownerFieldNote`. The parcel number is `blklot` (block
131
+ plus lot, text, letter suffixes possible); the situs is split into `from_address_num`/`to_address_num` and
132
+ `street_name`/`street_type`, so an address lookup composes House Number plus Street Name, and a parcel whose
133
+ range spans several house numbers matches only at its low end. Licence PDDL (the ArcGIS item's and DataSF's).
134
+ `credit: "City and County of San Francisco (DataSF, SFGIS)"`. The `notes` name the fallback, SF Planning's
135
+ PlanningData MapServer layer 23. `relatedPropertyCitySlug: "san-francisco"` links the record to the city's
136
+ permit and violation datasets, whose registry entry now carries `countySlug: "san-francisco-county"`.
137
+
138
+ ### Recorded as unreachable: Laurel County, Kentucky
139
+
140
+ `ky-laurel` (21125) ships with `hasPublicRest: false` and `coverage.reason: "no-public-service"`. The PVA
141
+ publishes parcels only through its qPublic website (laurelpva.com), which has no query service and answers
142
+ automated requests with a browser challenge; the state holds Laurel's parcels "Without Ownership" and has no
143
+ statewide layer yet. The record answers a search for the county on 2026-10-19 with the PVA's site and phone
144
+ number, so `findCounty()` resolves it and `coverageAnswer()` says why there is nothing to query.
145
+
146
+ ### Retired: Thurston County, Washington
147
+
148
+ `wa-thurston` now has no endpoint (`hasPublicRest: false`, `coverage.reason: "licence-restricted"`). The
149
+ county's reply of 2026-10-19 set out its data terms as non-negotiable: term 8 bars making the data available to
150
+ third parties naming Thurston County as the source, and term 10 bars commercial owner mailing lists
151
+ (https://www.thurstoncountywa.gov/map-viewer-disclaimer/disclaimer-data). The endpoint is withheld from the
152
+ atlas; the record keeps its GeoData Center contact and now carries the county's email,
153
+ GeoData@thurstoncountywa.gov, which is where to ask the county to allow it. Owner names were never on the
154
+ county's public layer, so the mailing-only set (Pima AZ, Riverside CA) is what remains. The hosted
155
+ county-access floor moves 53067 to `all-off` the same day, so the hosted API and MCP answer `paused` for it on
156
+ every surface.
157
+
158
+ ### Retired: Charleston County, South Carolina
159
+
160
+ `sc-charleston` (45019) now has no endpoint either (`hasPublicRest: false`, `coverage.reason:
161
+ "licence-restricted"`). Charleston County GIS told us on 2026-10-21 that it "does not authorize this use of our
162
+ data", so the endpoint is withheld from the atlas; the record keeps its GIS contact and carries the county's
163
+ email, WebMasterGIS@charlestoncounty.org, which is where to ask the county to allow it. The hosted county-access
164
+ floor moves 45019 to `all-off`, so the hosted API and MCP answer `paused` for it on every surface.
165
+
166
+ ### The county office's email
167
+
168
+ - `CountyPausedAnswer.county_office` and `CoverageAnswer.county_office` are now
169
+ `{ department, url, email }`, each `string | null`. The email is the record's `contact.email`
170
+ (`AccessRecordLike.contact.email?`, `CoverageRecordLike.contact.email?`); a record without one gives
171
+ `email: null`, the key always present. It is how a person asks the county for the data the answer
172
+ cannot give.
173
+ - `coverageAnswerText()` prints the email on the County office line after the url, so every text surface
174
+ (the hosted MCP, the `uks` CLI 0.1.8) shows it.
175
+ - Under `free_discovery`, `pausedAnswer()` reads the email by the rule the department and url already
176
+ follow: an address at one of the record's endpoint hosts is null, a plain address stays, and a retired
177
+ record (no endpoint) has no host to name, so its email is shown under every preset, `all-off` included.
178
+
179
+ ### `credit` on the county record, shown wherever the data is
180
+
181
+ - `CountyRecord.credit?: string`: the statement the publisher's terms require or request wherever its data
182
+ is shown, verbatim and never shortened. Four classes carry one. The ten Colorado counties on the State's
183
+ `Colorado_Public_Parcels` service (El Paso, Jefferson, Arapahoe, Adams, Douglas, Larimer, Boulder, Weld,
184
+ Mesa, Park) carry "State of Colorado Public parcels were accessed on 2026-10-17 from
185
+ https://gis.colorado.gov/public/rest/services/Address_and_Parcel". `ca-san-francisco` carries "City and
186
+ County of San Francisco (DataSF, SFGIS)". `mo-greene`, whose layer is the City of Springfield's, carries
187
+ the City's "Use of Data" disclaimer in full (https://www.springfieldmo.gov/disclaimer, now the endpoint's
188
+ `licenseUrl`), because the City requires any person, company or agency using data from its open data
189
+ systems to include it wherever the data is distributed. The same text stays in `notes` after `Credit:`.
190
+ `pa-berks` carries "Information is based on geospatial data from the County of Berks." (added 2026-10-24):
191
+ clause 4 of the County's data agreement requires that sentence on every product derived from its data, so
192
+ it goes with every Berks row, file and label sheet, the free single-parcel lookup included.
193
+ - Where it shows: with everything served from the record's layer. The hosted REST API carries it as
194
+ `county.credit` on `/api/atlas/lookup` and `/api/atlas/county`, as `credit` on every `/api/atlas/bulk` row
195
+ the county's layer answered, and as `county.credit` on a covered `/api/atlas/radius` answer. The hosted MCP
196
+ carries it inside `county` on `get_county_endpoint` and `find_county_by_fips` (the whole record), as
197
+ `atlas_match.credit` on `find_parcel_by_address`, per county on `list_counties_in_state`, as `credit` on an
198
+ `enrich_address` answer, and as `county.credit` on `radius_owners`. The site prints "Data credit: ..."
199
+ beside the rows on Parcel Lookup (a pasted credited layer included, and in its PDF), Find My Parcel Number,
200
+ Owner Lookup, the radius notice tool (and a `Data credit` column in its label CSV) and bulk enrichment (and
201
+ a `Data credit` column in its CSV); the ArcGIS to GeoJSON converter writes it as a top-level `credit` member
202
+ of the FeatureCollection; the county page shows it under its own heading. A withheld record carries none,
203
+ since it shows no data; the paid rows still served for a county whose record is withheld from discovery
204
+ carry it unless it names the county's service (Colorado's does, so it is null there). A consumer of this
205
+ package reads `credit` and shows it as is.
206
+ - The validator's cap on `credit` is 600 characters (`CREDIT_MAX` in `scripts/atlas-rules.mjs`), bounding a
207
+ paragraph rather than a sentence, because Springfield's statement is 452 characters and a required statement
208
+ is never truncated.
209
+ - A `credit` holds only WinAnsi characters (printable ASCII, U+00A0..U+00FF and the 27 Windows-1252 additions
210
+ such as curly quotes and dashes; `creditCharacterProblem` in `scripts/atlas-rules.mjs`), the set a PDF's
211
+ standard Helvetica can print, so the statement prints verbatim in every PDF export.
212
+
213
+ ### `sharedService` on the endpoint record
214
+
215
+ `EndpointRecord.sharedService?: true` (added 2026-10-24): one county registers a layer of the service, but the
216
+ record's own notes say the service covers other counties or the whole state, so its other layers may be any
217
+ county's. Four endpoints carry it: Chittenden VT (VCGI's statewide `FS_VCGI_VTPARCELS_WM_NOCACHE_v2`, "covers
218
+ all 14 Vermont counties"), Kauai HI (the State's `ParcelsZoning` MapServer, where Honolulu, Maui and Hawaii
219
+ County parcels are separate layers), Denver CO (`ODC_PROP_PARCELS_A`, a statewide composite) and Anchorage AK
220
+ (`AK_Parcels`, the statewide layer). The site's `findServiceCounty`, which places an unregistered layer of a
221
+ one-county service as that county's, places a layer of a marked service as no county's, and the registry
222
+ validator requires the marker whenever a record's prose says so. The site's free tools still ask the county
223
+ access switch about every county registered on a marked service before reading one of its layers (a pause on
224
+ Chittenden holds VCGI's other layers): the data is at least that county's, though the layer gets no county and
225
+ no credit. Absent is the other value; `false` is never written. The other 178 single-registration services were
226
+ surveyed (their service metadata and layer lists fetched) and none other covers another county.
227
+
228
+ ### San Francisco city datasets: data.sf.gov
229
+
230
+ The two DataSF Socrata datasets in `city-property/san-francisco.json` (building permits `i98e-djp9`, notices
231
+ of violation `nbtm-fbw5`) now point at `data.sf.gov`. The old host, `data.sfgov.org`, answers 301 with
232
+ `Access-Control-Allow-Origin: https://data.sfgov.org`, which a browser refuses, so the site's fetches failed;
233
+ the new host answers 200 with `Access-Control-Allow-Origin: *` (verified 2026-10-20 with `$limit=1` and an
234
+ Origin header on both ids). The registry validators now reject `data.sfgov.org` anywhere in a registry file.
235
+
236
+ ### Package metadata
237
+
238
+ Metadata only. The package author and the license holder now name UrbanKit Studio LLC.
239
+
240
+ ## 0.6.41 — 2026-10-24
241
+
242
+ New Jersey: owner names and owner mailing addresses are withheld for all six counties. Data change only, no
243
+ code or API change, and no county joins or leaves (`totals` do not move).
244
+
245
+ - **All six NJ counties** (Bergen 34003, Camden 34007, Essex 34013, Monmouth 34025, Morris 34027, Ocean 34029)
246
+ now carry `capabilityOverrides.owner_name` and `capabilityOverrides.owner_mailing_address` as `restricted`,
247
+ citing Daniel's Law (N.J.S.A. 56:8-166.1, P.L. 2020, c.125, as amended), which restricts disclosure of
248
+ the home addresses of judicial officers, law enforcement officers, child protective investigators and
249
+ prosecutors, and their household family members. A parcel layer does not identify which parcels belong to
250
+ those persons, so the package leaves both roles out. `ownerFieldNote` is added on Monmouth and extended on
251
+ the other five.
252
+ - The mailing columns (`ST_ADDRESS`, `CITY_STATE`, `ZIP_CODE`) are dropped from `searchFields` and
253
+ `sampleQuery` on Bergen, Camden, Essex, Morris and Ocean. On Monmouth the owner columns (`Owner_Name`,
254
+ `Owner_Street`, `Owner_Csz`) are dropped, and `sampleQuery` now searches `Location` instead of
255
+ `Owner_Name`. Anything that reads these records to build an owner or mailing query for an NJ county now
256
+ finds no such field and a `restricted` status.
257
+ - Monmouth's earlier note described Daniel's Law as covering domestic-violence victims and cited
258
+ P.L. 2021, c.371. Both are corrected: the statute covers the groups above and is cited as above.
259
+ - Consumers of this package (the stdio MCP server `@urbankitstudio/mcp-atlas` among them) pick the change up
260
+ on a dependency bump to 0.6.41.
261
+
262
+ ## 0.6.40 — 2026-10-18
263
+
264
+ Five counties join, three county records move and one sentence is added. No breaking change. `totals` move
265
+ from 494 counties / 494 endpoints / 467 with an endpoint to 499 / 497 / 470. North Carolina now carries all
266
+ 100 of its counties (Hoke and Swain were the two the 0.6.37 release left out), Ohio eleven and Illinois ten.
267
+
268
+ ### New counties
269
+
270
+ - **Rock Island County, IL** (`il-rock-island`, 17161), from the county GIS department's ArcGIS Online
271
+ feature service (services9.arcgis.com, item 9cae8a64ab0e4cea99758f741ca43b3c, "Parcels", owner
272
+ thirty34jb, the county GIS account): `Parcels` FeatureServer layer 0, `parcel_layer`, 65,948 rows, last
273
+ edited 2026-10-01, tax year 2025. The county's own host, gis.rockislandcountyil.gov
274
+ (`Hosted/Parcels/FeatureServer/0`), now answers "Token Required", so the record points at the ArcGIS
275
+ Online copy. Searchable by `parcel_number`, `owner1_name`, `taxbill_name`, `site_address`, `Site_City`
276
+ and `owner1_address1`; the layer also carries the owner and tax-bill mailing addresses, the assessed
277
+ values (EMV, EAV, farm and non-farm land and building), the last sale, class, township, legal
278
+ description, zoning, year built and square footage. 65,763 rows carry a parcel number and every one of
279
+ them an owner name; 58,640 carry a site address. The native spatial reference is Illinois State Plane
280
+ West, feet (WKID 3436), so a point query passes `inSR=4326`; the layer pages, declares
281
+ `maxRecordCount: 2000` and answers with CORS open. `license: "public"`: the item's license text reads
282
+ "For use by the general public", and the county's Data Release Policy is as-is and no-warranty with a
283
+ source credit of Rock Island County GIS, no resale ban and no fee stated. Credit Rock Island County GIS.
284
+ Verified 2026-10-18.
285
+ - **Hoke County, NC** (`nc-hoke`), from the county's ArcGIS Server
286
+ (`maps.hokecounty.org`, `OperationalLayers` MapServer layer 6, "Parcels",
287
+ 29,570 rows). The NC OneMap statewide layer has no owner or address values for
288
+ Hoke. Searchable by `TWN_PIN`, `OWNER_NAME` and `PHY_ADDRES`; it also lists the
289
+ second owner name, the owner's mailing lines (`MAILING_AD`, `MAILING__1`) with
290
+ `CITY`, `STATE_OR_C` and `ZIP_CODE`, calculated acres, `COST_TOTAL` (the layer
291
+ gives no definition) and the deed date. The layer has no site-city or site-ZIP
292
+ column: `CITY`, `STATE_OR_C` and `ZIP_CODE` are the owner's mailing city, state
293
+ and ZIP. It pages and declares `maxRecordCount: 1000`. 95 rows have no owner,
294
+ and one parcel number repeats. `license: "unknown"`: the county publishes no
295
+ license text, only a liability notice on its GIS page.
296
+ - **Swain County, NC** (`nc-swain`), from the county GIS office's ArcGIS Online
297
+ feature service (`Swain_County_Parcels` FeatureServer layer 0,
298
+ `Parcel_Polygons_Public`, 13,554 rows, last edited 2026-09-29). The NC OneMap
299
+ statewide layer has no parcel numbers for Swain. Searchable by `PARCEL_ID`,
300
+ `Name__1` (the owner name, which the layer does not define) and
301
+ `ParcelPropertyAddress`; it also lists the second name, the owner's mailing
302
+ lines and city, state and ZIP (the layer has no site city), acres, land,
303
+ building and total assessed value, and the deed date. 599 parcel numbers repeat
304
+ across rows (798 extra rows, at most 45 under one number) and 46 rows have none.
305
+ `license: "unknown"`: the county item's license text is a warranty disclaimer
306
+ that places no limit on use.
307
+ - **Licking County, OH** (`oh-licking`), from the county's own ArcGIS Server
308
+ (`gis.lickingcounty.gov`, `Auditor/Parcels` MapServer layer 0, 83,782 rows; the
309
+ latest transfer is dated 2026-10-15). Searchable by `Parcel`, `OwnerName` and
310
+ `SiteAddress`; it also lists the site city and ZIP, the four owner mailing
311
+ columns, municipality, township, tax acres, land use, property class, market
312
+ and net total value, year built, and the latest transfer date and amount. 1,189
313
+ rows are shapes with no auditor record, and every parcel number is unique. The
314
+ layer pages and declares `maxRecordCount: 2000`. `corsEnabled: true`: on
315
+ 2026-10-17 the server echoed the request origin on the layer and on `/query`,
316
+ though an earlier probe the same day found no CORS header. `license: "unknown"`:
317
+ the county's GIS downloads page carries a reference-only, no-warranty
318
+ disclaimer and states no limit on redistribution.
319
+ - **Madison County, OH** (`oh-madison`), from the County Engineer's ArcGIS Online
320
+ feature service (`Parcels_Engineer` FeatureServer layer 13,
321
+ `Parcels_Parcels_join`, 25,931 rows, last edited 2026-10-15), which joins the
322
+ Engineer's parcels to the Auditor's. Searchable by `TAXPIN` (the auditor's
323
+ parcel number; `PARCEL_NO` is the Engineer's own sequence number and repeats on
324
+ up to 223 rows, so it is not listed), `NAME` (the owner), and the site address,
325
+ which the layer splits into `Number` and `Street` (the bare street name, with no
326
+ street type). It also lists the site city and ZIP (blank on 40.4% of rows), the
327
+ owner's mailing street, city, state and ZIP (`Owner_Cont`, `Owner_Co_1`,
328
+ `Owner_Co_4`, `Owner_Co_5`; the layer's `Taxpayer_*` columns are a different,
329
+ tax-bill address and are not listed), acres, the transfer date and the tax
330
+ district. 287 rows carry the placeholder parcel numbers 99-99999.999 and
331
+ 88-88888.888 and no owner, and 1,600 other parcel numbers repeat across rows, so
332
+ a parcel-number lookup can return several rows. `license: "unknown"`: the
333
+ parcel item and the county's GIS Data Download hub state no terms, while two
334
+ other items in the same ArcGIS Online organization (the Comprehensive Plan
335
+ layers) carry a notice ending "Reproduction in any form is prohibited without
336
+ the expressed consent of Madison County"; the record's notes quote both.
337
+
338
+ The four records probed on 2026-10-17 carry in their `notes` the share of rows filled for every column the
339
+ atlas lists, read that day, and the extent and county-seat checks (each layer's extent falls inside its
340
+ county, and a point or box at the county seat intersects parcels). `lastVerified` is the probe date, not a
341
+ data date: none of those four layers dates its roll. Each answered one bounded query again on 2026-10-18.
342
+
343
+ ### Records moved
344
+
345
+ - **Fulton County GA (13121) repointed.** The county's on-premises server (gismaps.fultoncountyga.gov,
346
+ PropertyMapViewer layer 11) stopped accepting connections on 2026-10-18. The record now points at Fulton County
347
+ GIS's own ArcGIS Online copy of the tax parcels, `Tax_Parcels/FeatureServer/0` on services1.arcgis.com (item
348
+ e581a072dca9442e884d3682bff03484, owner Fulton_County_GIS, public, last edited 2026-09-22): 373,306 parcels,
349
+ TaxYear 2026, the same nine fields with the same labels, maxRecordCount 1,000, pagination on, CORS open. The
350
+ 42 rows whose owner reads RESTRICTED stay withheld. Use the unsuffixed `Tax_Parcels` service, not the
351
+ `Tax_Parcels_2025` and `Tax_Parcels_2024` snapshots beside it.
352
+ - **Greenville County SC (45045) retired, with the reason.** `hasPublicRest: false`, `endpoints: []`,
353
+ `coverage.reason: "owner-data-sold"`, checked 2026-10-18: the county removed every public map service from
354
+ gcgis.org (the services root lists none; the registered layer answers "Service not found") and sells its tax
355
+ parcels with ownership as countywide vector data on a USB drive, $500 in shapefile format, under a data order
356
+ form and release agreement; `coverage.countyProduct` carries that page. Owner lookups are on the county's
357
+ Real Property Search, and no statewide South Carolina layer carries owner names. The record had been
358
+ `unreachable` since 2026-08-29.
359
+ - **Montgomery County KY (21173) retired, with the reason.** `coverage.reason: "login-required"`, checked
360
+ 2026-10-18: around 2026-10-15 the county's GIS host, the Gateway Area Development District, made the
361
+ `Montgomery_Parcels` FeatureServer private, and the layer now answers "Token Required". The organisation's
362
+ public layers carry no owner names. `contact.url` is the county's site. The county's KRS 61.874 pause on paid
363
+ surfaces is live data from the hosted API and is unchanged.
364
+ - **`countyDownMessage(countyName)`**: the one sentence every surface gives when a county's own GIS server is
365
+ down (the liveness probe observed `down`, the operator paused the county with reason `upstream-broken`, or a
366
+ query failed on the county's side): "Fulton County's GIS server is down on the county's end, so its parcel data
367
+ can't be reached right now. Try again later, or use the county office's own site." Without a name it reads
368
+ "The county's GIS server…". It says nothing about money: whether a call was charged is that call's fact, so each
369
+ surface appends its own charge clause. Exported from the package entry. On the hosted REST API, a lookup on
370
+ a county whose own server the liveness probe currently reads as down is served with this sentence and is
371
+ not charged on any tier (Leo, call 11, 2026-10-18); a reading older than six hours, or none, charges as
372
+ before.
373
+ - **`coverageOverridesPause(record)`**: true for a record with no endpoint whose `coverage.reason` is
374
+ anything but `licence-restricted`. The hosted API's lookup, county and MCP discovery tools answer such a
375
+ county with its `coverage` block even while the county access switch pauses it: the pause exists for a
376
+ county we could serve, and this one has nothing to serve, so the record's own reason is the honest answer
377
+ (Montgomery KY). A licence-restricted retired county (Shelby TN) keeps the paused answer, and a county
378
+ whose endpoint returns is paused again with no edit to the switch. A contradictory record
379
+ (`hasPublicRest: false` beside a non-empty `endpoints` array, a shape `validate-atlas` rejects) keeps the
380
+ pause too, so an override never serves endpoints the record itself lists. Exported from the package entry.
381
+
382
+ ## 0.6.39 — 2026-10-18
383
+
384
+ The paused-county answer. No breaking change, no data change, `totals` unchanged.
385
+
386
+ A county can be in the atlas, with a working endpoint, and still be paused for some uses: its own terms
387
+ forbid commercial use, a state law lets the county require a contract, or the county asked. The hosted
388
+ API now answers for such a county with a `coverage` block whose `status` is `"paused"`, and this release
389
+ exports that shape and the function that builds it:
390
+
391
+ - `ACCESS_FLAGS`: the seven surfaces a county can be paused on, each on its own: `paid_lookup` (REST lookup by a
392
+ paying caller, PAYG or a paid plan), `paid_enrich` (MCP `enrich_address`), `paid_bulk` (REST bulk rows),
393
+ `paid_lists` (REST radius, MCP `radius_owners`), `free_site_lookup` (the site's free lookup tools, and REST
394
+ lookup by an anonymous or free-tier caller), `free_site_lists` (the site's free whole-layer download) and
395
+ `free_discovery` (the county record beyond its identity in county metadata and the MCP discovery tools).
396
+ - `ACCESS_PRESETS` and `PRESET_FLAGS`: `"paid-off"` (the four paid surfaces), `"paid-lists-off"` (`paid_bulk` and
397
+ `paid_lists`; single-parcel lookups still work) and `"all-off"` (all seven; the hosted API and MCP answers
398
+ serve only the record's identity: no endpoint URLs, notes or other fields). `flagsFor(preset)`, `pausedFlags(flags)` and `presetOf(flags)` (`"custom"` for
399
+ any other set) convert.
400
+ - `free_discovery` governs the hosted API and MCP answers only. The endpoint URLs in this package's bundled data
401
+ are governed by the data release: a county whose endpoints must not be published becomes a no-endpoint record
402
+ there (0.6.37 removed `tn-shelby`'s endpoint).
403
+ - `ACCESS_REASONS`: `licence-terms`, `state-law`, `county-request`, `data-quality`, `upstream-broken`, `other`.
404
+ - `accessBlocks(entry, flag)`: whether that surface is paused for the county (no entry: never).
405
+ - `pausedAnswer(entry, record, page)` and the `CountyPausedAnswer` type: `indexed: true`, `status: "paused"`,
406
+ `preset`, `paused` (the paused surfaces), `reason`, `why`, `citation_url`, `since`, `county_office`,
407
+ `free_alternatives` (the free county page, null when the free lookup or discovery is paused) and `request: null`.
408
+ **`CountyPausedAnswer.why` is `string | null`, as is `citation_url`**, although the operator's
409
+ `CountyAccessEntry.why` is a `string`: a consumer that shows `why` must handle null (the hosted site shows a
410
+ generic sentence).
411
+ When `free_discovery` is paused, a `county_office` department or url, the `why` or the `citation_url` that names
412
+ one of the record's endpoint hosts or an ArcGIS REST path (a `rest/services` path segment, so not
413
+ `forest/services`, a `sharing/rest` path, an `arcgis.com/home/item` page, or a Map, Feature, Image, GP,
414
+ VectorTile or Scene server) is null; pass the record with its `endpoints` (`AccessRecordLike.endpoints`). The
415
+ value is read with its percent-escapes decoded, in several forms: plain; folded (NFKC, lower case, the
416
+ characters a reader does not see dropped: format characters such as a soft hyphen or a zero-width space, and
417
+ the other default-ignorables such as a combining grapheme joiner or a variation selector; fullwidth and
418
+ ideographic stops as dots); and each again with its control characters deleted, as the URL parser deletes a
419
+ tab or newline. A url a withheld answer links (the `citation_url`, the office url) is also read as the URL
420
+ parser makes it, the link a browser follows. Two passes over the forms decide it, and either one nulls. The
421
+ first parses each dotted name as a host (in the URL parser's ASCII form, without a port or a trailing dot): it
422
+ names the service when it equals an endpoint host (both without a leading `www.`) or is a name under one
423
+ (`fiscalhub.gis.` beside `gis.`). The second looks for an endpoint host without its `www.` with no ASCII
424
+ letter or digit on either side, so any punctuation, symbol, space, `-` or `_` beside it still names it. An
425
+ endpoint host is matched as an ASCII key, so every endpoint host in the atlas is plain ASCII (no `xn--` label,
426
+ no `%`); an IDN host would need the matcher extended first. A parent host (the
427
+ county's own domain beside its `gis.` server) or a sibling of a plain host (an `assessor.` beside `gis.`)
428
+ stays; it fails closed on a name under a `www.` endpoint's domain (`maps.` beside `www.`) and on a longer name
429
+ that begins with the host. Otherwise the office contact stays: a hub or open-data portal, or a department
430
+ named after its GIS organisation, is how a paused county's office is reached. A value over 2,048 characters,
431
+ before or after folding, is null.
432
+
433
+ Which counties are paused is live data, not bundled: read `GET https://urbankitstudio.com/api/atlas/status`
434
+ (its `paused` list) or `?fips=` for one county. A pause or un-pause therefore needs no release of this package.
435
+
436
+ Fixed: `countySlugFromName` drops an apostrophe (straight `'`, `’` U+2019 or `ʼ` U+02BC) before slugging, so
437
+ `"Prince George's"` gives `prince-georges-county`, the slug the atlas stores, not `prince-george-s-county`, which
438
+ `findCounty` does not find. `requestHint` slugs a county name by the same rule. Every bundled county name now gives
439
+ its stored slug (494 of 494; it was 493). Found through mcp-atlas.
440
+
441
+ ## 0.6.38 — 2026-10-17
442
+
443
+ This release gives one answer for a county the atlas cannot serve. Every record
444
+ with `hasPublicRest: false` now carries a `coverage` block with a reason code,
445
+ one sentence and the date we last checked. `coverageAnswer()` turns that record,
446
+ or the lack of one, into the answer the hosted MCP server, the REST API and the
447
+ `uks` CLI give, with a ready-to-send request for the county. Fifteen counties
448
+ are newly recorded with their reason, and 27 records now say why they are not
449
+ served. The SDK lookups also accept a two-letter state abbreviation. Every
450
+ change is additive. No existing field, export or return value changes.
451
+
452
+ `totals` move from 479 counties and 494 endpoints to 494 counties and 494
453
+ endpoints. 467 counties have a verified endpoint, as before.
454
+
455
+ ### The coverage answer
456
+
457
+ - `CountyRecord.coverage?: CountyCoverage | null` holds `reason`, one of
458
+ `COVERAGE_REASONS` (`owner-data-sold`, `no-public-service`, `login-required`,
459
+ `licence-restricted`, `attributes-missing`, `owner-blank`, `service-down`,
460
+ `data-stale`, `technical`), plus `summary`, one sentence of at most 280 characters written
461
+ to be shown as is, and `checked`, the YYYY-MM-DD of the last check. Two keys
462
+ are optional. `recheck` is the YYYY-MM-DD of the next planned check (Columbia
463
+ PA carries one). `countyProduct` names the county's own paid channel, and only
464
+ with `owner-data-sold`, never a reseller.
465
+ - `licence-restricted` (additive) means the county publishes a parcel
466
+ service, but its terms do not allow us to serve it. Shelby TN carries it.
467
+ - `countyHasEndpoint(record)` is the one "is this county served" rule:
468
+ `hasPublicRest` is true and at least one endpoint is registered.
469
+ - `coverageAnswer(record, ref)` returns null for a served county. Otherwise it
470
+ returns a `CoverageAnswer` with `indexed: false` and a `status`.
471
+ `"not-in-atlas"` means the atlas has no record, and says nothing more, since a
472
+ misspelled name lands there too. `"no-public-data"` means the atlas records
473
+ the county and it publishes nothing we can serve. The block also carries
474
+ `reason`, `why`, `checked`, `recheck`, `county_office` (from the record's
475
+ `contact`), `county_product`, `request`, and `message`, which is the whole
476
+ answer as text.
477
+ - `coverageAnswerText(block)` renders that text, one point per line, and it is
478
+ what every text surface prints verbatim. `NOT_IN_ATLAS_WHY` and
479
+ `NO_PUBLIC_DATA_WHY` are its two fallback sentences.
480
+ - `requestHint(ref, status?)` builds the `request` block: a body for
481
+ `POST https://urbankitstudio.com/api/atlas/request` (`COUNTY_REQUEST_URL`) and
482
+ the hosted `request_county` tool, with `COUNTY_REQUEST_MESSAGE` or
483
+ `COUNTY_RECHECK_REQUEST_MESSAGE`. `ref` is
484
+ `{ state, county?, fips? }`. The state can be an abbreviation, a name, a slug
485
+ or a two-digit FIPS code, and any spelling of one county gives the same body.
486
+ It returns null when the ref names no county the endpoint accepts (no valid
487
+ FIPS code, and a state or county with nothing to slug).
488
+ - All of these are pure, with no fetch, no clock and no randomness, so one
489
+ record and ref give the same answer on every surface. The hosted MCP server,
490
+ the REST API and the `uks` CLI 0.1.4 run copies of this same module.
491
+ - Types exported: `CoverageReason`, `CoverageStatus`, `CountyCoverage`,
492
+ `CountyProduct`, `CoverageAnswer`, `CoverageTextInput`, `CoverageRecordLike`,
493
+ `EndpointPresenceLike`, `CountyRequestRef`, `CountyRequestHint`.
494
+ - On the hosted REST API, a plan-key lookup that lands on a county we do not
495
+ serve no longer draws from the key's quota: the draw is given back, and the
496
+ answer says "not billed" only when the credit landed.
497
+
498
+ ### Newly recorded counties (15, no endpoint)
499
+
500
+ - `owner-data-sold`: Lancaster PA (42071), and Kenton (21117), Campbell (21037)
501
+ and Pendleton (21191) KY, whose parcels LINK-GIS publishes and sells.
502
+ - `attributes-missing`: Columbia PA (42037), Trimble (21223) and Hardin (21093)
503
+ KY, Pueblo CO (08101), Pierce WA (53053), Loudoun VA (51107) and Forsyth GA
504
+ (13117).
505
+ - `no-public-service`: Knox (47093) and Rutherford (47149) TN, which
506
+ Tennessee's statewide parcel data leaves out.
507
+ - `data-stale`: Williamson TN (47187), which the statewide data also leaves
508
+ out, and whose only public parcel layer is a copy from 2003.
509
+ - `technical`: Scott KY (21209).
510
+
511
+ ### Coverage blocks on existing no-endpoint records (12)
512
+
513
+ Winnebago IL, Casey KY, Iron MI, Hillsborough and Rockingham NH, Dutchess NY,
514
+ Garfield OK, Providence RI, Webb TX, Jackson MO, Montgomery OH and Shelby TN.
515
+ The two New Hampshire records are also corrected. NH GRANIT's statewide parcel
516
+ mosaic is public, with no login, but it has no owner field, so both are
517
+ `attributes-missing`. Shelby TN is `licence-restricted`, because ReGIS
518
+ publishes a parcel service whose terms bar public display without its
519
+ written consent.
520
+
521
+ - tn-shelby: notes no longer link the county's ArcGIS service. They cite the
522
+ county's ReGIS page (https://www.shelbycountytn.gov/467/REGIS-Regional-GIS)
523
+ and keep ReGIS's quoted terms, and no field of the record names the service's
524
+ host or path.
525
+
526
+ ### State abbreviations in the SDK lookups (additive)
527
+
528
+ `findState`, `findCounty` and `listCountiesByState` still look a state up by
529
+ its slug, and a two-letter abbreviation is also accepted, in either case:
530
+ `findCounty("IL", "kane-county")` finds Kane. The 50 states and DC read as
531
+ their slug by the rule the hosted MCP server and the REST API use
532
+ (`stateSlugFromAbbrev` in the coverage module). `"pr"`, `"gu"` and any other
533
+ input are read as a slug unchanged, so they find nothing, as before. Slugs stay
534
+ exact: `findState("Illinois")` is still undefined.
535
+
536
+ ## 0.6.37 — 2026-10-17
537
+
538
+ This release adds 83 North Carolina counties, all from the NC OneMap parcel
539
+ compilation. Hoke and Swain are left out. The statewide layer carries no owner
540
+ or address values for Hoke and no parcel numbers for Swain. The release also
541
+ adds Union County OH, from the county auditor's service. Martin NC and Union OH
542
+ were the demand triggers. This is a data release with no API change.
543
+
544
+ `totals` move from 395 counties and 413 endpoints to 479 counties and 494
545
+ endpoints, and 467 counties have a verified endpoint. Four records that pointed
546
+ at another state's layer are corrected, and two of them now have no endpoint.
547
+ Shelby TN's endpoint is withheld (see below). North Carolina now carries 98 of
548
+ its 100 counties, and Ohio carries 9.
549
+
550
+ ### New counties
551
+
552
+ - **83 North Carolina counties**, all on NC OneMap's statewide parcel layer
553
+ (`NC1Map_Parcels` MapServer layer 1 on services.nconemap.gov, 5,938,900 rows
554
+ across the 100 counties), in the same shape as the ten counties already
555
+ registered on it (Forsyth, Guilford, Cumberland, Montgomery, New Hanover,
556
+ Iredell, Union, Onslow, Johnston, Brunswick): the same URL, scoped per county
557
+ with `scopeWhere: "cntyname = '<County>'"`, and no `supportsPagination` or
558
+ `maxRecordCount` key (the layer pages, 5,000 rows per request). Each
559
+ `sampleQuery` is a county-scoped `parno` equality (2.0 to 2.8 s on release
560
+ day). Chowan's rows appear doubled: all eight parcel numbers checked match two
561
+ rows each, so its sample returns two.
562
+ - Columns are listed from coverage measured on release day, counting a value of
563
+ only spaces as blank: `siteadd` (searchable), `scity`, `szip`, `parval`,
564
+ `gisacres` and `parusedesc` where at least 50% filled. Seven counties have no
565
+ site address in the statewide layer (Avery, Bladen, Cabarrus, Franklin,
566
+ Orange, Perquimans, Richmond) and carry a reviewed
567
+ `situs_address: not_published` record; search them by owner or parcel number.
568
+ - Mailing columns (`mailadd`, `mcity`, `mstate`, `mzip`) are listed where
569
+ `mailadd` is at least 80% filled and a 25-row sample was street-like on at
570
+ least 90% of rows (four counties at 22 of 25 passed a 50-row resample from a
571
+ later objectid window): 73 counties. The other ten carry a reviewed
572
+ `owner_mailing_address: not_published` record that says why.
573
+ - `license: "public"` with the NC OneMap home page as `licenseUrl`, as on the ten
574
+ counties already registered; the `notes` cite NC OneMap's terms page (a free
575
+ and unrestricted use policy, no warranty), credit the county tax assessor's
576
+ office and NC OneMap (NC CGIA), give each county's measured fill rates, and say
577
+ that no field dates a county's roll: `transfdate` is the state's transform
578
+ date, and Hyde's (2025-03-25) and Clay's (2025-04-17) are 18 months old or
579
+ more.
580
+ - **Union County, Ohio**, from the county's own ArcGIS Server
581
+ (`ParcelAllSaleinfoBuildingInfo` MapServer layer 1, 40,059 rows; sales through
582
+ 2026-09-21). Searchable by `ParcelNo`, `DeedOwner` and `Address`; it also lists
583
+ the site city, `ZipCode` (which holds "CITY OH ZIP", so it is labelled as the
584
+ site city, state and ZIP), the three mailing lines (the third is spelled
585
+ `Mailaddress3` on the layer), appraised value, acres, land use, and sale date
586
+ and price. The layer pages and declares `maxRecordCount: 2000`. 850 rows have
587
+ no owner (90 of them no parcel number either), and 1,045 parcel numbers
588
+ repeat across rows, so a parcel-number lookup can return several rows.
589
+ `license: "unknown"`: the county publishes no license text; its GIS data page
590
+ (the `licenseUrl`) offers the parcels as a free shapefile.
591
+
592
+ ### Corrected records (another state's layer)
593
+
594
+ An extent sweep found four records whose layer belongs to a same-named county in
595
+ another state. Each layer was live and returned real owner rows, so liveness
596
+ checks stayed green.
597
+
598
+ - **Jackson County, Missouri** (`mo-jackson`) pointed at Jackson County,
599
+ Michigan (`gis.mijackson.org`). The county's own current layer
600
+ (`jcgis.jacksongov.org`, `ParcelsAscendRelate` FeatureServer layer 1, 311,163
601
+ parcels) carries parcel ids only, so the record now has no endpoint
602
+ (`hasPublicRest: false`), as Dutchess NY does.
603
+ - **Montgomery County, Ohio** (`oh-montgomery`) pointed at Montgomery County, New
604
+ York. The Auditor's own layer (`gis.mcohio.org`, `AUDGIS_B1` MapServer layer 7)
605
+ carries owners, but only under qualified join names such as
606
+ `SDE.WEB_CAMA.OWNER_NAME1`, which the registry's field-name rule does not
607
+ accept, so the record has no endpoint until it does.
608
+ - **Summit County, Ohio** (`oh-summit`) pointed at Summit County, Utah. It now
609
+ uses the county's own `parcels_web_GEODATA_Tax_Parcels` FeatureServer layer 0
610
+ on `scgis.summitoh.net` (265,902 parcels): owner `ownernme1`, site address
611
+ `siteaddress`, mailing `pstladdress`, `pstlcity`, `pstlstate` and `pstlzip5`,
612
+ and the current market value. It declares `maxRecordCount: 3000`;
613
+ `license: "unknown"`, because neither the layer nor its items publish licence
614
+ text.
615
+ - **Fayette County, Kentucky** (`ky-fayette`) pointed at the Fairbanks North Star
616
+ Borough, Alaska. It now uses LFUCG's own `Parcel` FeatureServer layer 0
617
+ (114,433 parcels: `PVANUM`, site `ADDRESS`, unit, property class, acres). That
618
+ layer publishes no owner name, so `owner_name` carries a reviewed
619
+ `not_published` record on an `endpoint_schema` basis, as Jefferson, Madison,
620
+ Oldham and Henderson do.
621
+ LFUCG's Terms of Use apply to the layer; the notes quote their indemnity
622
+ clause (section VI.B) in full.
623
+
624
+ ### Withheld: Shelby County, Tennessee
625
+
626
+ `tn-shelby` now has no endpoint (`hasPublicRest: false`). The ReGIS parcel
627
+ service's item information says "This data may NOT be displayed publicly, sold
628
+ or transferred to any other entity without the written consent of ReGIS", so the
629
+ endpoint is withheld from the atlas pending that consent. The server also
630
+ requires unsafe legacy TLS renegotiation, which standard clients refuse. The
631
+ record keeps its ReGIS contact.
632
+
633
+ ### Credit lines
634
+
635
+ The registry has no attribution field yet, so each credit is a sentence at the
636
+ end of the record's `notes`, starting `Credit:`. The README's new Attribution
637
+ section lists the same credits.
638
+
639
+ - The ten Colorado counties on the State's `Colorado_Public_Parcels` service (El
640
+ Paso, Jefferson, Arapahoe, Adams, Douglas, Larimer, Boulder, Weld, Mesa, Park)
641
+ carry the attribution statement the service's copyright text names as
642
+ sufficient: "State of Colorado Public parcels were accessed on 2026-10-17 from
643
+ https://gis.colorado.gov/public/rest/services/Address_and_Parcel".
644
+ - Montgomery MD and Prince George's MD credit their county's Planning Department
645
+ (M-NCPPC), which asks for attribution.
646
+ - Wake NC (CC BY 4.0), Sonoma CA (CC BY-SA 3.0 US, "County of Sonoma"),
647
+ Sacramento CA and the District of Columbia (CC BY 4.0) credit the publisher
648
+ their licence names.
649
+ - Union OH's notes quote the county's GIS disclaimer (prepared for tax
650
+ appraisal, no liability, no limit on use).
651
+ - Montgomery MD's notes explain its 4.6-degree extent: 24 of 351,954 points share
652
+ one placeholder coordinate in West Virginia. The layer is the county's own, so
653
+ it takes no county scope.
654
+
655
+ ## 0.6.36 — 2026-10-14
656
+
657
+ City search only. No breaking change, `totals` unchanged.
658
+
659
+ ### Street typed with a house number (Los Angeles violations, Fort Worth permits, Scottsdale violations)
660
+
661
+ These tables keep the house number and the street name in separate columns. A typed address now
662
+ follows the same street rules as Dallas-style tables, with the number kept as its own clause:
663
+
664
+ - A spelled-out direction with its type is the street: `2600 South Blvd` and `100 North Ave` search
665
+ `South` and `North` (they searched `Blvd` and `Ave`).
666
+ - A lettered street keeps its letter: `1700 E Q St`, `100 Q St` and `100 N A St` search `Q`, `Q` and
667
+ `A`. A lone direction letter is a lettered street only in the two-token form before `St` or `Ave`
668
+ (`100 E St` searches `E`); with a direction before it (`100 N E Blvd`) the letter is the street.
669
+ - An abbreviated direction (`N`, `E`, `NE`) plus any other street type is a direction and a street:
670
+ `100 N Court` and `100 E Blvd` search `Court` and `Blvd`. A spelled-out direction is the street name
671
+ (`100 North Court` searches `North`).
672
+ - A number plus one stray token with no street type (`100 %`, `100 E`, `100 St`) is searched as typed,
673
+ never as a bare `%` or `E`. The name columns hold the bare street name, so that search finds no
674
+ rows rather than unrelated rows at that number.
675
+
676
+ ## 0.6.35 — 2026-10-14
677
+
678
+ City search normalization and one new county. No breaking change; one additive
679
+ optional field on the dataset record types (`houseNumberSpaceRuns`, below).
680
+ `totals` move to 395 counties and 413 endpoints (386 with a verified endpoint).
681
+
682
+ ### New county
683
+
684
+ - **Montgomery TN (47125)**, parcel id and situs address only: the county's
685
+ public parcel layer carries no owner or mailing fields, so owner lookups
686
+ return nothing there.
687
+
688
+ ### City search
689
+
690
+ `buildCityQuery` (and the hosted
691
+ `get_city_datasets` `query_url`) now build a more tolerant street match, read
692
+ live on 2026-10-14 against the five datasets that missed.
693
+
694
+ - **Double-spaced addresses (Orlando only).** A dataset opts in with the new
695
+ optional registry field `houseNumberSpaceRuns` (the longest run of spaces to
696
+ try, 1 to 8), set only where a table was measured to store such a run:
697
+ Orlando code cases, permits and zoning cases, at 4. For those, when a typed
698
+ street starts with a house number the query ORs the typed text with one to
699
+ four spaces after the number (`4274 Tatum St` also tries `4274 Tatum St`
700
+ through four spaces), and with the typed text itself when that has a longer
701
+ run or a tab, in parentheses so the dataset scope still applies. Every other
702
+ dataset keeps the one typed literal, exactly as before: the OR group measured
703
+ 1.6 to 1.9x slower on Chicago violations and Orlando permits, which is not
704
+ worth paying where the table has no such run. A group whose encoded length
705
+ would pass about 3,000 bytes falls back to the single literal. Spaces only: a word cannot
706
+ come between the number and the street, so `4274 Tatum St` does not find
707
+ `4274 OLD TATUM ST` and `12 Main` does not find `12 Old Main`. The bound
708
+ is the longest run measured upstream on 2026-10-14, with one step of
709
+ headroom: three in Orlando permits and zoning cases, two in code cases. Orlando
710
+ stores a run there on 99,433 of 146,669 code-case addresses, 55,657 of
711
+ 1,105,323 permit addresses and 221 of 407 zoning-case addresses, so a typed
712
+ full address returned nothing and agents were told to search the street name
713
+ alone. As before, the house number is not anchored on its left: `12 Main` also
714
+ finds `3312 MAIN`. (140,344 code-case addresses carry a double space
715
+ somewhere, mostly before the city name; 99,433 have the run right after the
716
+ number.)
717
+ - **Split-street datasets, street typed without a house number.** Los Angeles
718
+ violations (`stno`/`stname`), Fort Worth permits (`Addr_No`/`Street_Name`) and
719
+ Scottsdale violations (`StreetNumber`/`StreetName`) keep the bare name in the
720
+ address column, so `Funtier Ct` matched as typed found nothing while `Funtier`
721
+ found 87 permits. The name-only path now takes one leading direction and one
722
+ trailing street type off, as the number-plus-name path already did. A trim that
723
+ would leave a lone letter, only a direction or street type, or no letter or
724
+ digit (`N St`, `E St`, `N E St`, a typed `%`) is not applied: the street is
725
+ matched as typed. The derived street name an agent query sends on a Dallas-style
726
+ table (`streetForDataset`) follows the same rule: `N St`, `N Q St` and
727
+ `N % St` are sent as typed, not as `St`, `Q` or `%`, and `100 N St` sends `N St`
728
+ (the name column never holds a house number). A street whose name is itself a
729
+ type or direction word keeps its strip when a real type follows it: `100 Court St`
730
+ sends `Court`, `Plaza Dr` sends `Plaza`. A spelled-out direction with its type
731
+ (`2600 South Blvd`, `100 North Ave`, `South St`) is the street South or North
732
+ and sends `South` or `North`. That holds on tables with no number column and a
733
+ type column (Dallas violations, Boston, San Francisco) for every input, and on
734
+ tables with a number column (Los Angeles, Fort Worth, Scottsdale) only for a
735
+ street typed without a house number; a table with no type column (Chicago
736
+ permits) keeps `100 North St` as `North St`. `N St` and `NE St` stay as typed. A house number
737
+ plus one unsafe token (`100 %`, `100 _`, `100 E`, `100 St`) is sent as typed,
738
+ never as the bare token.
739
+ - **Linear parsing.** The house-number parsers no longer backtrack on a long run
740
+ of digits or of spaces followed by a newline.
741
+ - **Ordinal street names.** A first token like `88th` or `3rd` is no longer read
742
+ as a house number with a glued unit, so `88th St` typed without a number is a
743
+ street name on the split datasets and `3rd Ave` is not cut to `Ave` on Dallas
744
+ violations.
745
+ - Dataset `coverage` and `queryNote` text for these datasets states the new
746
+ behaviour; the "search by street name alone" workaround is gone.
747
+
748
+ ## 0.6.34 — 2026-10-14
749
+
750
+ Mississippi statewide: 80 new counties and a second endpoint for Hinds from the
751
+ MARIS/MDEQ 2024 cadastral layers. These are the first records to declare
752
+ `supportsPagination: false` with `maxRecordCount` 2000 and a county
753
+ `scopeWhere`. Data release, no API change. `totals` move from 314 counties and
754
+ 331 endpoints to 394 and 412 (385 with a verified endpoint). Mississippi is the
755
+ first state with every county on record, 82 of 82.
756
+
757
+ ### New counties
758
+
759
+ - **80 Mississippi counties**, each on one of the two MARIS/MDEQ 2024 statewide
760
+ cadastral MapServer layers on gis.mississippi.edu (East: 44 counties,
761
+ 1,127,050 parcels; West: 38 counties, 878,999 parcels), scoped per county with
762
+ `scopeWhere: "STCNTYFIPS = '28xxx'"`. Both layers refuse pagination
763
+ (`supportsPagination: false`, `maxRecordCount: 2000`; read `offersPagination()`
764
+ before adding either parameter), answer at most 2,000 rows per request with
765
+ `exceededTransferLimit` on a cut, and scan the whole layer on any predicate but
766
+ an FID lookup (1.3 s to more than 20 s, more on the first call after hours idle). Each
767
+ `sampleQuery` is a FIPS-scoped FID equality with no pagination parameter; a
768
+ FID read answers in under a second where a PARNO equality scans the layer.
769
+ - Mailing columns (MAILADD1, MCITY1, MSTATE1, MZIP1) are listed only where
770
+ MAILADD1 is at least 80% populated and a sample read on release day was
771
+ street-like on at least 90% of rows: twelve counties (Clarke, Covington,
772
+ Forrest, Harrison, Jackson, Jones, Lowndes, Oktibbeha, Tishomingo and Wilkinson
773
+ on a 25-row sample; Adams and Union on a 50-row resample from a second FID
774
+ window after the first sample landed at 22 of 25). The other 68 carry a
775
+ reviewed `owner_mailing_address: not_published` record that says why.
776
+ - `license: "unknown"` with the MARIS cadastral data disclaimer as `licenseUrl`.
777
+ The data is public, free and keyless, and no explicit license is stated. The
778
+ `notes` credit the county tax assessors and MARIS/MDEQ; `dataVintage` names
779
+ the 2024 release (TAXYEAR reads 2024 on every row).
780
+
781
+ ### Changed
782
+
783
+ - **Hinds MS** keeps its own-host primary (it paginates) and gains the
784
+ statewide West layer as a second endpoint; both read PARNO. **DeSoto MS**
785
+ keeps its single own-host endpoint: its primary reads PIN and the statewide
786
+ layer PARNO, and a county whose endpoints disagree on the parcel-id column
787
+ could hold no protected-parcel notice entry. Both primaries' `corsEnabled`
788
+ resolves from `null` to `true`, measured with an Origin header. Hinds gains a
789
+ reviewed `owner_mailing_address: not_published` record: the mailing columns
790
+ are empty on both of its layers.
791
+
792
+ ## 0.6.33 — 2026-10-14
793
+
794
+ City data. No API change, no breaking change, `totals` unchanged. The city
795
+ registries grow from 20 to **30 cities** with code-violation or building-permit
796
+ datasets and from 17 to **19** with zoning-case dockets. Every endpoint, address
797
+ column, date column and scope filter was re-read live on 2026-10-14, and each
798
+ new dataset carries `latestRecordDate`, the newest record seen that day.
799
+
800
+ ### New code-violation and building-permit cities
801
+
802
+ - **Los Angeles CA**: LADBS open code-enforcement cases (`u82d-eh7z`; closed
803
+ cases sit in a separate dataset that is not searched) and building permits
804
+ issued since 2020 (`pi9x-tg5x`). The violations table splits the street, so
805
+ a typed "number name" is matched on `stno` and `stname`.
806
+ - **Fort Worth TX**: code violations and development permits (ArcGIS tables).
807
+ The permits table's one-line address column is empty upstream, so permits
808
+ match the street name on `Street_Name` and the house number on `Addr_No`.
809
+ Trade permits are included.
810
+ - **Mesa AZ**: code-compliance cases and issued building permits (Socrata).
811
+ - **Minneapolis MN**: code violations and permits (ArcGIS).
812
+ - **Wichita KS**: MABCD code cases and permits (the city's own ArcGIS server).
813
+ The permits layer is scoped to `Jurisdiction='Wichita'`. The county name is
814
+ Sedgwick County, which the atlas does not carry, so no `countySlug`.
815
+ - **Cleveland OH**: Building and Housing violation notices and building permits
816
+ (ArcGIS), each with its Accela Citizen Access link.
817
+ - **New Orleans LA**: code-enforcement cases and permits (Socrata, CC0).
818
+ - **Orlando FL**: code-enforcement cases scoped to the housing, lot, zoning,
819
+ abatement and commercial case types, and permit applications (Socrata).
820
+ - **Cincinnati OH**: code-enforcement complaints scoped to
821
+ `display_agency like 'Cincinnati%'` (the table is Hamilton County-wide) and
822
+ building permits (Socrata).
823
+ - **Scottsdale AZ**: closed code complaints since 2021-09 (split street, matched
824
+ on `StreetNumber` and `StreetName`) and building permits (the city's own
825
+ ArcGIS server).
826
+
827
+ ### New zoning-case cities
828
+
829
+ - **Orlando FL**: Planning Applications scoped to the Zoning application type.
830
+ Existing and proposed zoning live in the comments prose, not in columns.
831
+ - **Scottsdale AZ**: the ZN (rezoning) series of the planning-case table.
832
+ Thin: about 6 rezonings since 2025-04.
833
+
834
+ ### County links
835
+
836
+ `relatedPropertyCitySlug` now names a city for Los Angeles CA, Tarrant TX,
837
+ Hennepin MN, Cuyahoga OH, Hamilton OH, Orleans LA and Orange FL counties;
838
+ `relatedZoningCitySlug` for Los Angeles CA, Tarrant TX, Orleans LA, Orange FL
839
+ and Maricopa AZ (Scottsdale; Maricopa's property link stays Phoenix).
840
+
841
+ ### Disclosures
842
+
843
+ - The hosted MCP `get_city_datasets` now returns `zoning_source` whenever the
844
+ zoning registry carries a `currentRecordsNote`, not only when its source URL
845
+ differs from the property source, so Orlando's and Scottsdale's zoning notes
846
+ reach agents on the default call. Additive: no field renamed.
847
+ - Dataset `coverage` now states the measured search limits: Orlando's
848
+ double-spaced addresses (search by street name alone), the split-street
849
+ datasets (LA violations, Fort Worth permits, Scottsdale violations: a street
850
+ typed without a house number is matched as typed, so a suffix finds nothing),
851
+ Wichita's null-jurisdiction permits, Cincinnati's departments with their
852
+ shares, and Scottsdale's sparse Zoning column.
853
+ - Types: `CitySource` (the zoning record's `source`) gains the optional
854
+ `currentRecordsNote` its data already carried. Additive.
855
+
856
+ ## 0.6.32 — 2026-10-14
857
+
858
+ New counties — Berks PA (42011), Bedford PA (42009), Salem city VA (51775),
859
+ Nueces TX (48355), Hidalgo TX (48215), Brazos TX (48041), Ector TX (48135) and
860
+ Webb TX (48479, a researched stub with no public REST). No API change, no
861
+ breaking change. `totals` move from 306 counties / 324 endpoints (298 with a
862
+ verified endpoint) to **314 counties / 331 endpoints** (305). Callers asked for
863
+ all eight.
864
+
865
+ ### New counties
866
+
867
+ - **Berks PA (42011)**, with a verified endpoint on PASDA, the Penn State
868
+ clearinghouse that hosts the county's data (`BerksCounty` MapServer layer 6,
869
+ 156,838 parcels): owner, situs, owner mailing address and the 14-digit parcel
870
+ id are searchable, and the layer answers cross-origin requests. The layer is a
871
+ January 2026 snapshot, and PASDA republishes under a new vintage stamp, so the
872
+ layer name or index may move; the `notes` say how to find it again.
873
+ - **Bedford PA (42009)**, with a verified endpoint on PASDA (`BedfordCounty`
874
+ MapServer layer 0, 34,225 parcels). The layer is the July 2023 vintage and
875
+ situs is filled on only about 26 percent of rows, so an address search misses
876
+ most parcels; owner, parcel id and a point lookup are the dependable paths. The
877
+ `notes` record both.
878
+ - **Salem city VA (51775)**, an independent city, with a verified endpoint on the
879
+ city's own ArcGIS Online (`TaxParcels` FeatureServer layer 5, 10,765 parcels,
880
+ edited the day before verification). The mailing street is `owner_address2`;
881
+ `owner_address1` is a care-of line on about 8 percent of rows.
882
+ - **Nueces TX (48355)**, with a verified endpoint on the ArcGIS Online
883
+ organisation of the appraisal district's GIS vendor (`NuecesCADWebService`
884
+ FeatureServer layer 0, 157,404 parcels, edited on the day of verification).
885
+ The site address is split across `situs_num` and `situs_street`, so an address
886
+ search runs on the street word; the `notes` say so.
887
+ - **Hidalgo TX (48215)**, with a verified endpoint on the regional 911 district's
888
+ server (`RGV911_Web_Apps_MIL1` MapServer layer 7, 334,098 parcels). Situs is
889
+ one full string; the mailing column is the street line only, with no mailing
890
+ city, state or ZIP.
891
+ - **Brazos TX (48041)**, with a verified endpoint on the City of Bryan's server
892
+ (`BRAZOS_COUNTY_CERTIFIED_PARCELS` MapServer layer 0, the 2025 certified roll,
893
+ 75,652 parcels, whole county). Empty source fields were written as the text
894
+ `NULL`, and most `SitusAddr` values end in ` NULL`; the data is shipped as
895
+ published and the `notes` say how to query around it.
896
+ - **Ector TX (48135)**, with a verified endpoint on the appraisal-software
897
+ vendor's server (`EctorCounty_Basemap` MapServer layer 0, 77,042 parcels).
898
+ The layer publishes NO situs column: owner, a one-line mailing address and the
899
+ PIN are searchable, and an address lookup has to go through a geocoded point.
900
+ - **Webb TX (48479)**, no public endpoint: `hasPublicRest: false` and an empty
901
+ `endpoints` array. Webb CAD's map host answers with a login page, and the only
902
+ public Webb layers on ArcGIS Online are partial. The `notes` record the
903
+ evidence as of 2026-10-14.
904
+
905
+ ## 0.6.31 — 2026-10-12
906
+
907
+ City datasets. New export, no breaking change, `totals` unchanged. The package
908
+ now bundles the two city registries urbankitstudio.com serves at
909
+ `/data/city-property/` and `/data/city-zoning/`: 20 cities with verified
910
+ code-violation or building-permit datasets and 17 with zoning-case dockets
911
+ (`data/city-property/`, `data/city-zoning/`), and describes how to query each
912
+ one the way the site does.
913
+
914
+ - `listCities(filter?)`, `findCity(ref, state?)` (slug, name, "Name, ST", the
915
+ NYC and DC aliases; ambiguity is a list), `getCityProperty(slug)`,
916
+ `getCityZoning(slug)`, `cityDatasets(slug, kind?)`, `citiesForCounty(stateSlug,
917
+ countySlug)`, `cityPath(slug, kind)`.
918
+ - `buildCityQuery(dataset, street)` returns the exact URL the site sends for a
919
+ dataset and a typed street (Socrata, ArcGIS, CKAN or CARTO), with the site's
920
+ quote-doubling guard; null when the lookup needs a geocoded point or a parcel
921
+ id first, in which case `dataset.queryTemplate` carries the request with a
922
+ `{LON},{LAT}` placeholder. Address lookups carry no template on purpose: a
923
+ street must be escaped for the provider's SQL, and only `buildCityQuery` does
924
+ that. LIKE wildcards (`%`, `_`) in a street stay active, as on the site; each
925
+ `queryNote` says so.
926
+ - Each `CityDatasetDescriptor` carries the provider, endpoint, lookup mode,
927
+ address and date columns, the scope `where` that is part of the dataset's
928
+ definition, the field map, status buckets, coverage, a freshness block, the
929
+ point template where one applies, a newest-first `sampleQuery` and a `queryNote`.
930
+ - County records' `relatedZoningCitySlug` and `relatedPropertyCitySlug` now
931
+ resolve inside the package through `getCityZoning` / `getCityProperty`.
932
+ - Split addresses. A dataset whose table keeps the house number in its own
933
+ column (`addressParts`: Dallas violations, Boston, Chicago, NYC and San
934
+ Francisco, plus the Boston and Indianapolis dockets) is matched on the street
935
+ NAME: `buildCityQuery` takes the typed house number off, then a leading
936
+ directional when the table has a direction column, then one trailing street
937
+ type when it has a type column (`streetForDataset`, `splitAddressStreetName`;
938
+ the roles follow the column order of `addressParts`). A name column holds
939
+ "MARILLA" while "1500" and "ST" sit elsewhere, so the typed line as one
940
+ substring finds almost nothing (Dallas, 2026-10-13: 101 of 109 Marilla rows
941
+ carry the type in `str_suffix`). The page's own URL stays `propertyQueryUrl`
942
+ / `casesQueryUrl`; a `streetNumber` mapping, which no shipped city sets yet,
943
+ matches number and name on both columns instead.
944
+ - Nothing fetches. For volume against a Socrata dataset, send your own
945
+ `X-App-Token`.
946
+ - Boston's zoning record no longer names `suffolk-county`, which the atlas
947
+ does not carry (Massachusetts is statewide); its county name stays.
948
+
949
+ ## 0.6.30 — 2026-10-12
950
+
951
+ 0.6.30: endpoint capability keys supportsPagination and maxRecordCount,
952
+ offersPagination() reader; no data change. Builds on 0.6.29. `totals` do not move, and no
953
+ bundled endpoint carries either key yet. No breaking change.
954
+
955
+ Some ArcGIS layers refuse pagination: any `resultRecordCount` or
956
+ `resultOffset`, in any letter case, gets HTTP 200 with
957
+ `{"error":{"code":400,"message":"Pagination is not supported."}}` instead of
958
+ rows. The atlas will register such layers (Mississippi's statewide parcel
959
+ layers are next), so the record can now say so before any of them ship.
960
+
961
+ ### Added
962
+
963
+ - `EndpointRecord.supportsPagination?: false`. Only the literal `false` is
964
+ written; absent means the layer pages. The name is Esri's own
965
+ (`advancedQueryCapabilities.supportsPagination`).
966
+ - `EndpointRecord.maxRecordCount?: number`, the layer's declared per-query row
967
+ cap, present on every flagged endpoint.
968
+ - `offersPagination(endpoint)`: `false` on a flagged layer. Check it before
969
+ adding either parameter; when it is `false`, send one request without them
970
+ and narrow the query when `exceededTransferLimit` comes back.
971
+
972
+ The registry's validator enforces, on every flagged endpoint: a positive
973
+ integer `maxRecordCount`, a `sampleQuery` with neither parameter, and the same
974
+ two values on every registration of the same layer URL.
975
+
976
+ ## 0.6.29 — 2026-10-12
977
+
978
+ New counties, no API change, no breaking change. Builds on 0.6.28. `totals`
979
+ move from 303 counties / 322 endpoints to **306 counties / 324 endpoints**
980
+ (298 with a verified endpoint). All three were asked for through the site's free
981
+ tools. Jackson MS (28059), also asked for, is not in this release: the only
982
+ public layer for it has no pagination support, and the service's own probes
983
+ send a row cap, so it waits for a release that handles that.
984
+
985
+ ### New counties
986
+
987
+ - **Racine WI (55101)**, with a verified endpoint on the county's own ArcGIS
988
+ Enterprise server (`DataPortal/Parcels_NAD83_2011_STATEPLANE` FeatureServer
989
+ layer 0, 84,684 parcels, updated nightly): owner, situs, owner mailing address
990
+ and parcel id are searchable, pagination works, and the layer answers
991
+ cross-origin requests.
992
+ - **Westmoreland PA (42129)**, with a verified endpoint on the county's own
993
+ ArcGIS Enterprise server (`Upgrade/Parcels_AS400_VIEW_Upgrade` MapServer layer
994
+ 2, 191,739 parcels). The owner column `NAME` is a fixed-width 30-character
995
+ string padded with trailing spaces, so an exact `=` never matches: search with
996
+ `LIKE 'SMITH%'`, and the server rejects `TRIM()`. The service lives in a folder
997
+ named `Upgrade`, so the URL may move; the `notes` say so.
998
+ - **Casey KY (21045)**, no public endpoint: `hasPublicRest: false` and an empty
999
+ `endpoints` array. The Casey PVA publishes only through qPublic, the state
1000
+ server carries no Casey parcel service, and Kentucky has no statewide parcel
1001
+ layer yet. The `notes` record the evidence as of 2026-10-12 and the PVA office
1002
+ to ask.
1003
+
1004
+ ## 0.6.28 — 2026-10-10
1005
+
1006
+ withholding: RESTRICTED markers and masked mailing. Data notes only, no API
1007
+ change, no breaking change, no change to `totals`. Four county notes now say
1008
+ how the service treats those values:
1009
+
1010
+ - New Hanover NC: an owner that reads "RESTRICTED" is a county redaction
1011
+ marker, so that row is withheld.
1012
+ - Fulton GA: 41 rows whose owner is the county's RESTRICTED marker are withheld.
1013
+ - Lake IL: 24 rows whose owner (taxpayer_name) is the county's RESTRICTED marker are
1014
+ withheld.
1015
+ - Wood WV: 263 mailing values carry asterisks. 251 of them are fully masked
1016
+ (some keep the ZIP, some carry none) and are served as no mailing address;
1017
+ the other 12 are real addresses and are served.
1018
+
1019
+ On the service side (not in this package), "NOT RESTRICTED" and "NOT
1020
+ CONFIDENTIAL" are not markers, and a mailing value counts as fully masked when
1021
+ it is only asterisks and separators, at least one asterisk, with at most a
1022
+ trailing ZIP or ZIP+4, or a run of three or more "#".
1023
+
1024
+ ## 0.6.27 — 2026-10-10
1025
+
1026
+ New counties and a New York outage fix, no API change, no breaking change. `totals` move from 294
1027
+ counties / 305 endpoints to **303 counties / 322 endpoints** (296 with a
1028
+ verified endpoint). Each new county sits on a statewide layer the atlas already
1029
+ carries, one registration per county scoped by that layer's county field:
1030
+ the NC OneMap parcel aggregation (`cntyname`, as Forsyth and Montgomery use),
1031
+ the Montana cadastral layer (`CountyName`, as Yellowstone uses) and the WVU
1032
+ parcel layer (`COUNTY`, the state's alphabetical code, as Kanawha uses).
1033
+
1034
+ ### New counties
1035
+
1036
+ - **New Hanover NC (37129)**, 103,489 parcels, owner on 99.8%.
1037
+ - **Iredell NC (37097)**, 109,254 parcels. The site city is blank on every row.
1038
+ - **Union NC (37179)**, 117,539 parcels. The site city is blank on every row.
1039
+ - **Onslow NC (37133)**, 95,297 parcels.
1040
+ - **Johnston NC (37101)**, 125,583 parcels. 68 owner rows carry an asterisk
1041
+ on a real name.
1042
+ - **Brunswick NC (37019)**, 151,753 parcels, situs on 85%. The site city is
1043
+ blank on every row.
1044
+ - **Gallatin MT (30031)**, 53,618 parcels, TaxYear 2026, situs on 83.7%.
1045
+ - **Berkeley WV (54003, code `02`)**, 62,652 parcels, situs on 98.3%.
1046
+ - **Wood WV (54107, code `54`)**, 58,263 parcels. 263 mailing addresses are
1047
+ masked with asterisks.
1048
+
1049
+ NC rows carry no reliable per-county date (`revisedate` is null and the layer
1050
+ publishes no editing info). West Virginia's layer says it was updated for Tax
1051
+ Year 2023. North Carolina goes from 9 to 15 counties, Montana from 4 to 5 and
1052
+ West Virginia from 4 to 6.
1053
+
1054
+ ### New York outage fix
1055
+
1056
+ The state parcel server (`gisservices.its.ny.gov`) has hung on every query
1057
+ since about 2026-10-09 16:00-18:00Z. Eight New York records gain the official
1058
+ NYSGIS mirror of the same layer on ArcGIS Online (`services6.arcgis.com`,
1059
+ FeatureServer/1, 2025 roll, last edited 2026-09-25, open CORS). Onondaga,
1060
+ Suffolk and Westchester list the mirror first and keep the state endpoint
1061
+ second; the five NYC boroughs list it third, behind MapPLUTO and the state
1062
+ endpoint. Counties do not change, so only the endpoint total moves (+8).
1063
+ The mirror carries no mailing address for the boroughs. A real owner name and
1064
+ street address in the Onondaga note was replaced with a generic description.
1065
+
1066
+ ## 0.6.26 — 2026-10-10
1067
+
1068
+ New counties, no API change, no breaking change. `totals` move from 286
1069
+ counties / 297 endpoints to **294 counties / 305 endpoints** (287 with a
1070
+ verified endpoint). All eight sit on the Colorado OIT statewide parcel
1071
+ composite already used by El Paso and Jefferson, one registration per county
1072
+ scoped by `countyName`. The layer is a state roll-up: each county's rows carry
1073
+ the `dateReceived` the state loaded them (all 2026-03-27 to 2026-05-19), so
1074
+ ownership can trail the county's own roll.
1075
+
1076
+ ### New counties
1077
+
1078
+ - **Arapahoe CO (08005)**, 231,438 parcels, situs on 93%, mailing on 98%.
1079
+ - **Adams CO (08001)**, 188,474 parcels, situs on 93%.
1080
+ - **Larimer CO (08069)**, 180,960 parcels, situs on 88%.
1081
+ - **Douglas CO (08035)**, 164,818 parcels, situs on 88%.
1082
+ - **Weld CO (08123)**, 163,491 parcels, situs on 81%.
1083
+ - **Boulder CO (08013)**, 144,551 parcels, situs on 99%.
1084
+ - **Mesa CO (08077)**, 78,984 parcels. 650 carry the placeholder owner
1085
+ "Confidential Owner"; those rows are withheld, not served as a name.
1086
+ - **Park CO (08093)**, 43,241 parcels, situs on 85%.
1087
+
1088
+ Pueblo (101,146 parcels) was checked and left out: the layer carries no situs
1089
+ address for it. Colorado goes from 3 to 11 counties.
1090
+
1091
+ ## 0.6.25 — 2026-10-10
1092
+
1093
+ New counties, no API change, no breaking change. `totals` move from 280
1094
+ counties / 291 endpoints to **286 counties / 297 endpoints** (279 with a
1095
+ verified endpoint). All six sit on the Arkansas GIS Office statewide parcel
1096
+ layer, one registration per county scoped by `countyfips`, and none of them
1097
+ publishes a mailing address (the layer has no mailing field).
1098
+
1099
+ ### New counties
1100
+
1101
+ - **Clay AR (05021)**, requested by a visitor through the county-request form.
1102
+ 16,324 parcels, owner populated on every row, situs on 69%.
1103
+ - **Saline AR (05125)**, 80,143 parcels, situs on 96%.
1104
+ - **Garland AR (05051)**, 76,870 parcels, situs on 99%.
1105
+ - **Faulkner AR (05045)**, 66,634 parcels, situs on 88%.
1106
+ - **Sebastian AR (05131)**, 60,283 parcels, situs on 91%.
1107
+ - **Craighead AR (05031)**, 54,896 parcels, situs on 87%.
1108
+
1109
+ Arkansas goes from 3 to 9 counties.
1110
+
1111
+ ## 0.6.24 — 2026-10-10
1112
+
1113
+ New counties, no API change, no breaking change. `totals` move from 277
1114
+ counties / 288 endpoints to **280 counties / 291 endpoints** (273 with a
1115
+ verified endpoint). Each was chosen for population and for publishing owner,
1116
+ mailing address, situs and parcel id on a public layer.
1117
+
1118
+ ### New counties
1119
+
1120
+ - **Chester PA (42029)**, with a verified endpoint on PASDA, Penn State's state
1121
+ clearinghouse (`pasda/ChesterCounty` MapServer layer 11, "Chester County
1122
+ Parcels 202610", 194,767 parcels, county-supplied). Owner, situs, owner
1123
+ mailing address and UPI are searchable, and the layer answers cross-origin
1124
+ requests. PASDA republishes monthly under a new vintage stamp, so the layer
1125
+ name (and possibly its index) rotates; the `notes` say how to re-resolve it,
1126
+ and that the older `maps.pasda.psu.edu/arcgis` copy is dead.
1127
+ - **Gaston NC (37071)**, with a verified endpoint on the county's own
1128
+ `PublicGIS/Parcels` MapServer layer 11 (118,132 parcels). The layer carries
1129
+ three owner vintages; the record lists only the current-owner `CURR_`
1130
+ columns, not the January-1 (`JAN1_`) or prior-year (`PRVYR`) ones. No CORS
1131
+ header, so it is server-side only.
1132
+ - **Clark WA (53011)**, with a verified endpoint on the county's own
1133
+ `MapsOnline/MailingLabels` FeatureServer layer 0 ("Owner Addresses", 196,284
1134
+ taxlots): owner, care-of, a complete owner mailing address, situs and the
1135
+ Assessor's property id, plus a parcel publish date (2026-10-09). The sibling
1136
+ `PropertyFinder` layer is not used because it has no date and no mailing
1137
+ city, state or ZIP. No CORS header, so it is server-side only.
1138
+
1139
+ ## 0.6.23 — 2026-10-10
1140
+
1141
+ New counties, no API change, no breaking change. Builds on 0.6.22. `totals`
1142
+ move from 273 counties / 284 endpoints to **277 counties / 288 endpoints**
1143
+ (270 with a verified endpoint). All four were asked for through the site's free
1144
+ tools.
1145
+
1146
+ ### New counties
1147
+
1148
+ - **James City VA (51095)**, with a verified endpoint on the county's own
1149
+ ArcGIS Server (`JCC/GIS_Data` FeatureServer layer 17, 36,478 parcels):
1150
+ owner, situs, owner mailing address and PIN are searchable, and the layer
1151
+ answers cross-origin requests. The county's licence text asks that a
1152
+ disclaimer be placed on all products derived from the data; the `notes`
1153
+ quote it. The county's ArcGIS Online copy of the layer requires a token and
1154
+ is not used.
1155
+ - **Hill TX (48217)**, with a verified endpoint on Hill CAD's
1156
+ `HillCADWebService` FeatureServer layer 0 (39,102 parcels), hosted on the
1157
+ appraisal district's GIS vendor's ArcGIS Online organisation. The situs is
1158
+ split over `situs_num` and `situs_street` (plus prefix and suffix); the
1159
+ record lists `situs_street` as the site-address column, so search an address
1160
+ by street name plus `situs_num`. 213 rows read `CONFIDENTIAL` as the owner
1161
+ (Texas Tax Code §25.025 withholding); the hosted API and MCP serve those rows
1162
+ with `withheld: "county-redacted"`, as described under 0.6.22. The
1163
+ vendor-hosted URL can move; the `notes` say so.
1164
+ - **Greene MO (29077)**, with a verified endpoint on the City of Springfield's
1165
+ `Maps/GisViewerGroups` MapServer layer 18 (123,884 parcels), which covers the
1166
+ whole county, not only the city. The owner mailing address is one combined
1167
+ string. The layer publishes no vintage and no terms.
1168
+ - **Georgetown SC (45043)**, parcel shapes only: the county's
1169
+ `GCGIS_OpenData` FeatureServer layer 2 (47,943 parcels) carries the TMS,
1170
+ plat and zoning and no owner, situs or mailing column, so the record carries
1171
+ an `ownerFieldNote` and `not_published` overrides for those three fields.
1172
+ The county publishes owners in a separate non-spatial table on the same
1173
+ service (layer 7, `PARCELATTRIBUTES`, keyed by TMS); the atlas registers one
1174
+ spatial layer per endpoint and does not join tables, so the `notes` describe
1175
+ the table and the two-query join rather than registering it.
1176
+
1177
+ ## 0.6.22 — 2026-10-10
1178
+
1179
+ New counties, no API change, no breaking change. Builds on 0.6.21. `totals`
1180
+ move from 257 counties / 268 endpoints to **273 counties / 284 endpoints**
1181
+ (266 with a verified endpoint). Kentucky goes from one county to 17: 13 publish
1182
+ owner names (Fayette plus 12 new) and 4 are parcel shapes only.
1183
+
1184
+ ### New counties — Kentucky, with owner names (12)
1185
+
1186
+ Boone (21015), Warren (21227), Daviess (21059), McCracken (21145), Clark
1187
+ (21049), Shelby (21211), Webster (21233), Franklin (21073), Simpson (21213),
1188
+ Henry (21103), Montgomery (21173) and Meade (21163). Each record carries the
1189
+ parcel id, the owner name, the owner mailing address, the situs address (Henry
1190
+ has none) and the county's other columns as documented `searchFields`, and a
1191
+ sample query on the owner column. Every layer answers cross-origin requests.
1192
+
1193
+ - **Withheld-address note.** Kentucky law lets judges (KRS 61.7991) and people
1194
+ in the Safe at Home program (KRS 14.304) have their home address withheld, so
1195
+ a county may blank an owner or label it `CONFIDENTIAL`. Each of the 12 records
1196
+ carries `capabilityOverrides.owner_name` and `owner_mailing_address` as
1197
+ `available` with a `statute` basis whose `note` says so, cites both statutes
1198
+ and names the address where a Safe at Home participant can send written
1199
+ notice, and says that judges and participants who send written notice have
1200
+ their parcel withheld from every UrbanKit surface. (The hosted API and MCP
1201
+ now serve a row with a county redaction marker in its owner, mailing or situs
1202
+ value as owner, mailing and situs null with `withheld: "county-redacted"`,
1203
+ keeping only the request fields and the parcel id; the package ships only the
1204
+ registry. A row withheld on written notice, `withheld: "notice-suppressed"`,
1205
+ is not charged; on `radius_owners` / `POST /api/atlas/radius`, which bill per
1206
+ row that carries an owner or mailing address, a county-redacted row is not
1207
+ charged either.) Read it with `reviewedCapability(county, "owner_name")`. These
1208
+ counties are `available`, not `restricted`.
1209
+ - **Franklin** publishes a 2019 snapshot (its newest `Year` is 2019); its
1210
+ record carries a `dataVintage` advisory that says so.
1211
+ - **Simpson**'s layer states "Data is not to be used for legal purposes"; the
1212
+ `notes` quote it. Its mailing street is `Address__1` (`Address_Li` is a
1213
+ second line used on 66 rows).
1214
+ - **Meade**'s service name ends in `_WFL1`, and that suffix is required.
1215
+ - In the Shelby, Franklin, McCracken, Webster, Henry, Montgomery, Meade and
1216
+ Simpson layers the plain `City`, `State` and `Zip` columns are the OWNER'S
1217
+ mailing city, state and ZIP, not the situs; the labels say so.
1218
+
1219
+ ### New counties — Kentucky, parcel shapes only (4)
1220
+
1221
+ Jefferson (21111, LOJIC's open-data layer), Madison (21151), Oldham (21185)
1222
+ and Henderson (21101). Their layers carry a parcel id and geometry (Henderson
1223
+ also a situs address) and no owner column, so each records `owner_name` as
1224
+ `not_published` on an `endpoint_schema` basis. Jefferson's record uses LOJIC's
1225
+ current layer rather than Louisville Metro's `New_AllParcels`, which has an
1226
+ address and a property class but was last edited in 2017.
1227
+
1228
+ ## 0.6.21 — 2026-10-08
1229
+
1230
+ Three demanded counties, no API change, no breaking change. `totals` move to
1231
+ **257 counties / 268 endpoints** (250 with a verified endpoint).
1232
+
1233
+ ### New counties
1234
+
1235
+ - **Lyon NV (32019)**, with a verified endpoint on the county's own ArcGIS
1236
+ Server (`Landbase/Land_Use_Parcel`, MapServer layer 10 `Land Use -- Parcel`,
1237
+ 35,343 parcels, owner on 99.5% of them, CORS echoes the caller's origin).
1238
+ Owner (`PANAME`), situs (`PHY_ADDR`), mailing address (`PMADD1`/`PMADD2`/
1239
+ `PMCTST`/`PZIP`) and the dashed `APN` are all on the layer. The owner column
1240
+ has no index: a `%SMITH%` search takes 13 to 15 seconds and an anchored
1241
+ `SMITH%` search about 4.5, which the `notes` record.
1242
+ - **Humboldt CA (06023)**, with a verified endpoint on the county's own ArcGIS
1243
+ Server (`Parcels/Parcels`, MapServer layer 0, 72,721 parcels). The layer is
1244
+ parcels only: it publishes the APN, situs address, zoning, general plan and
1245
+ use code, but no owner or mailing columns, and the `notes` say so.
1246
+ - **Fresno CA (06019)**, parcels only, with a verified endpoint on the county's
1247
+ own ArcGIS Server (`FC_PARCEL_SELECT`, MapServer layer 0, 442,045 rows). The
1248
+ county states it cannot provide ownership information on the internet under
1249
+ Cal. Gov. Code 6254.21 (now 7928.205), so the record carries an
1250
+ `ownerFieldNote` and `capabilityOverrides` marking `owner_name` and `owner_mailing_address`
1251
+ `restricted` (`county_cited_statute`, attributed to the County of Fresno), and
1252
+ neither column is in `searchFields`. `reviewedCapability` and
1253
+ `isReviewedUnservable` report both restricted. The APN, situs address, land
1254
+ use and assessed values are searchable.
1255
+
1256
+ ## 0.6.20 — 2026-10-08
1257
+
1258
+ New counties, no API change, no breaking change. `totals` move to **254
1259
+ counties / 265 endpoints** (247 with a verified endpoint).
1260
+
1261
+ ### New counties
1262
+
1263
+ - **Chesterfield VA (51041)**, with a verified endpoint on the county's own
1264
+ ArcGIS Online organisation (`Cadastral_ProdA`, layer 3 `ParcelsEnriched`,
1265
+ 150,624 parcels, edited daily, CORS `*`). The county's older Hub `Parcels`
1266
+ layer is deprecated and frozen at 2024-07-30; the record does not use it.
1267
+ - **Newport News VA (51700)**, an independent city, with a verified endpoint on
1268
+ the city's `Operational/Parcel` MapServer (54,444 parcels). The layer
1269
+ publishes no edit date, which the `notes` say.
1270
+ - **Bradford PA (42015)**, with a verified endpoint on PASDA's
1271
+ `BradfordCounty` MapServer layer 5 (34,935 parcels, county-supplied data).
1272
+ The owner field is truncated at 25 characters, the address columns are the
1273
+ owner's mailing address, and the layer carries no situs address; the `notes`
1274
+ say so.
1275
+ - **Montgomery NC (37123)**, with a verified endpoint on the NC OneMap
1276
+ statewide parcel layer, scoped by `scopeWhere` to
1277
+ `cntyname = 'Montgomery'` (30,283 parcels).
1278
+ - **Dutchess NY (36027)**, registered with `hasPublicRest: false`: the county's
1279
+ parcel layer carries geometry and ids only, the New York State tax parcel
1280
+ service has no Dutchess rows, and owner data is sold by the county's Real
1281
+ Property Tax Service Agency. The `notes` record the evidence and the contact.
1282
+
1283
+ ## 0.6.19 — 2026-10-06
1284
+
1285
+ New county, no API change, no breaking change. `totals` move to **249
1286
+ counties / 261 endpoints** (243 with a verified endpoint).
1287
+
1288
+ ### New county — Polk OR (41053)
1289
+
1290
+ Asked for through the site's free tools on 2026-10-04. The county's
1291
+ Hub-advertised REST path (`maps.co.polk.or.us/gis/`) has returned HTTP 500 on
1292
+ every request since at least 2026-10-05, but the same ArcGIS Server answers on
1293
+ its second web adaptor, `/arcserv2/`, and the record registers its
1294
+ `Klop_core/Acct_35` taxlot layer: 36,413 taxlots with the assessment account
1295
+ joined on, so owner name, owner mailing address, situs, Map Taxlot, ORTaxlot and
1296
+ account number are all searchable, and the layer answers cross-origin
1297
+ requests (`corsEnabled: true`), so browser clients query it directly. The
1298
+ `notes` record the outage and the adaptor it works around.
1299
+
1300
+ ## 0.6.18 — 2026-10-06
1301
+
1302
+ New county, no API change, no breaking change. `totals` move to **248
1303
+ counties / 260 endpoints** (242 with a verified endpoint).
1304
+
1305
+ ### New county — Mendocino CA (06045)
1306
+
1307
+ The County's own ArcGIS Online organisation (Mendocino County GIS) publishes a
1308
+ county-wide assessor parcel layer, `Parcels_Public_` FeatureServer layer 0:
1309
+ 62,051 parcels, data last edited 2026-06-08, CORS open. APN (8-character text),
1310
+ the 10-character APNFULL and the situs address are searchable. The layer
1311
+ publishes no owner name and no owner mailing address, so the record carries an
1312
+ `ownerFieldNote` and `not_published` capability overrides for both, and an
1313
+ owner search returns nothing for this county. The City of Ukiah's organisation
1314
+ hosts two Ukiah-area clips of the same data; neither is county-wide, so neither
1315
+ is listed. `addedAt` is 2026-10-06.
1316
+
1317
+ ## 0.6.17 — 2026-10-05
1318
+
1319
+ New county, one new optional field, no breaking change. `totals` move to **247
1320
+ counties / 259 endpoints** (241 with a verified endpoint).
1321
+
1322
+ ### New county — Schenectady NY (36093)
1323
+
1324
+ Asked for through the site's free tools on 2026-10-04 and not in the NYS Tax
1325
+ Parcels Public statewide layer (a count on `COUNTY_NAME='Schenectady'` returns
1326
+ 0 where the Albany control returns 112,804). The record registers the public
1327
+ hosted layer published by Spatial Data Logic, the vendor behind the county's
1328
+ property system: 59,235 parcels, owner names published, parcel id (tax-map
1329
+ print key), situs and owner mailing address. It carries a `dataVintage`
1330
+ advisory, because the layer's data was last edited on 2023-09-28.
1331
+
1332
+ ### New field — `addedAt`
1333
+
1334
+ A county may now carry `addedAt` (`YYYY-MM-DD`), the date it entered the
1335
+ atlas. It is optional (counties added before this release carry none) and,
1336
+ unlike `lastVerified`, it never moves. Schenectady is the first record to carry
1337
+ it.
1338
+
1339
+ ### Re-checked, still no public REST — Iron MI (26071), Garfield OK (40047)
1340
+
1341
+ Both were asked for and both stay `hasPublicRest: false`; their `notes` now
1342
+ record the 2026-10-05 re-check (Michigan's state ArcGIS server lists no parcel
1343
+ layer; Garfield's Spatialest GeoServer still has WFS disabled, and Oklahoma has
1344
+ no statewide parcel service).
1345
+
1346
+ ## 0.6.16 — 2026-10-04
1347
+
1348
+ Data refresh. No API change. `totals` unchanged at **246 counties / 258
1349
+ endpoints**. Five New Jersey county records on the two NJOGIS statewide
1350
+ services now state what those services serve. All 67 Florida registrations
1351
+ on the FDOR statewide cadastral replace a `sampleQuery` the layer rejects
1352
+ with one scoped to the county. The site's downloadable directory now scopes its sample queries,
1353
+ which is not part of this package.
1354
+
1355
+ ### Data refresh — Bergen NJ lists the mailing-address columns its shared layer serves
1356
+
1357
+ - **New Jersey:** Bergen County (FIPS 34003) sits on the NJ Office of GIS
1358
+ statewide composite, `Parcels_Composite_NJ_WM` FeatureServer layer 0, scoped
1359
+ `COUNTY='BERGEN'`. Essex County uses the same layer, and Essex's record
1360
+ already listed the mailing address on the tax record. Bergen's did not, so
1361
+ it understated what the layer returns. Measured live 2026-10-04, with
1362
+ Essex as the control:
1363
+
1364
+ | `COUNTY=` scope, test | BERGEN | ESSEX |
1365
+ |---|---|---|
1366
+ | all rows | 281,646 | 178,089 |
1367
+ | `ST_ADDRESS<>''` | 281,646 | 178,089 |
1368
+ | `CITY_STATE<>''` | 281,646 | 178,089 |
1369
+ | `ZIP_CODE<>''` | 281,612 | 178,088 |
1370
+ | `OWNER_NAME<>''` | 0 | 0 |
1371
+ | `PROP_LOC<>''` | 281,646 | 178,089 |
1372
+ | `CALC_ACRE>0` | 244,163 | 152,381 |
1373
+ | `YR_CONSTR>0` | 263,852 | 156,919 |
1374
+ | `BLDG_DESC<>''` | 241,129 | 121,070 |
1375
+
1376
+ `OWNER_NAME IS NOT NULL` returns every row in both counties, because the
1377
+ column holds empty strings. Test it with `<>''`.
1378
+ - **`searchFields`:** adds `ST_ADDRESS` ("Mailing Street Address",
1379
+ searchable), `CITY_STATE` ("Mailing City and State") and `ZIP_CODE`
1380
+ ("Mailing ZIP Code"), worded exactly as Essex words them. Adds `CALC_ACRE`
1381
+ ("Calculated Acreage"). **Removes `OWNER_NAME`**, as Essex does: it is
1382
+ blank on every row statewide, and the owner-field classifier takes the first
1383
+ owner-looking column, so listing it pointed owner lookups at a blank
1384
+ column. The column's absence of data stays on record through the existing
1385
+ `ownerFieldNote` and the reviewed `capabilityOverrides.owner_name`
1386
+ (`restricted`, Daniel's Law, quoted from the service description). Both are
1387
+ unchanged apart from the row count in the note.
1388
+ - `sampleQuery` now returns the mailing columns. `corsEnabled` `null` →
1389
+ `true` (`Access-Control-Allow-Origin: *`). `lastVerified` 2026-05-15 →
1390
+ 2026-10-04. `licenseUrl` and `contact.url` were `https://www.njogis.com/`,
1391
+ which no longer resolves (NXDOMAIN). They now point to the layer's ArcGIS
1392
+ item and to NJGIN, as Essex's do. `notes` rewritten to the measured facts.
1393
+ The old notes claimed "20,553+ parcels", and the scope holds 281,646.
1394
+ - **What a consumer sees:** the site's derived capability for
1395
+ `owner_mailing_address` on Bergen moves from `not_published` to
1396
+ `available`, the same as Essex. Owner-name search stays unavailable.
1397
+ - **Essex** (FIPS 34013) now lists `YR_CONSTR` ("Year of Construction") and
1398
+ `BLDG_DESC` ("Building Description") with Bergen's labels. Both are
1399
+ populated for Essex: 156,919 and 121,070 of 178,089 rows. With that change
1400
+ Bergen and Essex list the same composite columns. `lastVerified` is now
1401
+ 2026-10-04 and `notes` carry the two counts.
1402
+ - **Ocean (34029), Camden (34007) and Morris (34027)** stay on their own
1403
+ service, `maps.nj.gov` `Framework/Cadastral/MapServer/0`. It publishes the
1404
+ same 3,481,240 statewide rows as the composite, and its Essex counts match
1405
+ the composite's exactly. Measured on that service 2026-10-04:
1406
+
1407
+ | `COUNTY=` scope, test | OCEAN | CAMDEN | MORRIS | ESSEX (control) |
1408
+ |---|---|---|---|---|
1409
+ | all rows | 299,346 | 191,292 | 177,533 | 178,089 |
1410
+ | `ST_ADDRESS<>''` | 299,346 | 191,292 | 177,533 | 178,089 |
1411
+ | `CITY_STATE<>''` | 299,346 | 191,292 | 177,531 | 178,089 |
1412
+ | `ZIP_CODE<>''` | 299,327 | 191,292 | 177,514 | 178,088 |
1413
+ | `PROP_LOC<>''` | 299,346 | 191,292 | 177,531 | 178,089 |
1414
+ | `OWNER_NAME<>''` | 0 | 0 | 0 | 0 |
1415
+ | `CALC_ACRE>0` | 260,309 | 169,212 | 166,312 | 152,381 |
1416
+ | `YR_CONSTR>0` | 258,549 | 162,106 | 156,490 | 156,919 |
1417
+ | `BLDG_DESC<>''` | 260,928 | 153,496 | 146,953 | 121,070 |
1418
+
1419
+ For each of the three:
1420
+ - **`OWNER_NAME` is removed** from `searchFields`. It was listed as
1421
+ searchable, and Ocean's notes called it "public, queryable".
1422
+ - **Fields added:** `ST_ADDRESS`, `CITY_STATE` and `ZIP_CODE` with Essex's
1423
+ labels. `CALC_ACRE`, `YR_CONSTR` and `BLDG_DESC` are listed with Bergen's
1424
+ labels.
1425
+ - **`PAMS_PIN` relabelled** "PAMS Parcel Identifier (MUN_BLOCK_LOT)". The old
1426
+ label, "PAMS Property Identification Number", did not match the site's
1427
+ parcel-id reader, so `apn` derived as `not_published`. It now derives as
1428
+ `available`, and so does `owner_mailing_address`.
1429
+ - **`capabilityOverrides.owner_name` stays `not_published`**, with its
1430
+ `endpoint_schema` basis. This service states no reason for the blank
1431
+ column. The Daniel's Law attribution comes from the composite service's
1432
+ description, so `ownerFieldNote` and `notes` quote it as that service's
1433
+ statement. They do not record it as a restriction this endpoint cites.
1434
+ - **Other changes:** `notes` and `ownerFieldNote` are rewritten from the
1435
+ measured counts. Camden's layer carries 36 municipality names. Its old
1436
+ notes said 37 municipalities. Unverified population figures and clerk links
1437
+ are dropped, including oceanrecorder.com, which no longer resolves.
1438
+ - `corsEnabled` `null` → `true`, because the service reflects the request
1439
+ origin. `sampleQuery` is now a `PROP_LOC` search that returns the mailing
1440
+ columns. The old one searched the blank owner column and returned nothing.
1441
+ `lastVerified` 2026-05-15 → 2026-10-04.
1442
+ - **Bergen and Essex** `capabilityOverrides.owner_name.basis.sourceUrl` and
1443
+ `ownerFieldNote.source.url` move from `https://www.nj.gov/njgin/`, which
1444
+ does not carry the Daniel's Law statement, to the composite's ArcGIS item,
1445
+ `https://www.arcgis.com/home/item.html?id=533599bbfbaa4748bf39faf1375a8a9c`.
1446
+ Its summary reads "The OWNER_NAME field is redacted per Daniel's Law for
1447
+ public web access." Essex's `notes` and `ownerFieldNote` are dated
1448
+ 2026-10-04, matching its `lastVerified`. The counts are unchanged: 178,089
1449
+ rows, 0 with an owner name, 3,481,240 statewide, all re-measured 2026-10-04.
1450
+
1451
+ ### Data fix — Florida FDOR sample queries use a capped scoped `where`
1452
+
1453
+ - **Florida:** 57 county records use the Florida Department of Revenue
1454
+ statewide cadastral (`Florida_Statewide_Cadastral` FeatureServer layer 0)
1455
+ as their first endpoint. Their `sampleQuery` was a point+distance query.
1456
+ On 2026-10-04 the layer rejects every distance-buffered query with HTTP 200
1457
+ and error 400 "24204: The spatial reference identifier (SRID) is not
1458
+ valid", whether the point is sent in wkid 4326 or 3857, so the sample
1459
+ returned nothing. Each record's `sampleQuery` is now
1460
+ `where=(CO_NO=<n>)&outFields=*&resultRecordCount=1`. That is the record's
1461
+ own `scopeWhere`, composed the way the site's directory composes it. Fired
1462
+ live 2026-10-04: all 57 answered HTTP 200 with one row carrying that
1463
+ county's own `CO_NO`, median about 1 second, slowest under 10 seconds.
1464
+ Without the cap the layer returns its 2,000-row maximum. No `scopeWhere`,
1465
+ URL, field list or note changed.
1466
+ - The 10 Florida records that register FDOR as a **second** endpoint, after
1467
+ their own county layer, get the same sample. Fired live 2026-10-04:
1468
+ Duval, Manatee, Sarasota, Pinellas and Pasco answered one row with their
1469
+ own `CO_NO` in about half a second. Broward answered after 55 seconds cold
1470
+ and in 50 ms once warm. Miami-Dade, Hillsborough, Palm Beach and Lee did
1471
+ not answer it: about 55 seconds each, then HTTP 200 with error 400 "Unable
1472
+ to perform query", twice each. On those large counties the layer also
1473
+ fails a cap of 101 or 501, except Hillsborough, which answers both. That
1474
+ matches the slow-scan limit the Florida notes describe, because the layer
1475
+ does not index `CO_NO`. The old point+distance sample failed on all ten.
1476
+
1477
+ ## 0.6.15 — 2026-10-03
1478
+
1479
+ SDK change, no data change. `totals` stay at **246 counties / 258
1480
+ endpoints**. This release lets an agent tell which layers take a typed search,
1481
+ and fixes the deep-link helper, which sent people to a search that cannot run
1482
+ (GitHub #905). One existing output changes; see "Changed".
1483
+
1484
+ 68 endpoints carry `attributeSearch: "unsupported"`: the 67 Florida counties on
1485
+ the shared FDOR cadastral layer, whose county column is not indexed, and
1486
+ Orleans Parish LA, whose server never finishes a table scan. Those layers
1487
+ answer spatial queries and return owner, address and parcel fields on every
1488
+ hit. They do not answer a `where` clause. Until now the package exposed the
1489
+ flag but no reader for it, and `buildParcelLookupDeepLink` ignored it.
1490
+
1491
+ ### Added
1492
+
1493
+ - **`offersAttributeSearch(endpoint)`**: false when the layer refuses a
1494
+ `where` clause. Query such a layer by geometry, as its `sampleQuery` shows.
1495
+ It gives the same answer as the reader urbankitstudio.com uses, on every
1496
+ bundled endpoint; a monorepo test holds the two equal.
1497
+ - **`countyOffersAttributeSearch(county)`**: true when ANY of the county's
1498
+ layers takes a `where` clause. Miami-Dade answers true through its own layer
1499
+ although its FDOR layer is flagged.
1500
+ - **`buildParcelLookupDeepLink(endpoint, base?, county?)`**: an optional third
1501
+ argument, the `CountyRecord` (or any `{ stateSlug, countySlug, endpoints? }`,
1502
+ exported as the `DeepLinkCounty` type). When given, the county's PRIMARY
1503
+ layer (`endpoints[0]`) links to the county route
1504
+ `/tools/parcel-lookup/<state>/<county>` if it is flagged or shared
1505
+ (`scopeWhere` set). Only that route knows the county, so only it can scope a
1506
+ shared layer and show a flagged one as lookup by location. It always loads
1507
+ the primary layer, so any other endpoint keeps the generic `?endpoint=` link
1508
+ to itself: Broward's flagged FDOR layer sits behind the county's own layer,
1509
+ as it does in nine other Florida counties. Passed without `endpoints`, the
1510
+ county is taken as vouching that the endpoint is its primary.
1511
+ - **`canonicalLayerUrl(url)`**: one key per layer, used for that comparison.
1512
+ An http(s) URL keys on host and path, with the scheme, a default port, the
1513
+ query string, the fragment, a trailing `/query`, repeated slashes, dot
1514
+ segments, trailing slashes and letter case folded away. urbankitstudio.com
1515
+ keys its own registry lookup the same way.
1516
+
1517
+ ### Changed — existing output, as a bug fix
1518
+
1519
+ - `buildParcelLookupDeepLink` no longer adds `fieldHint` for a flagged layer,
1520
+ with or without the new argument. This changes what an existing call
1521
+ returns: Orleans Parish lost `fieldHint=OWNERNME1`, which pre-selected an
1522
+ owner search that never answers. The FDOR layers had no `fieldHint` before.
1523
+ Calls on unflagged layers return exactly what they returned in 0.6.14. It
1524
+ ships as a patch because the old link opened a search that never answers,
1525
+ and before 1.0 a bug fix that changes a returned string is still a patch.
1526
+ - The `attributeSearch` doc comment counted 67 flagged registrations. It now
1527
+ says 68 and names Orleans Parish.
1528
+ - The README's deep-link example showed `fieldHint=TaxName` for Kane County.
1529
+ The helper never produced that, because `TaxName` does not match its owner
1530
+ rule. The example now shows the real output.
1531
+
1532
+ ## 0.6.14 — 2026-10-03
1533
+
1534
+ Data refresh. No API change. `totals` move from **241 counties / 253
1535
+ endpoints** to **246 counties / 258 endpoints**; 240 of the 246 now carry a
1536
+ verified endpoint (was 235 of 241). Five endpoints are added: four new counties
1537
+ and Clackamas OR, which was already on record without one. Garfield OK is
1538
+ added without an endpoint. Only Sullivan TN adds an owner-name search. The
1539
+ other four endpoints answer by parcel number and, except Tulare, by address,
1540
+ and each records why owner data is absent in `capabilityOverrides` and
1541
+ `ownerFieldNote`.
1542
+
1543
+ ### Data refresh — five endpoints, one county without one
1544
+
1545
+ - **New Jersey:** Essex County (FIPS 34013, Newark), on the NJ Office of GIS
1546
+ statewide Parcels and MOD-IV composite (`Parcels_Composite_NJ_WM`, layer 0),
1547
+ scoped `COUNTY='ESSEX'` (178,089 of 3,481,240 rows). Searchable by PAMS
1548
+ parcel identifier (`PAMS_PIN`), property location (`PROP_LOC`),
1549
+ municipality (`MUN_NAME`) and the mailing street on the tax record
1550
+ (`ST_ADDRESS`). **Not searchable by owner:** `OWNER_NAME` is an empty string
1551
+ on every row statewide, and the NJ Office of GIS states in the service
1552
+ description that it is redacted per Daniel's Law for public web access, so
1553
+ the column is not listed and `owner_name` is `restricted`, attributed to
1554
+ NJOGIS, as for Bergen. Licence: public. Because the composite is now shared,
1555
+ Bergen's registration gains the matching `scopeWhere` `COUNTY='BERGEN'`.
1556
+ - **Oregon:** Clackamas County (FIPS 41005, Oregon City), already on record
1557
+ without an endpoint, now carries the county's `Taxlots_CMap` FeatureServer,
1558
+ layer 0 (163,927 taxlots). Searchable by parcel number (`PARCEL_NUMBER`),
1559
+ map and taxlot (`TLNO`), site address (`SITUS`) and city (`SITUS_CITY`).
1560
+ **Not searchable by owner and no mailing address:** the layer has neither
1561
+ column and the county publishes owner names on no public layer, routing owner
1562
+ questions to Assessment and Taxation; `owner_name` and
1563
+ `owner_mailing_address` are `not_published`. No scope (county layer).
1564
+ Licence: disclaimer only, recorded `unknown`.
1565
+ - **Nebraska:** Phelps County (FIPS 31137, Holdrege), on Nebraska's statewide
1566
+ `StatewideParcelsExternal` MapServer, layer 0, scoped `County_ID='137'`
1567
+ (7,664 of 1,154,898 rows; 59 are blank-attribute polygons, which the
1568
+ sample query skips with `Parcel_ID<>''`). Searchable by parcel ID (`Parcel_ID`,
1569
+ `State_PID`), situs address (`Situs_Address`) and legal description.
1570
+ **Not searchable by owner and no mailing address:** the statewide layer has
1571
+ neither column, and the county's gWorks vendor server exposes only a parcel
1572
+ key; `owner_name` and `owner_mailing_address` are `not_published`. CORS
1573
+ reflects the request origin. Licence: none stated, recorded `unknown`.
1574
+ - **California:** Tulare County (FIPS 06107, Visalia), on the county's
1575
+ `Parcels_(Public_View)` FeatureServer, layer 0 (166,875 parcels).
1576
+ Searchable by APN only (`APN`, nine-digit text, and `APNFormatted`) plus
1577
+ legal description. **No owner, no site address, no mailing address** in the
1578
+ layer; `owner_name`, `owner_mailing_address` and `situs_address` are
1579
+ `not_published`. `maxRecordCount` is 1000. No scope (county layer). Not the
1580
+ City of Tulare's layer at maps.tulare.ca.gov. Licence: none stated, recorded
1581
+ `unknown`.
1582
+ - **Tennessee:** Sullivan County (FIPS 47163, Blountville), on the City of
1583
+ Johnson City's regional `ParcelPublishing/TaxParcels` MapServer, layer 0,
1584
+ scoped `COUNTYNAME='Sullivan County'` (78,958 of 183,538 rows; the bare value
1585
+ `Sullivan` matches nothing). Searchable by owner name (`OWNER`), parcel ID
1586
+ (`GISLINK`, `ID`), property address (`ADDRESS`, stored street first and house
1587
+ number last) and owner mailing address (`MAILADDR`). CORS reflects the
1588
+ request origin. **Licence: `restricted`.** The item's terms say delivered
1589
+ products "are not to be resold by the purchaser for any reason and may not be
1590
+ reproduced without the written permission of the staff of the Johnson City
1591
+ GIS Division"; the record holds only the query URL and the county is served
1592
+ by live query.
1593
+ - **Oklahoma:** Garfield County (FIPS 40047, Enid), **no endpoint**
1594
+ (`hasPublicRest: false`). The assessor's Spatialest viewer is backed by a
1595
+ GeoServer that serves WMS images only with WFS disabled, and Oklahoma has no
1596
+ statewide parcel REST layer. The `notes` record the search, with controls, so
1597
+ it is not repeated.
1598
+
1599
+ All six states were already registered in `POPULATED_STATES`, so `src/*.ts`
1600
+ is unchanged and only `data/` grew. Deliberately a **patch**, for the reason
1601
+ 0.6.5 gave: adding county rows to an already-registered state is the 0.6.1
1602
+ shape, and `^0.6.0` reaches everyone already installed.
1603
+
1604
+ ## 0.6.13 — 2026-10-03
1605
+
1606
+ Data refresh. No API change. `totals` unchanged at **241 counties / 253
1607
+ endpoints**; one county moves from `unreachable` to `live`.
1608
+
1609
+ ### Data refresh — Orleans Parish LA is live again, for location queries
1610
+
1611
+ - **Louisiana:** Orleans Parish (FIPS 22071, New Orleans), `apps/property3`
1612
+ MapServer layer 15, `status` **`unreachable` → `live`**, `lastVerified`
1613
+ 2026-10-03. The layer was never down: it refuses any query that names its
1614
+ `outFields` (every named list answers HTTP 200 carrying ArcGIS error 400
1615
+ "Failed to execute query." in about 0.2 s) and never finishes a table scan
1616
+ (`1=1`, `OBJECTID>0` and `PARCELID IS NOT NULL` all ran past 40 s for a
1617
+ single row), which is the shape the old sample query and 463 consecutive
1618
+ probes used. With `outFields=*` the same layer answers a point + 30 m query
1619
+ in 0.28 s (40 features) and an 800 m envelope with geometry in 0.65 s (501
1620
+ features). The record now carries **`attributeSearch: "unsupported"`**: no
1621
+ `where`-clause search is offered, so a consumer that reads the flag offers
1622
+ lookup by location only. The owner columns (`OWNERNME1`, `OWNERNME2`) and the
1623
+ mailing address (`PSTLADDRESS`, `PSTLCITY`, `PSTLSTATE`, `PSTLZIP5`) are
1624
+ returned on every spatial hit. `sampleQuery` is an exact `PARCELID`
1625
+ equality with `outFields=*`, the one predicate shape that answers quickly.
1626
+ `corsEnabled` `null` → `true` (the header reflects the request origin).
1627
+ `notes` rewritten to the measured facts. Consumers that name their columns
1628
+ must send `*` to this layer; the site does so through its wildcard list.
1629
+ **TLS caveat, stated in `notes`:** the server sends an incomplete
1630
+ certificate chain (leaf without its Sectigo intermediate). Clients that
1631
+ fetch the missing intermediate themselves (curl on Windows, Chromium and
1632
+ Safari through the OS store) connect; Node's `fetch` fails the handshake
1633
+ with `UNABLE_TO_VERIFY_LEAF_SIGNATURE` (measured 2026-10-04). UrbanKit's
1634
+ billed enrichment and radius paths run on Vercel's Edge runtime, which
1635
+ cannot be given a certificate authority and whose handling of an incomplete
1636
+ chain is undocumented and not yet measured, so assume they cannot serve this
1637
+ parish until a keyed test on the deployed function says otherwise; today an
1638
+ Orleans row ends `county-unavailable` and is refunded, unchanged from
1639
+ before. The liveness probe pins the intermediate, so its ok describes the
1640
+ layer, not those paths. A Node consumer of this record must trust the
1641
+ Sectigo intermediate before any query runs. Tracked in urbankitstudio#906.
1642
+ **Field order:** `PARID` (the assessor's per-unit parcel id) is listed first
1643
+ and `PARCELID` (the GeoPIN, a lot id shared by every condo unit on a lot)
1644
+ second, so a consumer that keys rows by the first parcel-id field keys by
1645
+ unit: a 30 m buffer in the CBD returns 40 units on 3 lots with 39 owners.
1646
+
1647
+ ## 0.6.12 — 2026-10-03
1648
+
1649
+ Data refresh. No API change. `totals` move from **239 counties / 251
1650
+ endpoints** to **241 counties / 253 endpoints**; 235 of the 241 now carry a
1651
+ verified endpoint (was 233 of 239).
1652
+
1653
+ ### Data refresh — two counties added
1654
+
1655
+ - **Minnesota:** Mille Lacs County (FIPS 27095, Milaca), with a verified
1656
+ endpoint on the county's own `MilleLacs_Public_101` MapServer, layer 35
1657
+ (20,083 parcels with a taxpayer name). Searchable by taxpayer name
1658
+ (`TAXPAYER_NAME`, stored `LAST/FIRST`, e.g. `SMITH/BRIAN & ASHLEY`), parcel
1659
+ number (`PARCELID`), site address (`PROPERTY_ADDRESS`, `CITY_TWP_NAME`) and
1660
+ the taxpayer mailing address (`TAXPAYER_ADDRESS_1` to `_4`). The layer also
1661
+ has an `OWNER_NAME` column, but it is blank on about 97 percent of rows and
1662
+ names a different party where it is filled, so it is deliberately not listed
1663
+ and owner search uses `TAXPAYER_NAME`. CORS
1664
+ reflects the request origin. No license is stated.
1665
+ - **Missouri:** the independent City of St. Louis (FIPS 29510, no county; not
1666
+ St. Louis County, 29189), with a verified endpoint on the Assessor's
1667
+ `Assessor_Public_Parcels` MapServer, layer 11. Searchable by owner name
1668
+ (`OwnerName`, truncated to 40 characters at the source), parcel id
1669
+ (`ParcelId`), site address (`SITEADDR`, which carries runs of internal
1670
+ spaces, so search on a house number plus a street word) and the owner
1671
+ mailing address (`OwnerAddr`, `OwnerCity`). **The server sends no
1672
+ `Access-Control-Allow-Origin` header, so `corsEnabled` is `false`:** a
1673
+ browser on another origin cannot read it, and it must be queried from a
1674
+ server. No license is stated.
1675
+
1676
+ Minnesota and Missouri were already registered in `POPULATED_STATES`, so
1677
+ `src/*.ts` is unchanged and only `data/` grew. Deliberately a **patch**, for
1678
+ the reason 0.6.5 gave: adding county rows to an already-registered state is
1679
+ the 0.6.1 shape, and `^0.6.0` reaches everyone already installed.
1680
+
1681
+ ## 0.6.11 — 2026-10-03
1682
+
1683
+ Repairs three registrations the liveness probe flagged (Monmouth NJ, Prince
1684
+ George's MD, Pinal AZ), records why Greenville SC stays unreachable, fixes the
1685
+ three Arkansas scopes that returned nothing without an error, reorders
1686
+ Washoe NV's owner fields, and declares two endpoint fields the bundled data
1687
+ already carried. `totals` are unchanged: 239 counties / 251 endpoints.
1688
+
1689
+ ### Fixed: Arkansas county scopes
1690
+ - The statewide Planning_Cadastre layer now stores `countyfips` as the 3-digit
1691
+ county code. `scopeWhere` and `sampleQuery` for Pulaski, Benton and
1692
+ Washington move from '05119'/'05007'/'05143' to '119'/'007'/'143'. The old
1693
+ values matched 0 rows, so every scoped attribute search for these counties
1694
+ returned nothing. The new scopes count 180,229, 175,795 and 114,791 rows
1695
+ (2026-10-03).
1696
+
1697
+ ### Fixed: endpoints
1698
+ - Monmouth County NJ moves to the county's new ArcGIS Online layer
1699
+ (Monmouth_County_Parcels FeatureServer/0 on services1.arcgis.com). The old
1700
+ services9 GEO_MC_Parcels service is gone. Field names change: PAMSPIN,
1701
+ Owner_Name, Location, Class, Net_Value, Land_Value, Impr_Value, Sale_Date,
1702
+ Sale_Price, Year_Built, Zone. Owner mailing is now Owner_Street plus
1703
+ Owner_Csz, replacing CityStateZip, so the `mailing` role a consumer derives
1704
+ moves from a city/state/zip column to a street column. Acreage is dropped.
1705
+ Value and date fields are strings on this layer.
1706
+ - Prince George's County MD moves to the Planning Department's hosted copy
1707
+ (Property_Flattened_Py FeatureServer/0 on services1.arcgis.com), with the
1708
+ same field names. The county host answers "Layer not found" on about every
1709
+ second request. The hosted copy refreshes periodically, not live.
1710
+ - Pinal County AZ: the county renamed every field to long-form names (for
1711
+ example PARCELID to Parcel_Identification_Number, OWNERNME1 to
1712
+ First_Owner_Name). All searchFields and the sampleQuery are re-mapped. The
1713
+ layer now publishes owner mailing address (Postal_Address, Postal_City,
1714
+ Postal_State, Postal_Zip_Code_5/4), so `owner_mailing_address` for Pinal
1715
+ moves from not_published to available.
1716
+ - Greenville County SC: notes only. It stays `unreachable` because the county
1717
+ deleted its GreenvilleJS services folder.
1718
+ - Washoe County NV: LASTNAME now precedes FIRSTNAME in `searchFields`, so a
1719
+ consumer that takes the first owner column searches surnames, as the
1720
+ record's own sampleQuery does. A derived owner value for a Washoe row is
1721
+ therefore the surname alone (was the first-name column).
1722
+
1723
+ ### Added: two optional fields on `EndpointRecord`
1724
+ - `scopeWhere?: string`: the predicate ANDed into every attribute search on a
1725
+ layer shared by several counties. The bundled data already carried it, and
1726
+ the type now declares it.
1727
+ - `attributeSearch?: "unsupported"`: marks a layer that cannot serve a
1728
+ `where`-clause search. Set today on the 67 Florida FDOR registrations.
1729
+
1730
+ ## 0.6.10 — 2026-10-03
1731
+
1732
+ Data refresh. No API change. `totals` move from **235 counties / 247
1733
+ endpoints** to **239 counties / 251 endpoints**; 233 of the 239 now carry a
1734
+ verified endpoint (was 229 of 235).
1735
+
1736
+ ### Data refresh — four counties added
1737
+
1738
+ - **Georgia:** Dougherty County (FIPS 13095, Albany), with a verified
1739
+ endpoint on Albany GIS's `Parcels_Public_View` FeatureServer, covering the
1740
+ city and county jurisdictions (38,007 parcels). Searchable by owner name
1741
+ (`Name`), parcel number (`ParcelNum`), site address (`Address`) and the
1742
+ owner mailing address (`MailAddress1`, `MailCity`). String fields are
1743
+ space padded, so match with `LIKE 'SMITH%'`. The layer is in wkid 102667,
1744
+ so request `outSR=4326` for longitude and latitude. No license is stated.
1745
+ - **Alabama:** Mobile County (FIPS 01097), with a verified endpoint on the
1746
+ Mobile County Revenue Commission's `MCRC_Public_Parcels` FeatureServer
1747
+ (214,040 parcels, updated 2026-09-22). Searchable by owner name (`Name1`,
1748
+ `Name2`), parcel number (`Parcel_Number`) and site address (`PropAddr1`).
1749
+ The layer has no owner mailing-address columns. String fields are space
1750
+ padded, so match with `LIKE 'SMITH%'`. The layer is in wkid 102630, so
1751
+ request `outSR=4326` for longitude and latitude. No license is stated.
1752
+ - **Oregon:** Malheur County (FIPS 41045, Vale), with a verified endpoint on
1753
+ the county's `County_GIS_Portal_WFL1` FeatureServer (20,570 tax lots).
1754
+ Searchable by owner name (`OWNERSNAME`), map and taxlot id (`MapTaxlot`),
1755
+ site address (`SITUSADD`, `CITY`) and the owner mailing address (`ADD1`,
1756
+ `CITYSTATE`). The layer name and index are month-stamped (`TL Jun2026`,
1757
+ layer 16) and the county republishes monthly. A client that finds the
1758
+ layer gone should list the FeatureServer's layers and pick the one whose
1759
+ name starts with `TL ` and which carries `OWNERSNAME`. No license is
1760
+ stated.
1761
+ - **Washington:** Island County (FIPS 53029, Coupeville), with a verified
1762
+ endpoint on the Assessor's `Geocortex/Base` MapServer layer 0. Searchable
1763
+ by owner name (`taxpayer`), parcel id (`ParcelNo`), site address
1764
+ (`physical_addr`) and the owner mailing address (`mailing_addr1`,
1765
+ `mailing_addr_city`). Values carry trailing spaces, so trim them. CORS
1766
+ reflects the request origin rather than answering `*`. No license is
1767
+ stated.
1768
+
1769
+ Georgia, Alabama, Oregon and Washington were already registered in
1770
+ `POPULATED_STATES`, so
1771
+ `src/*.ts` is unchanged and only `data/` grew. Deliberately a **patch**, for
1772
+ the reason 0.6.5 gave: adding county rows to an already-registered state is
1773
+ the 0.6.1 shape, and `^0.6.0` reaches everyone already installed.
1774
+
1775
+ ## 0.6.9 — 2026-09-26
1776
+
1777
+ Adds the District of Columbia and fixes one helper. `totals` move from
1778
+ **234 counties / 246 endpoints** to **235 counties / 247 endpoints**; 229 of
1779
+ the 235 now carry a verified endpoint (was 228 of 234). `totals.states` moves
1780
+ from 50 to 51 because the index now lists DC beside the 50 states. It is not
1781
+ a state, so count `listStates()` entries by `slug` if you need states only.
1782
+
1783
+ ### Added: the District of Columbia
1784
+
1785
+ - `data/index.json` gains a populated `district-of-columbia` entry (abbrev
1786
+ `DC`), and `data/district-of-columbia.json` carries one record: `District
1787
+ of Columbia`, FIPS 11001, countySlug `district-of-columbia`, on DC GIS's
1788
+ Owner Polygons layer (Property_and_Land_WebMercator FeatureServer, layer 40).
1789
+ - Registered in `POPULATED_STATES`, so `findState("district-of-columbia")`,
1790
+ `findCounty("district-of-columbia", "district-of-columbia")`,
1791
+ `listCountiesByState` and `atlas.byStateSlug` resolve it. `listStates()`
1792
+ returns 51 entries.
1793
+ - The District has no counties. Its one record is the District itself, the
1794
+ county equivalent the Census Bureau lists as 11001. The layer maps land, one
1795
+ polygon per lot. A condominium building is a single lot on it, usually with
1796
+ no owner name, so an owner search finds lot owners, not unit owners.
1797
+
1798
+ ### Fixed: `countySlugFromName` knew four unit names
1799
+
1800
+ It recognised County, Parish, Borough and Census Area and appended " County"
1801
+ to every other name: `countySlugFromName("Anchorage Municipality")` returned
1802
+ `anchorage-municipality-county` while the bundled record is
1803
+ `anchorage-municipality`. The same was true of the nine Connecticut planning
1804
+ regions and Massachusetts' `Statewide` record. It now uses the site's rule,
1805
+ which also knows Municipality, City, Statewide, Planning Region and District
1806
+ of Columbia, and it trims the name first. A name ending in `City` now slugs as
1807
+ it stands, which is right for the independent cities and wrong for two
1808
+ Virginia counties the registry does not carry, James City County and Charles
1809
+ City County, whose base names end in City: the old rule gave
1810
+ `james-city-county`, this one gives `james-city`. If either joins the
1811
+ registry its stored name is the full `James City County`. If you stored slugs
1812
+ this function produced for one of those names, they change; the bundled
1813
+ `countySlug` values do not.
1814
+
1815
+ This is a **patch** for the reason 0.5.3 gave: `^0.6.0` has to reach everyone
1816
+ already installed.
1817
+
1818
+ ## 0.6.8 — 2026-09-26
1819
+
1820
+ Data refresh. No API change. `totals` move from **233 counties / 245
1821
+ endpoints / 50 states** to **234 counties / 246 endpoints / 50 states**;
1822
+ 228 of the 234 now carry a verified endpoint (was 227 of 233).
1823
+
1824
+ ### Data refresh — one county added, one record corrected
1825
+
1826
+ - **Ohio:** Mahoning County, with a verified endpoint. Its parcel layer
1827
+ answers only `outFields=*`: a query that names its fields is refused
1828
+ with "Failed to execute query." The record's `sampleQuery` already sends
1829
+ `outFields=*` and its `notes` record the quirk, so a client that builds
1830
+ its own query should request every field and read the columns it needs.
1831
+ - **Mississippi:** Hinds County's layer (MDEQ OPCGIS, HINDS_PARCELS) now
1832
+ refuses a named `outFields` list the same way, measured 2026-09-26 on
1833
+ the enrichment, radius and liveness query shapes; it answered a named
1834
+ list in May. Its `sampleQuery` now sends `outFields=*`, its `notes` say
1835
+ so, and `lastVerified` moves to 2026-09-26. The record also stops listing
1836
+ `MAILADD1`, `MCITY1`, `MSTATE1` and `MZIP1` under `searchFields`: those
1837
+ columns exist in the layer's schema but are blank on every one of its
1838
+ 114,814 parcels (`returnCountOnly` with `<field> IS NOT NULL AND <field>
1839
+ <> ''` counts 0 for each, against 114,547 for `OWNNAME` and 114,514 for
1840
+ `SITEADD`, measured 2026-09-26), so the county serves owner names and
1841
+ site addresses, not mailing addresses, and a client reading
1842
+ `searchFields` for a mailing column no longer finds one there. No count
1843
+ changes.
1844
+
1845
+ Ohio was already registered in `POPULATED_STATES`, so `src/*.ts` is
1846
+ unchanged and only `data/` grew. Deliberately a **patch**, for the reason
1847
+ 0.6.5 gave: adding county rows to an already-registered state is the 0.6.1
1848
+ shape, and `^0.6.0` reaches everyone already installed.
1849
+
1850
+ ## 0.6.7 — 2026-09-26
1851
+
1852
+ Data refresh. No API change. `totals` move from **227 counties / 241
1853
+ endpoints / 50 states** to **233 counties / 245 endpoints / 50 states**;
1854
+ 227 of the 233 now carry a verified endpoint (was 223 of 227).
1855
+
1856
+ ### Data refresh — six counties in four registered states
1857
+
1858
+ - **Michigan:** Lenawee County, with a verified endpoint, and Iron County,
1859
+ which ships with `hasPublicRest: false`. Iron's map viewer has no public
1860
+ query API, and Michigan has no statewide parcel layer to fall back on.
1861
+ - **California:** Riverside County. Its public layer carries APN, situs and
1862
+ mailing address but no owner name, and `capabilityOverrides.owner_name`
1863
+ records that as `not_published`.
1864
+ - **Maryland:** Baltimore County, the county rather than Baltimore City.
1865
+ - **Oregon:** Coos County, with a verified endpoint, and Clackamas County,
1866
+ which ships with `hasPublicRest: false` because the county does not
1867
+ publish owner names online.
1868
+
1869
+ All four states were already registered in `POPULATED_STATES`, so `src/*.ts`
1870
+ is unchanged and only `data/` grew. Deliberately a **patch**, for the reason
1871
+ 0.6.5 gave: adding county rows to an already-registered state is the 0.6.1
1872
+ shape, and `^0.6.0` reaches everyone already installed.
1873
+
1874
+ ## 0.6.6 — 2026-09-22
1875
+
1876
+ **No API change and no data change.** `totals` stay at **227 counties / 241
1877
+ endpoints / 50 states**, `src/` is untouched, and `data/` is byte-identical to
1878
+ 0.6.5. If you are on 0.6.5 you gain nothing functional by upgrading.
1879
+
1880
+ What you gain is that you can **verify where this came from**. 0.6.6 is the
1881
+ first release of this package published with a [provenance
1882
+ attestation](https://docs.npmjs.com/generating-provenance-statements): npm
1883
+ attaches one only on an OIDC publish from a **public** repository, and every
1884
+ release up to and including 0.6.5 was published from a private one, so none of
1885
+ them could have had it. From 0.6.6 the package publishes from
1886
+ [`urbankitstudio/atlas`](https://github.com/urbankitstudio/atlas), and
1887
+ `npm view @urbankitstudio/atlas` will show the signed link back to the commit
1888
+ and workflow run that built the tarball.
1889
+
1890
+ The package is still **authored** in the private monorepo — `data/` is generated
1891
+ from the parcel registry that lives there — and mirrored out. So the source of
1892
+ truth has not moved; only the thing that presses publish has.
1893
+
1894
+ Deliberately a **patch**, for the same reason 0.5.3, 0.6.1 and 0.6.5 gave: the
1895
+ README says to pin `^0.x`, and `^0.6.0` resolves to `>=0.6.0 <0.7.0`, so a patch
1896
+ reaches everyone already installed. A release that changes no API and no data is
1897
+ exactly what a patch is for, and a `0.7.0` would reach nobody without a manual
1898
+ bump — which would be the wrong shape for a change whose whole point is that
1899
+ existing users get a verifiable artifact.
1900
+
1901
+ ## 0.6.5 — 2026-09-20
1902
+
1903
+ Data refresh. No API change. `totals` move from **171 counties / 174
1904
+ endpoints / 50 states** to **227 counties / 241 endpoints / 50 states**;
1905
+ 223 of the 227 now carry a verified endpoint (was 166 of 171).
1906
+
1907
+ ### Data refresh — Florida
1908
+
1909
+ - **Florida: 11 → 67 counties, 10 → 77 endpoints.** Every Florida county is
1910
+ now registered, not just the largest ten. Florida was already a populated
1911
+ state (registered in `POPULATED_STATES` since before this release), so
1912
+ `findState`/`findCounty`/`listCountiesByState`/`atlas.byStateSlug` needed
1913
+ no source change — only `packages/atlas/data/florida.json` and
1914
+ `data/index.json` grew. `src/*.ts` is unchanged by this release.
1915
+
1916
+ Deliberately a **patch**, for the reason 0.5.3 and 0.6.1 gave: the README
1917
+ says to pin `^0.x`, and `^0.6.0` resolves to `>=0.6.0 <0.7.0` — so this
1918
+ reaches everyone already installed, and a `0.7.0` would not. Adding county
1919
+ rows to an already-registered state is exactly the 0.6.1 shape (Connecticut's
1920
+ nine planning regions, the four NYC boroughs), not the 0.4.0/0.5.0 shape
1921
+ (registering a *new* state key, which changes what an existing function
1922
+ returns for inputs it previously answered `undefined` for).
1923
+
1924
+ ## 0.6.4 — 2026-09-18
1925
+
1926
+ Data corrections. No API change. `totals` stay at **171 counties / 174
1927
+ endpoints / 50 states**; five of those counties publish no endpoint, and the
1928
+ README's coverage sentence is now checked against the data by a test, because
1929
+ the prose had been wrong twice.
1930
+
1931
+ - New York: the four boroughs that shared the id `ny-new-york` are now
1932
+ `ny-bronx`, `ny-kings`, `ny-queens` and `ny-richmond`.
1933
+ - South Carolina: Charleston County's parcels moved from `MapServer/61` to
1934
+ `MapServer/4`; re-pointed and re-verified 2026-09-17.
1935
+ - Texas: Fort Bend renamed its fields (`OwnerName` → `Owner_Name` and the
1936
+ rest); re-pointed rather than dropped, re-verified 2026-09-12.
1937
+ - California: San Diego's record now explains that every owner name comes back
1938
+ as the literal `Protected Per CA Gov Code 7928.205`.
1939
+ - Manifest: keywords gain `real-estate` and `property`; the README carries the
1940
+ Socket badge.
1941
+
1942
+ `^0.6.0` reaches everyone already installed.
1943
+
1944
+ ## 0.6.3 — 2026-09-03
1945
+
1946
+ Metadata only. No data or API change. The package author, the license
1947
+ holder and the README credit now name UrbanKit Studio instead of an
1948
+ individual. `^0.6.0` reaches everyone already installed.
1949
+
1950
+ ## 0.6.2 — 2026-09-02
1951
+
1952
+ Data corrections only. No API change. `totals` are now **171 counties / 174
1953
+ endpoints / 50 states**. A patch for the reason 0.6.1 gave: `^0.6.0` reaches
1954
+ everyone already installed.
1955
+
1956
+ ### Fixed — Oregon
1957
+
1958
+ - **Multnomah County pointed at Umatilla County's tax lots.** The layer's own
1959
+ description reads "a Umatilla County GIS layer depicting land partition and
1960
+ ownership within Umatilla County" (copyright Umatilla County GIS, 44,849 tax
1961
+ lots, and Multnomah's own layer counts 284,409). Multnomah now carries Multnomah County's
1962
+ own Tax Parcels layer on services5.arcgis.com: 56 fields, owner `NAME`,
1963
+ mailing `ADDR1`/`CITY`/`STATE`/`ZIP`, `SITUSADDR`, `DEED_DATE`, `SALE_DATE`,
1964
+ `SALE_PRICE`. Owner query verified live 2026-09-02.
1965
+ - **Lane County's path answered 404 "Service not found" on 2026-09-01** and was live again on
1966
+ 2026-09-02 (75 fields, owner query returns rows). The entry keeps its path, re-verified, and its
1967
+ notes name the county's other publication of the same parcels, `AT/AddressParcelSales/MapServer/2`
1968
+ (26 fields), as the fallback if the path drops again.
1969
+
1970
+ ### Added
1971
+
1972
+ - **Umatilla County, OR (41059)**: the tax lots layer that had been filed
1973
+ under Multnomah, now under its own county (owner `MAILING_NA`, mailing
1974
+ address, situs fields).
1975
+
1976
+ ### Tooling
1977
+
1978
+ - `scripts/audit-atlas-county-names.mjs` asks every endpoint who it says it is
1979
+ (layer and service metadata) and flags one that names another county of its
1980
+ state and not its own. Across all 174 endpoints on 2026-09-02 the only flags
1981
+ are statewide layers (Montana cadastral, Utah LIR) and New York City's
1982
+ MapPLUTO, which name several counties by design.
1983
+
1984
+ ## 0.6.1 — 2026-08-29
1985
+
1986
+ Data corrections only; no API change. `totals` are now **170 counties / 173
1987
+ endpoints / 50 states** (0.6.0 shipped 155 / 155 / 50).
1988
+
1989
+ Deliberately a **patch**, for the reason 0.5.3 gave: the README says to pin
1990
+ `^0.x`, and `^0.6.0` resolves to `>=0.6.0 <0.7.0` — so this reaches everyone
1991
+ already installed, and a `0.7.0` would not. Every change below corrects a claim
1992
+ a consumer could act on, and a correction is worth nothing if it does not
1993
+ arrive.
1994
+
1995
+ ### Fixed — claims about what a county can answer
1996
+
1997
+ - **Six counties no longer advertise owner-name search** their layers cannot
1998
+ answer (#530), and eight owner advisories that turned out to be FALSE were
1999
+ reverted (#541) — the audit predicate behind them (`COL <> ''` on an
2000
+ Oracle-backed layer) compares against NULL and is never true. The corrected
2001
+ audit re-ran across all 170 counties (#543).
2002
+ - **El Paso County CO** declares `situs_address` and `owner_mailing_address`
2003
+ as `not_published`: the columns exist and are null on every row, verified
2004
+ against a Denver control on the same layer (#543, #545).
2005
+ - **Cuyahoga County OH** re-pointed to the Fiscal Office's live layer after the
2006
+ county stopped its MyPLACE service; mailing address is now available there
2007
+ (#558). **Greenville County SC** and **Orleans Parish LA** are marked
2008
+ `unreachable`, with the measurement in `notes` (#558).
2009
+
2010
+ ### Added — counties and scoping
2011
+
2012
+ - **Connecticut** as its nine planning regions (Connecticut has no counties as
2013
+ Census geography; the retired county FIPS codes never resolved) (#538).
2014
+ - **Four NYC boroughs** and **Wayne, Macomb, Montgomery, Bucks, Chatham** (#529,
2015
+ #531).
2016
+ - Every county sharing a statewide layer now carries a `scopeWhere` (#529,
2017
+ #534, #558) — previously an owner search on a shared layer could return
2018
+ statewide matches under one county's name.
2019
+
2020
+ ### Housekeeping
2021
+
2022
+ `package-lock.json` still said 0.5.3 (never bumped at 0.6.0); the README said
2023
+ 155 counties; and this package's own test suite pinned 155 and had been red,
2024
+ unrun, since the data grew — `prepublishOnly` would have refused to publish.
2025
+ All corrected here, and a root-suite guard now keeps the pins honest.
2026
+
2027
+ ## 0.6.0 — 2026-08-13 (entry written retroactively on 2026-08-29)
2028
+
2029
+ Shipped without a changelog entry; this records it. `CountyRecord` in the
2030
+ published SDK had eleven fields while the registry it mirrors has sixteen —
2031
+ everything a human had reviewed (`ownerFieldNote`, `dataVintage`,
2032
+ `capabilityOverrides`, the related-city slugs) was in the bundle but
2033
+ unreachable through the public type. 0.6.0 exposed the reviewed capability to
2034
+ SDK consumers and documented why it does not derive (#409). Minor, because the
2035
+ public type widened.
2036
+
2037
+ ## 0.5.3 — 2026-08-05
2038
+
2039
+ Says plainly that `status` is not a live signal. No data changes; `totals` are
2040
+ unchanged (155 counties / 155 endpoints / 50 states).
2041
+
2042
+ Deliberately a **patch**, not a minor. The README tells you to pin `^0.x`, and
2043
+ for a `0.x` release `^0.5.2` resolves to `>=0.5.2 <0.6.0` — so a `0.6.0` would
2044
+ not reach anyone already installed. The point of this release is to correct a
2045
+ misleading claim, which is worth nothing if it does not arrive.
2046
+
2047
+ ### Fixed — a claim that read as a live signal
2048
+
2049
+ `status` and `lastVerified` describe what was true **when the version you
2050
+ installed was published**. This package ships static JSON and makes no network
2051
+ calls, so it cannot know anything more recent. `status: "live"` therefore means
2052
+ "a check passed before publish", not "this endpoint is up right now".
2053
+
2054
+ The gap is real and was measured: on 2026-08-04, Will County IL was being served
2055
+ as `status: "live"`, `lastVerified: "2026-06-27"` while the liveness probe had it
2056
+ **down with 4 consecutive failures**.
2057
+
2058
+ ### Added
2059
+
2060
+ - **`LIVE_STATUS_URL`** — exported as a runtime value so nobody hardcodes the
2061
+ endpoint. Points at `https://urbankitstudio.com/api/atlas/status`: keyless,
2062
+ refreshed every two hours.
2063
+ - Type-level documentation on `EndpointStatus`, `status` and `lastVerified`
2064
+ explaining what each actually asserts — visible in your editor on hover.
2065
+ - README section covering the two answers people get wrong: **`found: false`**
2066
+ means no observation exists (the county may be untracked, or never probed), and
2067
+ a **503** means the health store was unreachable. Neither says the endpoint is
2068
+ up. Treat both as unknown.
2069
+
2070
+ ## 0.5.2 — 2026-08-05
2071
+
2072
+ Data content patch. Published without a changelog entry at the time; recorded
2073
+ here after the fact.
2074
+
2075
+ ### Data refresh
2076
+
2077
+ - Resynced `packages/atlas/data/` to `public/data/atlas/`, clearing **all
2078
+ remaining dead-host endpoint references**. Montana moved to
2079
+ `gisservice.mt.gov/.../MapServer/1`. Verified against the published tarball:
2080
+ 58 files, 51 state JSONs, zero dead-host references.
2081
+
2082
+ ## 0.5.1 — 2026-07-20
2083
+
2084
+ Data content patch. No new counties, no API changes — `totals` are unchanged
2085
+ (155 counties / 155 endpoints / 50 states).
2086
+
2087
+ ### Data refresh
2088
+
2089
+ - Resynced `packages/atlas/data/` to `public/data/atlas/`, which had drifted
2090
+ ahead since the 0.5.0 sync (2026-06-28):
2091
+ - **King County, WA** — `notes` updated to replace the dead-end reference to
2092
+ `tise-fzs9` (a Metro vanpool-ridership dataset, not parcels) with the two
2093
+ live Socrata tables that carry King County's PIN format: Property Legal
2094
+ Descriptions (`4854-i48r`, filter by `parcel_number`) and Real Property
2095
+ Tax Receivables (`dkna-i698`, filter by `account_number`). Resolve an
2096
+ address to a PIN via eReal Property or Parcel Viewer first, then query
2097
+ either table with that PIN.
2098
+ - **Will County, IL** — added `ownerFieldNote`, a `CountyAdvisory` pointing
2099
+ consumers to the Will County Supervisor of Assessments' property search
2100
+ portal, since the public parcel REST layer exposes only the PIN (no
2101
+ owner-name or address field).
2102
+ - `data/index.json`: content (totals, `lastUpdated`, `states`) synced
2103
+ verbatim from `public/data/atlas/index.json`; the `version` field is kept
2104
+ on the SDK's own release lineage (0.5.0 → 0.5.1) rather than copied from
2105
+ the main site's independent content-version field (which reads `0.3.0`
2106
+ and tracks separately) — consistent with how this field has been set at
2107
+ every prior sync (0.1.0 → 0.5.0).
2108
+
2109
+ ### Breaking changes
2110
+
2111
+ None. Data-content correction only. Pin `^0.x` to track patches.
2112
+
2113
+ ## 0.5.0 — 2026-06-28
2114
+
2115
+ Data resync to the live atlas (**128 → 155 counties**). All 50 US states are
2116
+ now populated.
2117
+
2118
+ ### Data refresh
2119
+
2120
+ - Synced `packages/atlas/data/` to `public/data/atlas/` (+27 counties). Eleven
2121
+ states gained their first verified endpoints: Arkansas, Delaware, Hawaii,
2122
+ Maryland, Montana, New Mexico, North Dakota, Oklahoma, South Dakota,
2123
+ West Virginia, and Wyoming. Populated coverage went from 39 to all 50 states;
2124
+ several existing states also grew. `data/index.json` totals now read 155
2125
+ counties / 155 endpoints across 50 states.
2126
+ - Registered the eleven new states in the SDK's `POPULATED_STATES` map so
2127
+ `findState` / `findCounty` / `listCountiesByState` / `atlas.byStateSlug`
2128
+ resolve them. The smoke test's full-coverage guard (every `populated: true`
2129
+ state resolves through `byStateSlug`) now pins all 50.
2130
+
2131
+ ## 0.4.0 — 2026-05-31
2132
+
2133
+ Data resync to the live atlas (**117 → 128 counties**) plus a new FIPS lookup
2134
+ and a registry fix that makes 14 already-bundled states actually reachable.
2135
+
2136
+ ### New API
2137
+
2138
+ - `findCountyByFips(fips)` — look up a county by its 5-digit county FIPS code
2139
+ (state FIPS + county FIPS, e.g. `"17089"` → Kane County, IL). Scans all
2140
+ populated states. Non-5-digit / unknown / non-string input returns
2141
+ `undefined`; it never throws.
2142
+
2143
+ ### Bug fixes
2144
+
2145
+ - **14 bundled states were unreachable.** v0.3.0 shipped data files for
2146
+ Alabama, Alaska, Connecticut, Idaho, Kentucky, Louisiana, Maine,
2147
+ Massachusetts, Mississippi, New Hampshire, New Jersey, Rhode Island,
2148
+ South Carolina, and Vermont, but they were never added to the SDK's
2149
+ `POPULATED_STATES` map — so `findState` / `findCounty` / `listCountiesByState`
2150
+ / `atlas.byStateSlug` returned `undefined`/empty for all of them despite the
2151
+ JSON being present in the package. All 39 populated states are now registered.
2152
+ A new smoke test pins that every `populated: true` state in the index resolves
2153
+ through `byStateSlug` so this can't regress.
2154
+
2155
+ ### Data refresh
2156
+
2157
+ - Synced `packages/atlas/data/` to `public/data/atlas/` (+11 counties). Notable
2158
+ growth since v0.3.0: Florida 7 → 11, Texas 5 → 10, Ohio 5 → 7. Package
2159
+ `data/index.json` totals now read 128 counties / 128 endpoints across 50
2160
+ states (39 populated).
2161
+
2162
+ ### Breaking changes
2163
+
2164
+ None. New function + additive data only. Pin `^0.x` to track minor adds.
2165
+
2166
+ ## 0.3.0 — 2026-05-16 (consolidated overnight expansion)
2167
+
2168
+ Atlas grew **36 → 117 counties** across **17 → 39 populated states**. Five
2169
+ parallel expansion batches landed PRs #34, #35, #36, #37 + earlier #28.
2170
+
2171
+ ### New counties (81 added since v0.1.0)
2172
+
2173
+ **Batch 1 (PR #28)** — 10 counties: Orange CA, San Bernardino CA, San Diego CA,
2174
+ El Paso CO, Jefferson CO, Marion IN, Mecklenburg NC, Multnomah OR, Davidson
2175
+ TN (Nashville), Salt Lake UT. Added IN/OR/TN/UT as new states.
2176
+
2177
+ **Batch 2 (PR #34)** — 10 Tier-2 metros: Erie NY (Buffalo), Monroe NY
2178
+ (Rochester), Onondaga NY (Syracuse), Hamilton OH (Cincinnati), Ada ID (Boise),
2179
+ Dane WI (Madison), Milwaukee WI, Bergen NJ, Hartford CT (statewide CT CAMA),
2180
+ Anchorage AK (statewide AK). Added ID/WI/CT/AK as new states.
2181
+
2182
+ **Batch 3 (PR #36)** — 15 Pacific NW + West: Lane OR (Eugene), Marion OR
2183
+ (Salem), Spokane WA, Snohomish WA, Thurston WA (Olympia), Davis UT, Utah UT
2184
+ (Provo), Weber UT (Ogden), Washoe NV (Reno), Pinal AZ, Santa Clara CA,
2185
+ Alameda CA, Contra Costa CA, Sacramento CA, Ventura CA. (CA-suppressed-owner
2186
+ counties indexed for completeness; OR/WA/UT/AZ varies by state law.)
2187
+
2188
+ **Batch 4 (PR #32)** — 11 Northeast entries: Suffolk MA + Middlesex MA +
2189
+ Worcester MA + Norfolk MA + Essex MA + Plymouth MA (via MassGIS L3 statewide
2190
+ composite, one entry), Chittenden VT (Burlington), Cumberland ME + York ME,
2191
+ Monmouth NJ + Ocean NJ + Camden NJ + Morris NJ (NJOGIS statewide + Monmouth
2192
+ direct), Hillsborough NH + Rockingham NH (no public REST documented),
2193
+ Providence RI (no public REST documented). Added MA/VT/ME/RI/NH as new states.
2194
+
2195
+ **Batch 5 (PR #37)** — 20 Southeast counties: Duval FL (Jacksonville),
2196
+ Manatee FL, Sarasota FL, Cobb GA (Atlanta NW), DeKalb GA (Atlanta east),
2197
+ Forsyth NC, Guilford NC, Cumberland NC, Buncombe NC, Durham NC, Greenville SC,
2198
+ Charleston SC, Jefferson AL (Birmingham), Shelby TN (Memphis), Hamilton TN
2199
+ (Chattanooga), Hinds MS (Jackson), DeSoto MS, East Baton Rouge Parish LA,
2200
+ Orleans Parish LA (New Orleans), Fayette KY (Lexington). Added AL/SC/MS/LA/KY
2201
+ as new states. Also: corrected Broward FL endpoint from FDOR statewide
2202
+ (partitioned, unreliable cross-county filter) to BCPA's own server.
2203
+
2204
+ **Batch 6 (PR #35)** — 15 Midwest counties: Kent MI (Grand Rapids), Oakland MI
2205
+ (Detroit NW), Ramsey MN (St. Paul), Waukesha WI, Brown WI (Green Bay),
2206
+ Hamilton IN (Indy suburbs), Allen IN (Fort Wayne), Montgomery OH (Dayton),
2207
+ Stark OH (Canton), Johnson IA (Iowa City), St. Louis County MO, Shawnee KS
2208
+ (Topeka), Douglas NE (Omaha), Lancaster NE (Lincoln), Sarpy NE. Added
2209
+ MI/IA/MO/KS/NE as new states.
2210
+
2211
+ ### New states (22 added since v0.1.0)
2212
+
2213
+ Alabama, Alaska, Connecticut, Idaho, Indiana, Iowa, Kansas, Kentucky,
2214
+ Louisiana, Maine, Massachusetts, Michigan, Mississippi, Missouri, Nebraska,
2215
+ New Hampshire, New Jersey, Oregon, Rhode Island, South Carolina, Tennessee,
2216
+ Utah, Vermont, Wisconsin.
2217
+
2218
+ ### Engineering / hygiene fixes
2219
+
2220
+ - **Data sync**: 5 counties (Pima AZ, Suffolk NY, Westchester NY, Dallas TX,
2221
+ Bexar TX) were in `public/data/atlas/` but missing from `packages/atlas/data/`
2222
+ in v0.1.0 and v0.2.0. Synced — these counties now ship in the npm package.
2223
+ - **State registration**: indiana, oregon, tennessee, utah, maine, massachusetts,
2224
+ new-hampshire, new-jersey, rhode-island, vermont, alaska, connecticut, idaho,
2225
+ wisconsin, michigan, iowa, missouri, kansas, nebraska, alabama, south-carolina,
2226
+ mississippi, louisiana, kentucky added to the SDK's POPULATED_STATES map.
2227
+ - **Auto-prerender**: `scripts/sync-prerender-routes.mjs` (in the main app, not
2228
+ the SDK) now auto-generates the reactSnap include list from atlas data — no
2229
+ more manual route-list updates per atlas expansion.
2230
+
2231
+ ### Known non-issues
2232
+
2233
+ - Several counties don't expose owner-name on their public REST layer due to
2234
+ state law (CA Gov Code 7928.205, WA RCW 42.56.070(8)) or county policy.
2235
+ These are still indexed for completeness — the atlas marks each county's
2236
+ searchable fields, so downstream consumers can filter for owner-name
2237
+ availability.
2238
+ - NH Hillsborough + NH Rockingham + RI Providence: no public REST documented.
2239
+ Indexed with empty endpoints so atlas consumers can show "Not Yet Indexed"
2240
+ banners + invite contributions.
2241
+
2242
+ ### Breaking changes
2243
+
2244
+ None. Backward-compatible additions only. Pin `^0.x` to track minor adds.
2245
+
2246
+ ## 0.2.0 — 2026-05-15
2247
+
2248
+ (Superseded by 0.3.0 — see above for the consolidated changelog covering all
2249
+ expansion batches that landed between v0.2.0 and v0.3.0 publish.)
2250
+
2251
+ - Added 10 new counties: Orange CA, San Bernardino CA, San Diego CA, El Paso
2252
+ CO, Jefferson CO, Marion IN, Mecklenburg NC, Multnomah OR, Davidson TN, Salt
2253
+ Lake UT
2254
+ - Added 4 new states: Indiana, Oregon, Tennessee, Utah
2255
+ - Atlas: 46 counties across 21 states
2256
+
2257
+ ## 0.1.0 — 2026-05-15
2258
+
2259
+ - Initial public release
2260
+ - 36 counties across 17 states