Novarly ParcelSource checks and corrections

How we check our answers

Every report on this site is public data read by software, and software can be wrong. A wrong answer about a house is worse than no answer, so this page is the record of what we do about it: the checks that run every day, what happens as each report is built, and every correction we have had to make.

The promise. If a report you paid for was wrong, we tell you, fix it, and refund it. No argument.

Checked every day

Every day we ask each data source a paid report depends on a question whose answer we verified by hand: UK sources at 06:20 UTC, US sources at 15:20 UTC, clear of the USDA's overnight maintenance. If an answer changes or a source stops responding, we get an email the same day. The answers we check against are the ones below; the times are the last run at which each one held.

Last run 21 Nov 2026, 15:20 (UK time) — 16 of 19 checks passing.

SourceWhat we ask itFeedsLast verified
postcodes.io (ONS Postcode Directory)NR12 0PW resolves to Happisburgh, Norfolk, in EnglandEvery UK productpassed 21 Nov 2026, 06:21
Mining Remediation Authority, Development High and Low Risk Areas (live) and the coal mining reporting area (the MRA’s open GeoPackage, in our database)Bollington Town Hall (SK10 5JR) is in a Low Risk Area inside the coal mining reporting area, and Trafalgar Square is outside the coalfieldHomeGround, SearchCheckpassed 21 Nov 2026, 06:21
Mining Remediation Authority, coal mining reporting area GeoPackage (data.gov.uk release)The file the MRA publishes is still the release loaded into our databaseHomeGround, SearchCheck. A newer release means we re-load it, not a missing answer, so no page is flaggedpassed 21 Nov 2026, 06:21
Mining Remediation Authority, recorded mine entriesA hand-verified shaft near Wigan is found, and Trafalgar Square has noneHomeGround, Sinkcheck UK, Extendcheckfailed 21 Nov 2026, 06:21
last held 12 Nov 2026, 06:21
Our composited mine-entries mapThe Bollington Town Hall map renders with its street basemap (a blank map is a fraction of the size)HomeGround PDFfailed 21 Nov 2026, 06:21
last held 12 Nov 2026, 06:21
OpenStreetMap tilesA known tile returns a real imageHomeGround PDF mappassed 21 Nov 2026, 06:21
Environment Agency NCERM 2024 coastal erosion projectionsHappisburgh is a coastal frontage with five scenarios on the Kelling Hard to Lowestoft shoreline planTidelinepassed 21 Nov 2026, 06:21
Environment Agency Flood Map for PlanningA point inside Flood Zone 3 at Market Harborough reads as Zone 3, and the LE16 9JL postcode centre on Bath Street, inside the bounding boxes of three zone outlines but more than 400 m from all of them, reads as Zone 1Tideline, HomeGround, Redline, SearchCheckpassed 21 Nov 2026, 06:21
US Census Bureau geocoderThe Empire State Building geocodes to its own coordinatesUS sinking report, Fieldworkpassed 21 Nov 2026, 15:20
USDA SSURGO soil surveyA field near Ames, Iowa returns its mapped soil complexFieldworkpassed 21 Nov 2026, 15:20
USGS Karst in the United States (2014), in our databaseDowntown Orlando sits on carbonate rock under a shallow coverSinkcheck US, HomeGround USpassed 21 Nov 2026, 15:20
Florida Geological Survey subsidence incident reportsA known Tampa cluster returns at least one incident within 1 kmSinkcheck US, HomeGround USpassed 21 Nov 2026, 15:20
Our own databaseIt answersEverythingpassed 21 Nov 2026, 06:21
Energy Performance Certificate registerSK10 5JR has certificates on fileHomeGroundpassed 21 Nov 2026, 06:21
planning.data.gov.ukBathwick has listed buildings within 100 m and sits in a conservation areaExtendcheck, Redline, Landcheck. While it fails, Redline and Landcheck are not soldpassed 21 Nov 2026, 06:21
BGS 1:50,000 geology (WMS GetFeatureInfo)Ealing returns London Clay Formation bedrockExtendcheckpassed 21 Nov 2026, 06:21
Environment Agency 1 m LIDAR first-return DSM and DTMBA2 4DB returns a full 100 m square with heights between 0 and 250 mExtendcheckpassed 21 Nov 2026, 06:21
OpenStreetMap building footprints (Overpass)BA2 4DB has at least one building footprint within 50 mExtendcheckfailed 21 Nov 2026, 06:21
last held 11 Nov 2026, 06:20
NASA Earthdata Login — the bearer token our ASF HyP3 InSAR orders are submitted withHyP3 accepts the token and it has at least 14 days left (the fulfilment cron re-issues it at 21)GroundTruth — an expired token means paid InSAR orders cannot be submitted or collectedpassed 21 Nov 2026, 06:21

Checked as it is built

Any data source can fail, and when one does our reports say “no data” for that section rather than guessing. That is right for a section that can honestly be empty, such as no landfill within range, and wrong for one a report cannot do without, such as the coalfield status on a coal mining report. So every paid report records which sections came back empty as it is built, and a missing must-have section raises an alert at once.

Checked after it is sold

Every purchase emails the founder the exact report the customer received, PDF included, so a person looks at every sale.

Kept, not just shown

The public feeds we read are live-only: the storm overflow hub publishes no history, and hazard maps are replaced in place. So we keep our own record. Storm overflow status changes, every 30 minutes, since 5 August 2026. Satellite fire detections over Britain and Ireland since 1 January 2026. And from 3 September 2026, the findings behind each address check listed below, free or paid, with the time we found them. When a report says something, we can show what we saw and when.

Archived: HomeGround US address report, HomeGround UK address report, Tideline coastal report, Mining report, Redline site appraisal, Fieldwork farm report, Wellcheck water-table report, Landcheck detectorist report, Extendcheck extension report, Sinkcheck (UK and US), SearchCheck, BNG exemption check, Swimcheck point check, Postcode comparison. Pages that only read a published table, such as the Consent Clock, the sinking maps and the heat rankings, are not per-address answers and are not archived.

Corrections

Every fault that reached a customer or a published page, what caused it, and what changed. Newest first.

16 November 2026 · Redline and Landcheck

Redline and Landcheck stayed on sale while a source they need was down

What happened.
Redline, our solar site report, and Landcheck, our detectorist land report, both read planning records from planning.data.gov.uk. Our daily check of that platform failed on 23 October 2026 and on every day from 7 to 16 November 2026. On those days both reports stayed on sale, and neither page said that a source was down. Without those records Redline can only say “Not enough data” in place of a verdict, and Landcheck cannot check for scheduled monuments.
Cause.
Extendcheck already held back any section whose source had failed its daily check. Redline and Landcheck are sold whole, and nothing held them back. The note that tells visitors a source is down was also not set to show on these two pages.
Fix.
While the planning check is failing, Redline and Landcheck are not sold. Each page says the report is paused in place of the buy button and names the source that is down, and visitors can leave an email to hear when it is back. A request for the whole Extendcheck report is now also refused while any part of it is down. Extendcheck’s “Constraints within 50 m” section is also held back when the map of recorded mine entries cannot be read, because that section promises mine entries.
Who was affected.
No one bought either report while the check was failing. Our last purchase of any report was on 2 September 2026. Every paid Redline page shown in that time was our public sample, and no paid Landcheck page was shown. Searching every daily check we keep, which start on 2 September 2026, these are the only days the planning check failed.
Since then.
Tests check that a paused report cannot be bought, that each paused page shows the note and the paused box in place of the buy button, and that every check able to pause a sale also shows its note on that page. The tests fail if a report is put back on sale while its source is down.

15 November 2026 · The Explorer map

The Explorer map showed no live fires when the fire feed did not load

What happened.
The Explorer map has a layer of live fires in the UK and Ireland. When the feed behind that layer did not load, the map drew no fires and its key looked as normal, so it read as if no fires had been seen in the last 7 days. Searching every request log we keep, which start on 21 October 2026, the feed did not load 5 times: three times from 21:19 GMT on 6 November and twice from 20:52 GMT on 13 November. An Explorer page served at 20:52 GMT on 13 November was followed 3 seconds later by one of those failures, so that page most likely showed no fires. The same is likely for one served at 21:19 GMT on 6 November.
Cause.
The map treated a failed load of the fire feed as an empty one. It did not check whether the feed had answered with an error, so any failure became an empty list of fires. The UK InSAR layer on the same map had the same fault, but no failed load of it appears in our logs.
Fix.
The map now checks that each layer loaded. When the fire feed does not load, the key says “The live fire feed did not load, so fires are not shown.” in place of its colours, so the map no longer looks as if there were no fires. The UK InSAR layer does the same.
Who was affected.
Most likely two Explorer pages, one served at 21:19 GMT on 6 November and one at 20:52 GMT on 13 November, each until it was reloaded. They showed no fires when the feed held 217 and 184 detections from the previous 7 days, 6 and 66 of them from the last 24 hours. Our page log, which counts only visits made in a browser, records no visit to the Explorer since 21 October 2026, but it does not count every visit, so we cannot say that no one saw them. The wildfire map most likely failed to load at the same times; when it does, it shows an empty grey box, not a map of no fires.
Since then.
Tests make the feed fail in each way we know of: an error with and without a body, no connection, and an answer that is not a list. They check that the key says the layer did not load, that a feed that really holds no fires still shows as empty, and that the map still uses the loader that tells the two apart. The tests fail if the old handling is put back.

14 November 2026 · Renewable planning council pages, the solar planning guide and the FAQ

The same database refusal also happened on 3, 6 and 7 November, and some pages stayed wrong for more than two days

What happened.
Our entry of 13 November 2026 described one short database refusal. Searching every request log we keep, which start on 21 October 2026, we found three more: for under a second at 20:26 GMT on 3 November, for about 13 seconds from 21:19 GMT on 6 November, and for about 12 seconds from 21:33 GMT on 7 November. Pages built in those seconds had the same faults. On 3 November, 2 renewable planning council pages lost their chart of decisions by year, and one of them also lost its list of schemes. On 6 and 7 November, 360 of our 510 council pages were affected: 289 said the page was not found, 33 showed no list of schemes and no chart, 36 lost the chart, and 2 lost the list. The solar planning guide lost its national figures from 21:19 GMT on 6 November until 21:33 GMT on 7 November. The FAQ lost its measured Green Belt figures from 21:33 GMT on 7 November until 11:32 GMT on 9 November, and may also have left out its answer on how much renewable capacity has permission but was never built.
Cause.
The same as in our 13 November entry: when a lookup failed, our code answered as if it had found nothing, and the page was kept like any other. That entry said such a page was kept for up to a day. That was wrong. A page is rebuilt only on the first visit after a day, and that visit is still given the old copy, so a page can stay wrong for much longer. We had searched the logs only around the refusal we found, so we missed these three.
Fix.
The fix in our 13 November entry, live since 23:57 GMT on 13 November 2026, covers all of these pages. A failed lookup now stops the page and the site keeps serving the last good copy, a refused connection is tried twice more, and each running copy of the site holds fewer connections. No page from these three refusals was still wrong by then: a new version of the site rebuilt every page at 21:35 GMT on 9 November, and another at 16:08 GMT on 5 November.
Who was affected.
On 6 and 7 November, the 360 council pages, the solar planning guide and the FAQ, each until a visit more than a day later rebuilt it, or until 21:35 GMT on 9 November at the latest. On 3 November, the 2 council pages, until 16:08 GMT on 5 November at the latest. From 21:22 GMT on 6 November to 20:42 GMT on 9 November our site answered “not found” 798 times, on 273 of the council pages. It served the solar planning guide twice while it was wrong and the FAQ three times. Our 13 November entry left out the same kind of count: its 61 pages answered “not found” 75 times, from 22:27 to 22:53 GMT on 13 November. Our page log, which counts only visits made in a browser, records no visit to any of these pages while they were wrong, but it does not count every visit, so we cannot say that no one saw them. No report was sold and no one used the “email me when it’s back” form near any of these times. For some pages built in these seconds we cannot tell, because their code did not record a failed lookup: this page, which is kept for up to an hour, on all three days; the wildfire map, kept for up to 5 minutes, on 3 and 6 November; 9 urban heat pages on 3 November, none of which later answered “not found”; and the consent clock page on 6 November.
Since then.
Before we publish an entry, we now search every log we keep for earlier cases of the same fault, not only the day we found it, and we count every wrong answer the site gave, not only the visits our page log records. The tests described in our 13 November entry already cover these pages.

13 November 2026 · Renewable planning council pages and the solar planning guide

A short database refusal left 120 council pages and the solar planning guide empty or missing

What happened.
For about 15 seconds from 20:52 GMT on 13 November 2026 our database refused new connections. Pages built in those seconds were kept for up to a day. Of our 510 renewable planning council pages, 61 said the page was not found, 29 showed no list of schemes and no chart, and 30 lost the chart of decisions by year. The solar planning guide lost its national figures. We checked the other pages that read the same database and are kept for a day or more, including all 3,533 renewable site pages, the 34 urban heat pages, the planning constraint pages, the consent clock page and the FAQ, and none of them was affected.
Cause.
When a lookup failed, our code answered as if it had found nothing. So a failed lookup became “not found” or an empty list, and the page was kept for a day like any other. The same mistake was in the code behind the renewable site pages, the urban heat pages, the planning constraint pages, the FAQ, this page, the wild swimming map and the wildfire map. It was also behind the “email me when it’s back” form and the emails it sends, which could have said a product was back when we could not tell. The consent clock page did say when one of its figures could not be read, but kept that line for a day. The paid Redline site report could leave out its recorded approval rates and its transmission queue without saying so. Both the free Redline check and the paid report could say no decisions were recorded for an authority whose record could not be read. Neither could tell a failed read of nearby schemes from none, so a site could screen as favourable without them.
Fix.
A failed lookup now stops the page instead of answering with nothing. The site then keeps serving the last good copy of the page, and a page with no good copy answers with an error that is not kept. When the database refuses a connection, the lookup is tried twice more after a short wait. Each running copy of the site now holds at most 3 database connections and closes each one after 5 seconds unused, where it used to hold up to 10 for 10 seconds. The “email me when it’s back” form now asks you to try again when it cannot check, and no “back” email is sent unless the checks were read. The Redline check and report now say which lookup could not be read: the recorded approval rates, the transmission queue, the authority’s record or the nearby schemes. A site whose nearby schemes could not be read no longer screens as favourable. The same change was made behind every page named above, and the fix rebuilt every page.
Who was affected.
The 120 council pages and the solar planning guide, from 20:53 GMT on 13 November 2026 until this fix. Our page log records no visit to any of them in that time. No report was sold or built while the database refused connections, and no one used the “email me when it’s back” form.
Since then.
Tests make the database refuse the lookups behind these pages, the form and the Redline report, all at once and, for most of them, one at a time. They check that each page stops rather than showing an empty answer, that the form asks for a retry, that the report names each lookup it could not read, that a site whose nearby schemes could not be read does not screen as favourable, and that an empty answer from a working database still reads as empty. Other tests pin the cap of 3 connections, the 5-second close, and the two retries, which are made only when the database refused the connection.

12 November 2026 · HomeGround coal mining report, its PDF and the coal mining town pages

A locked mine-entry map stopped the paid coal mining report and left the town-page maps blank

What happened.
On the morning of 12 November 2026 the Mining Remediation Authority’s map of recorded mine entries began answering every request with a short web page saying “Token Required” in place of a map tile. It had answered normally at 06:21 UTC, when our daily check last read it. From then until this fix, no paid HomeGround coal mining report could be built, wherever the address: the report page and its PDF would fail with an error. The mine-entry map on a coal mining town page did not load unless a copy made before the lock was still cached, and nothing on the page said so. The interactive map in the paid report would have shown streets with no mine-entry layer, again with no note. Sinkcheck and Extendcheck read the same layer; both already said the mine-entry part was unchecked when it failed, so their answers were incomplete but not wrong.
Cause.
The service sends its “Token Required” page with a success status, not an error status. We checked only the status, so we took the page for a tile and tried to read it as an image. That failed, and the failure stopped the whole report rather than just its mine-entry line.
Fix.
Every tile is now checked to be a real PNG or JPEG image that decodes before it is used, and a mine-entry tile must also be see-through like the real layer, so an opaque image is not read as mine entries everywhere. When the mine-entry layer cannot be read, the paid report is still built and says recorded mine entries were not checked, which is not a finding that there are none; the PDF says the same and notes that the mine-entry map is not included. The town pages and the interactive map now say when the mine-entry layer did not load. The written summary at the top of the paid report may only mention mine entries to say they were not checked; any other claim about them is replaced by a summary built from the report’s own rows. The daily check now names what the service sent back, and the warning it shows on our product pages when the layer fails now covers Extendcheck as well as HomeGround and Sinkcheck.
Who was affected.
No report was sold while the layer was locked, and no paid coal mining report was rendered. Our page log records no visit to a coal mining town page or a paid coal mining report on 12 November. We found the fault the same morning. The Mining Remediation Authority has not said whether the lock is permanent; until the layer answers again, coal mining reports say mine entries were not checked.
Since then.
Tests give each reader a “Token Required” page, an image label over a web page, a corrupt image and an opaque one, and check that the report, the PDF, the town page, the map route and the daily check each say the layer was not read rather than failing or reading as clear of mine entries. A test also checks that a summary claiming no shafts nearby is replaced while the layer is unread.

12 November 2026 · HomeGround coal mining check, report, PDF and town pages, and the HomeGround free screen

We said planning applications in the coal mining reporting area go to the Mining Remediation Authority, and called the area statutory

What happened.
Where an address lies in the Mining Remediation Authority’s coal mining reporting area, the free coal mining check, the paid report and its PDF said that the area is statutory and that planning applications there go to the Authority. The coal line of the HomeGround free screen said the same, and so did the fallback text of the paid report’s written summary. Of our 60 coal mining town pages, 59 said it of their town. Neither claim is true of this area. The Authority’s user guide calls it the known extent of coal mining, used to decide whether a coal mining report is needed when a property is sold. The Authority’s planning advice runs on a different map, its Development High Risk Area, where most planning applications need a coal mining risk assessment, and the Authority says it is not consulted on householder work such as a home extension. Our guide page and the town pages also said planning applications in a Development High Risk Area need that assessment, without saying that householder work does not.
Cause.
The words were written when the coal mining check launched on 23 July 2026 and were not checked against the Authority’s user guide, which says what the area is for. The area is published under a title with the word “consultation” in it, and planning consultation runs on the separate Development High Risk Area, once called the Development Referral Area. Our words gave the area the High Risk Area’s planning role. The HomeGround free screen copied the line on 10 September. No test asked what the area means. The Pennington Flash drawing was checked against the guide on 11 November and says what the guide says, but the check stopped at that drawing.
Fix.
Each of these places now says the address is in the coal mining reporting area, the known extent of coal mining, where a coal mining report is used when a property is sold. The guide gives that name for the area itself. None calls it statutory or says anything about planning applications. When the area cannot be read, the free coal mining check, the paid report and its PDF say the coal mining reporting area was not checked; the HomeGround free screen then says nothing about the area. The facts we give the language model that writes the summary use the same words, and a summary that calls the area statutory or says applications go to the Authority is now rejected in favour of the fallback text. SearchCheck, the methodology page, the list of daily checks on this page and the summary we publish for AI assistants name the area the same way. The guide page and the town pages now say most planning applications in a Development High Risk Area need a coal mining risk assessment, and householder work does not.
Who was affected.
Every free coal mining check, paid report and PDF for an address inside the area since 23 July 2026, the HomeGround free screen for such an address since 10 September, and 59 town pages since 23 July. Our archive of answers, kept since 3 September, holds 25 free coal mining checks, 15 of them, at 14 points, inside the area; 5 HomeGround free screens, one of them inside it; and 24 answers inside it for the paid sample report for Pontypool, which is ours. Of the two reports sold, both on 2 September 2026, one is for an address inside the area, so its page and PDF carried the claim. Its link now shows the corrected words; a PDF saved before the fix keeps the old line. The other is outside the area. The summaries the language model wrote are not kept, so we cannot count those that repeated the claim. No figure, risk band or verdict changed: the fault was in what we said the area means.
Since then.
A test now reads every page, town page and PDF, and the code behind them, on both sites, and the summary we publish for AI assistants. It fails if any of them puts the word “statutory” up to 40 characters before “reporting”, “coal mining” or “coalfield”, or up to 30 characters after “reporting area” or “coalfield area”, in one sentence; uses the area’s old name; puts “referred” straight after “planning applications”, with or without “here”, “there” or “are” between; or says anything goes, is sent, passed or referred to the Authority, whether named as the Mining Remediation Authority, the Coal Authority, the Authority or the MRA. A second test, added with the drawing fixes of the same day, reads every drawing on both sites and every recipe that builds one, and fails on the old name or on applications being referred. Others pin the new line on each surface, and pin the summary check. That check rejects a summary that puts the address in the area when it is not; puts “statutory” near the word “area”; uses the old name; says an application goes, is sent, passed or referred to the Authority; says the Authority is consulted or that anyone must consult it; or says anything is referred to it. When we describe what a dataset means, we now quote its publisher’s user guide and cite it in the recipe or the code, as the Pennington Flash drawing does.

12 November 2026 · Homepage, Little Dawley drawing

The Little Dawley drawing said planning applications in its coal area go to the Mining Remediation Authority

What happened.
The coal sheet of the Little Dawley drawing called the area it draws statutory, and its key said planning applications there go to the Mining Remediation Authority. Neither is true of this area. The Authority’s user guide calls it the known extent of coal mining, used to decide whether a coal mining report is needed when a property is sold. The Authority’s planning advice runs on a different map, its Development High Risk Area, which this sheet does not draw.
Cause.
The words were written into the recipe that builds the drawing and were not checked against the Authority’s user guide. The Pennington Flash drawing, which draws the same area, was checked against the guide on the morning of 11 November and its words were corrected, but the Little Dawley recipe was not brought into line.
Fix.
The coal sheet now reads “All of it lies inside the coal mining reporting area, the Mining Remediation Authority’s known extent of coal mining.” Its key names the coal mining reporting area as “the known extent of coal mining, where a coal mining report is used when a property is sold”. Its source line now names the Authority’s coal mining reporting area file. The caption of the relief picture of the same ground, above the drawing on both homepages, now uses the same name. The share is unchanged: the whole sheet, checked again against the Authority’s published file.
Who was affected.
The drawing on the homepages of parcel.novarly.com and earth.novarly.com, from about 22:29 GMT on 10 November 2026 until the fix. No report or free check reads the drawing.
Since then.
A new test reads every drawing on both sites and every recipe that builds one, and fails on the old name or on planning applications being referred, and another pins the Little Dawley coal caption, key and source line.

12 November 2026 · Extendcheck extension report

The paid Extendcheck report said a home extension near a mine entry needs a coal mining risk assessment

What happened.
Where a recorded mine entry lay within 100 m, the constraints table of the paid Extendcheck report said the Mining Remediation Authority will want a coal mining risk assessment, and that its permission is needed to disturb ground near the entry. Until 26 October 2026 the row used the Authority’s old name, the Coal Authority. GOV.UK guidance says householder work such as a home extension does not need a coal mining risk assessment, though the council or building control can ask for one, and that the Authority is not consulted on householder applications. The permit it describes is needed for work that could disturb or enter coal seams, or affect old coal mine workings or mine entries, not for any ground near an entry.
Cause.
The sentence was written when Extendcheck launched on 10 September 2026 and was not checked against the GOV.UK guidance on home extensions in former coal mining areas.
Fix.
After “A recorded mine entry within 100 m.” the row now reads: “Householder work such as a home extension does not need a coal mining risk assessment, but the council or building control can ask for one. The Mining Remediation Authority is not consulted on householder applications. You need its permit before any work that could disturb or enter coal seams, or affect old coal mine workings or mine entries. GOV.UK guidance says to avoid building over or close to a mine entry.” The code cites the two GOV.UK guidance pages it follows.
Who was affected.
Every paid Extendcheck report rendered since 10 September 2026 for an address with a recorded mine entry within 100 m. No Extendcheck report has been sold, so no customer saw it. Our own sample report and any other paid render showed it only where a mine entry lies within 100 m. No verdict or other row changed.
Since then.
A test pins the new sentence and fails if the old claim comes back.

12 November 2026 · Block drawings on every page that shows one

Nine drawings did not link the Open Government Licence, and six sheets left out the attribution it asks for

What happened.
No credit line on the Bathwick, Happisburgh, Ilkley, Little Dawley, Pennington Flash, Petersfield, Ripon, Up Exe or Warham Camp drawings linked the licence; the Doncaster and Kirkstall drawings did. The coal sheets of the Doncaster, Kirkstall, Little Dawley and Pennington Flash drawings gave the licence’s name but not its sentence, which is the credit the licence asks for when, as with the Mining Remediation Authority, a provider gives no statement of its own. The Warham Camp grid sheet names Norfolk, checked against the Office for National Statistics county boundaries, but gave no credit for them. The Petersfield grid sheet gives the place and position of its sample postcode from postcodes.io, whose postcode data is under the licence, but gave no credit for it.
Cause.
The licence asks a reuser to give the attribution statement each provider specifies, to use its own sentence where a provider gives none, and “where possible” to link to it. Only the Doncaster and Kirkstall recipes set the link, the coal lines gave the licence’s name in place of its sentence, the Warham Camp recipe credited no source for the county it checks, and the Petersfield recipe none for the postcode it places. No test required either the link or the sentence on every drawing.
Fix.
Every line that named the licence as “Open Government Licence v3.0” now ends with its sentence, “Contains public sector information licensed under the Open Government Licence v3.0”, and links the licence. The Warham Camp grid sheet now carries the two statements the Office for National Statistics asks for with its boundaries. The Petersfield grid sheet now carries the Ordnance Survey, Royal Mail and National Statistics statements that postcodes.io gives with its postcode data, then the licence’s sentence, and links the licence.
Who was affected.
Every block drawing, from when it was first published until this fix. No report or free check reads the drawings.
Since then.
Tests now require every such line on every drawing to end with the licence’s sentence and link it, every grid sheet whose recipe reads the Office for National Statistics boundaries to carry its two statements, and the Petersfield grid sheet to carry the statements postcodes.io gives. The Petersfield build now stops if Ordnance Survey’s address for its OpenData licence no longer leads to version 3.0 of the Open Government Licence.

12 November 2026 · Redline, SearchCheck and HomeGround pages, Up Exe, Kirkstall and Doncaster drawings

Three flood notes said only Zone 2 takes in the outlines of past floods

What happened.
The flood note on the Up Exe, Kirkstall and Doncaster drawings said “It draws both zones as if there were no flood defences, and Zone 2 also takes in the outlines of floods on record.” The Environment Agency builds both zones from flood models and from the outlines of past floods. At Up Exe, 80.1% of the ground under the outlines marked recorded is drawn as Zone 3, and on all three drawings most of it is.
Cause.
The note took its wording from the Agency’s definition of Zone 2, which takes in “accepted recorded flood outlines”. The Agency’s page for the map also says the zones are “created using local flood model outputs, recorded flood outlines and national flood model information”, and each outline it serves says whether it is modelled, recorded or both. Nothing checked the note against those outlines.
Fix.
Each note now says what the Agency says and gives two figures worked out from the outlines it serves. At Up Exe it reads “The Agency builds both zones from flood models and from the outlines of past floods. It marks each outline as modelled, recorded, or both: those marked recorded, alone or with a model, cover 17.0% of the sheet, and 80.1% of that ground is Zone 3.” At Kirkstall the figures are 7.0% and 56.7%, and at Doncaster 9.8% and 64.7%. A second count from the same outlines, using the same cell-centre rule as the drawings, gives the same figures. The drawings, their keys and their other figures have not changed.
Who was affected.
The Up Exe drawing on the Redline page, from about 09:30 GMT on 11 November 2026, the Kirkstall drawing on the SearchCheck page, from about 15:00 GMT, and the Doncaster drawing on the HomeGround page, from about 17:31 GMT the same day. Each until the fix. No report or free check reads the drawings.
Since then.
Tests now pin the sentence and its two figures on each drawing, forbid the old wording, and hold each pair of figures to what the drawn zones allow. Each build now stops if the Agency’s page no longer says the zones are made from recorded flood outlines.

12 November 2026 · Extendcheck page, Bathwick drawing

The Bathwick drawing’s 21 June sun key added up to 99.9%

What happened.
The key of the 21 June sun sheet on the Bathwick drawing read 4.4%, 5%, 9.5%, 12.4% and 68.6%. Its five rows share out all of the open ground, so they should add up to 100%, but they added up to 99.9%.
Cause.
The recipe that builds the drawing rounded each row of the two sun keys to a tenth of a percent on its own. The test on every key in every drawing checks rows that give shares of the whole sheet. These rows give shares of the open ground, so it did not check them.
Fix.
The recipe now shares out the tenths so the rows of each sun key add up to 100%, with each row within a tenth of its true share, as the other drawings do. The “Under 2 hours” row now reads 4.5%. The other rows, the 21 December key and both captions have not changed: 81.0% of the open ground still gets six hours or more on 21 June.
Who was affected.
The drawing on the Extendcheck page, from about 23:20 GMT on 10 November 2026 until the fix. No report or free check reads the drawing: it is read only by the page.
Since then.
A test now requires both sun keys to add up to 100%, and each caption’s figure to be the sum of the rows it covers.

12 November 2026 · SearchCheck and HomeGround pages, Kirkstall and Doncaster drawings

Four credit lines on two drawings gave the year we fetched the data, not the year of the edition used

What happened.
The radon credit on the Kirkstall drawing read “Contains British Geological Survey materials © UKRI 2026. Radon Potential classification UK Health Security Agency © Crown copyright 2026. Open Government Licence v3.0”. The edition drawn, the Radon Indicative Atlas GB v3, was published in December 2022. The Office for National Statistics credit on the grid line of the Kirkstall and Doncaster drawings ended “Contains OS data © Crown copyright and database right 2026”, but the boundaries it credits are the December 2024 edition. The ground stability credit on the Doncaster drawing read “Contains British Geological Survey materials © UKRI 2026. Open Government Licence v3.0”, but the British Geological Survey’s record of version 8 of its GeoSure hex grid, the edition drawn, says it was created in 2019.
Cause.
Each of these credits, as the providers word them, leaves a year to fill in. The recipes that build the drawings filled it with the year we fetched the data, 2026, not the year of the edition used.
Fix.
The Kirkstall recipe now reads the year from the British Geological Survey’s record of the radon atlas, and stops if that record is for a different version from the one drawn. The radon credit now reads “Contains British Geological Survey materials © UKRI 2022. Radon Potential classification UK Health Security Agency © Crown copyright 2022. Contains public sector information licensed under the Open Government Licence v3.0”, which also gives the Open Government Licence’s own sentence in full. The Office for National Statistics credit now takes its year from the edition its title names, “(December 2024)”, and the build stops if the title names none. Both grid lines now end “Contains OS data © Crown copyright and database right 2024”. The Doncaster recipe now reads the year of the ground stability credit from the British Geological Survey’s record of each of the six hazard grids of version 8, and stops unless all six give the same year. All six say they were created in 2019. The ground stability credit now reads “Contains British Geological Survey materials © UKRI 2019. Contains public sector information licensed under the Open Government Licence v3.0”. The drawings and their figures have not changed.
Who was affected.
The radon credit on the Kirkstall drawing on the SearchCheck page, from about 15:00 GMT on 11 November 2026. The grid and ground stability credits on the Doncaster drawing on the HomeGround page, from about 17:31 GMT, and the grid credit on the Kirkstall drawing, from about 18:14 GMT, the same day. Each until the fix. No report or free check reads the drawings.
Since then.
Tests now require 2022 in the radon credit, 2024 in both grid credits and 2019 in the ground stability credit, and forbid the Kirkstall and Doncaster recipes to fill those years from the day of the fetch. The Office for National Statistics credit no longer takes a fetch day at all.

12 November 2026 · SearchCheck and HomeGround pages, Kirkstall and Doncaster drawings

Two flood captions named a source of flooding for ground where the Environment Agency gives none

What happened.
The flood caption on the Kirkstall drawing read “Where rivers could flood, as planners must read it: 13.7% of the sheet.” and the one on the Doncaster drawing read “Where rivers or the sea could flood, as planners must read it: 41.5% of the sheet.” Each share takes in outlines of floods on record whose source the Agency gives as unknown. They cover 0.2% of each sheet, so the captions named a source the Agency does not give for that ground.
Cause.
The captions described the zones by where the water comes from. The Agency builds its Flood Zones 2 and 3 from flood models and from the outlines of past floods, and it does not give a source for every outline. Nothing checked the captions against the sources the Agency gives on each sheet.
Fix.
Both captions now name what is drawn and keep their figures: “Flood Zones 2 and 3, as planners must read them: 13.7% of the sheet.” at Kirkstall and “Flood Zones 2 and 3, as planners must read them: 41.5% of the sheet.” at Doncaster. The notes, which give the share of unknown source, the drawings and their figures have not changed.
Who was affected.
The Kirkstall drawing on the SearchCheck page, from about 15:00 GMT on 11 November 2026, and the Doncaster drawing on the HomeGround page, from about 17:31 GMT the same day. Each until the fix. No report or free check reads the drawings.
Since then.
Tests now pin both captions, require each note to give the share of unknown source, and forbid either caption to name rivers or the sea.

12 November 2026 · Wellcheck water-table check and report

Depths to water were called depths below ground, when each is measured from the borehole’s measuring point

What happened.
Wellcheck shows a depth to water at each Environment Agency monitoring borehole it reads. The free check and the paid report called that depth “below ground”. A level at the datum read as “at ground level”, and water standing above it read as “ABOVE ground level”, with “artesian conditions” stated as a fact. The Wellcheck page and our FAQ said the same: that each borehole has a surveyed ground datum, and that the check shows the depth below ground. The depth is not measured from the ground. It is measured from the point the Agency measures that borehole from, which is not always the ground.
Cause.
The Environment Agency’s hydrology service gives each borehole a datum, which it defines as “the datum level for the measurement point”. We read that datum as the ground level at the borehole. The two need not be the same.
Fix.
What each depth is measured from is now named. A depth reads as below “the measuring point”, a level at the datum as “at the measuring point”, and the headline depth for water above it as “ABOVE the measuring point, which suggests confined (artesian) pressure”, because water above that point is a sign of confined pressure, not proof of it. The range and driest-ever lines say only “above the measuring point”. The report says each depth is below the point the Agency measures from at that borehole, which is not always the ground. The Wellcheck page, the FAQ and llms.txt say the boreholes have a surveyed measuring point, and a borehole with no usable datum says the feed gives it no usable measuring-point height.
Who was affected.
Every Wellcheck free check and paid report that showed a depth to water, from launch on 20 August 2026 until this fix, and the Wellcheck page and FAQ answer over the same time. The figures themselves were right: each is the borehole’s datum less the measured level, and that calculation has not changed. Only the words for what they were measured from were wrong. Where a measuring point is not at the ground, the depth below the ground differs from the figure shown by the distance between that point and the ground.
Since then.
Tests render the station card and require “below the measuring point” and never “below ground”, and require the headline depth for water above the point to read as a sign of confined pressure, not as artesian conditions. Another test reads the Wellcheck page, the report, the station cards, the FAQ, llms.txt and the groundwater code, and fails on “below ground”, “ground datum” or “ABOVE ground level”.

11 November 2026 · England flood zone: Tideline, HomeGround, Redline, SearchCheck, Compare and Extendcheck

A flood zone read from a polygon’s bounding box, and a daily check that passed because of it

What happened.
For England addresses the planning flood zone line could name a zone the point is neither in nor near. Our own daily test postcode in Market Harborough read as “Flood Zone 3 — high probability, mapped within ~100 m of this point” when the nearest Zone 2 outline is about 370 m away and the nearest Zone 3 about 470 m. The same line feeds Tideline, the HomeGround coal mining report and free screen, Redline, SearchCheck, Compare and Extendcheck. In Redline a wrong Zone 3 was also listed as an adverse factor and could hold the verdict at constrained; in SearchCheck it marked a flood search as indicated. Separately, since its launch on 10 September the paid Extendcheck report added a “Flood Zone 2” row, with a note that a flood risk assessment is needed, to the constraints table of every address in Zone 1. The Extendcheck sample card on the site said its Bath address was in Flood Zone 3; it reads as Zone 1.
Cause.
The Environment Agency service answers a box query with every zone whose bounding rectangle touches the box, not every zone whose outline does. Flood zones are long, river-shaped polygons, so their rectangles take in a great deal of dry ground, and we took any zone returned as the answer without testing its outline. The request also asked for at most five zones and was never checked for a cut-off answer, the box reached about 130 m north and south but only 80 m east and west, and a reply with no feature list was read as Zone 1. The 26 October correction changed the wording to “within ~100 m” and kept the box, reasoning that the daily check depended on it. That check’s reference point was itself a false positive, so it passed every day because of the fault. The Extendcheck row came from matching the words “Zone 2” inside the sentence “no Zone 2/3 mapped”.
Fix.
Every outline the service returns is now tested against the point. The zone that contains the point is the answer, and a higher zone whose edge lies within 100 m is reported beside it with its measured distance, for example “Flood Zone 1 (lowest risk) at this point — Zone 2 is mapped about 50 m away”. The search box now reaches about 120 m north, south, east and west of the point and up to 99 zones are read. A failed, cut-off or unreadable answer is never shown as Zone 1: Tideline, HomeGround and Redline say the zone could not be read, SearchCheck calls the flood search unresolved, and Compare and Extendcheck say the zone is not available. Where the point is in Zone 1 with a higher zone within 100 m, each product now says so. The HomeGround free screen shows an amber note; in the paid report the flood marker turns amber and names the highest zone mapped nearby with its distance, and the zone scale draws each nearby zone with a dashed outline. SearchCheck marks the flood search as indicated and says why. Redline lists a nearby zone as a neutral factor and will not return a favourable verdict when the zone nearby is Zone 3; the flood line in its PDF is amber, not green. Compare says so in its flood row and makes no call between two addresses whose possible zones overlap. Extendcheck takes its flood rows from the zone at the point and adds a row, marked as mapped nearby, for a higher zone within 50 m. The sample card is corrected.
Who was affected.
Every England flood answer archived since 3 September 2026 was re-run on 11 November with the corrected read and checked against the service’s own point and 100 m geometry queries; the two agree at every address. Eight England points carry an archived flood answer. Three were told Zone 3 with no Zone 2 or 3 within 100 m. One is NR12 0PW, the Norfolk postcode behind our own Tideline and Redline samples (31 renders, all ours): the nearest zone is about 140 m away. Redline listed Zone 3 there as an adverse factor; its verdict rests on a listed building at the point, not on the flood zone. The other two are free checks, of a Leicestershire postcode (two views, nearest zone about 370 m away) and a Norfolk postcode (one view, about 115 m). Two were told Zone 3 and are in Zone 1 at the point with Zone 3 genuinely close. One of these is the postcode of one of the two reports sold, on 2 September 2026, the day before the archive began; its archived answer is from a free check of the same postcode on 10 September. Located at its postcode centre, that report now reads Zone 1 with Zone 2 under 5 m away and Zone 3 about 15 m away, so the caution it carried stands but the zone at the point was overstated. It is corrected at its existing link. The other is a free check with Zone 2 about 40 m away and Zone 3 about 70 m. Two more, both free checks, were told Zone 1 and still read Zone 1. The eighth is BA2 6RG in Bath, our own Extendcheck sample address and not a customer’s: it printed Zone 3 on nine renders between 10 and 24 September and reads Zone 1 today under both the old and the new read, so we cannot say which fault produced those nine; its six renders since 26 October carried the unfounded Zone 2 row. The second sold report’s postcode is not among the eight, as it has no archived flood answer; it reads Zone 1 under both the old and the new read. No Extendcheck report has been sold. The cut-off, the narrow box and the unreadable-reply faults could each have understated a zone; no archived answer shows it. Reports rendered before the archive began on 3 September, and pages viewed without an archived answer, may have carried an overstated zone.
Since then.
Tests pin the outline test, the measured distance to a nearby zone, the cut-off and unreadable-reply handling, the Zone 1 sentence producing no Extendcheck flood row, and each product’s handling of a zone mapped nearby. The read is also checked against the Environment Agency’s own outlines for Market Harborough by a second calculation that shares no code with it. The daily check now asks two questions the old read could not both pass: a point inside Zone 3 at Market Harborough must read Zone 3, and a point in Zone 1 must read Zone 1 with nothing nearby. On 12 November 2026 the Zone 1 question moved to the LE16 9JL postcode centre on Bath Street, Market Harborough. It lies inside the bounding boxes of three zone outlines and more than 400 m from all of them. Before that the Zone 1 point was our test postcode described above. When it fails, the alert it sends us names what went wrong at each point that failed. We send two other box queries to the same Environment Agency feature service. The historic landfill read reports the point as on a site when it falls inside the site’s outline, and a point inside a hole in the site also counts as on it; otherwise it reports a site when a corner point of the outline is within 500 m. The coastal erosion read asks for a box reaching about 1.7 km north and south and 1.0 km east and west, counts any frontage the service returns, however far away it is, and measures each frontage’s distance to its nearest corner point. Neither checks for a cut-off answer, and both read a reply with no feature list as nothing there.

11 November 2026 · Homepage, Little Dawley drawing

The Little Dawley drawing showed more radar tiles than its key counted, gave the main rock only part of its share, and called its made ground levelled

What happened.
The satellite radar sheet of the Little Dawley drawing showed 156 steady tiles and 5 sinking ones, but its key counted 138 steady cells and 4 sinking. The extra tiles were parts of cells whose centres lie off the sheet, showing along its edges. The rock caption said “6 units, 63.8% of the sheet Halesowen Formation (mudstone, siltstone and sandstone).” The British Geological Survey’s map splits that formation by rock type, so it has two of the six rows in the key, and together they cover 86.8% of the sheet. The six rows carry four names, which the caption did not say. The made-ground caption said “Ground that people have tipped, filled or levelled”. On this sheet the map marks made ground and infilled ground but no landscaped ground, so “levelled” said more than the map shows.
Cause.
The recipe that builds the drawing fetches the European Ground Motion Service’s 100 metre cells from a box reaching 150 metres beyond each edge of the sheet. It counted only the cells centred on the sheet, but drew the steady and sinking cells it fetched even when their centres lay off the sheet. The rock caption printed the first row’s share as if it were the whole formation’s. The made-ground caption’s words were fixed in the recipe, not taken from the kinds the map marks on this sheet.
Fix.
The recipe now leaves out a cell centred off the sheet before it counts or draws it, so the sheet shows 138 steady tiles and 4 sinking, one for each cell in the key. The caption’s counts, 142 cells measured and 4 sinking, have not changed. The rock caption now reads “6 units of rock under 4 names. Halesowen Formation makes up 2 of them and 86.8% of the sheet.” The made-ground caption now reads “Ground the geological map marks as made ground or infilled ground: 26.3% of the sheet.” The still image shown before the drawing loads was made again from the corrected file.
Who was affected.
The drawing on the homepages of parcel.novarly.com and earth.novarly.com, from about 22:29 GMT on 10 November 2026 until the fix. No report or free check reads the drawing.
Since then.
A test counts the separate tiles drawn on each radar sheet and requires one for each cell its key counts. Another checks each drawing’s bedrock caption against its key. A count of units must match the rows. When a name has more than one row, the caption must say how many names the rows carry, and the share it gives that name must be that of all its rows. A third requires each made-ground caption to name the kinds its key lists, and nothing more.

11 November 2026 · Homepage, both sites

The homepage counted seven public records in a drawing made from five public datasets

What happened.
The homepages headed the Little Dawley drawing “One piece of ground, seven public records.” The drawing has seven sheets, but the first is an empty sheet of the National Grid with no record on it, and the rock and made-ground sheets come from the same British Geological Survey map. Its six sheets of data come from five public datasets. The text under the heading also said “A report is this stack, read at one address.”, but each report reads the records it needs, not this stack.
Cause.
The heading was written by hand from the number of sheets in the drawing. That count took the empty grid as a record and one map as two.
Fix.
The heading now reads “One piece of ground, five public datasets.” and the sentence reads “A report reads records like these at one address.”
Who was affected.
The homepages of parcel.novarly.com and earth.novarly.com, from about 22:29 GMT on 10 November 2026 until the fix. No report or free check was affected.
Since then.
A test counts the sheets with data in the drawing’s file and the datasets they come from, and requires the parcel.novarly.com heading to give that number of datasets. It also forbids “A report is this stack”.

11 November 2026 · Block drawings on five pages

Five drawings gave key shares that did not add up to their captions or to 100%

What happened.
In five drawings the shares in a key, each rounded on its own, did not add up to the figure the caption gives for them together, or to 100% where they cover the sheet. On the Doncaster drawing the flood key read 5.6% and 35.8% under a caption of 41.5%. On the Petersfield drawing the source protection zones read 1.2%, 6.1% and 21.8% under a caption of 29.2%. On the Little Dawley drawing the made-ground key read 25.5% and 0.9% under a caption of 26.3%. The rock keys of the Pennington Flash and Ripon drawings added up to 100.1%.
Cause.
The recipes rounded each row of a key to a tenth of a percent on its own, and nothing checked the rows against the caption.
Fix.
The recipes now share out the tenths so the rows add up to the caption’s figure, with each row within a tenth of its true share. Five rows moved by a tenth: Flood Zone 2 at Doncaster now reads 5.7%, the outer zone at Petersfield 6.2%, made ground at Little Dawley 25.4%, the Wilmslow Sandstone Formation at Pennington Flash 0.4% and the Brotherton Formation at Ripon 42.7%. This fix changed no caption.
Who was affected.
The Little Dawley drawing on the homepages of parcel.novarly.com and earth.novarly.com, from about 22:29 GMT on 10 November 2026. The Pennington Flash, Petersfield and Ripon drawings on the coal mining, Wellcheck and Sinkcheck UK pages, from about 09:30 GMT on 11 November 2026. The Doncaster drawing on the HomeGround page, from about 17:31 GMT on 11 November 2026. Each until the fix. No report or free check reads the drawings.
Since then.
A test checks every key in every drawing. Rows that cover the sheet must add up to 100%, and rows under a caption that gives their total must add up to that total. Either may fall short by at most a tenth for each row too small to print a figure.

11 November 2026 · Tideline page, Happisburgh drawing

The Happisburgh drawing gave the upper-end figure as 158 to 160 metres, though only the 158 metre stretches colour ground on it

What happened.
The upper-end sheet of the Happisburgh drawing said “158 to 160 metres by 2105”, and its key read “158 to 160 m”. Of the four stretches of coast on the sheet, only the two at 158 metres add ground beyond the 2105 zone. The upper-end zone of the two at 160 metres lies wholly inside the 2105 zone on this sheet, so all the ground coloured for the upper end comes from the 158 metre stretches. The range put a figure on ground the drawing does not colour.
Cause.
The recipe that builds the drawing took the range from every stretch whose upper-end zone touches the sheet, not from the stretches that add ground to the strip it colours. The fact-check before it shipped confirmed the Agency’s figures, not which of them the coloured strip shows.
Fix.
Each strip now takes its figures from the stretches that add ground to it. The key now reads “158 m”, and the caption says “158 metres by 2105 on two of the four stretches of coast on this sheet. On the other two the figure is 160 metres, but on this sheet their upper-end zone lies wholly inside the 2105 zone, so no more ground is coloured there.” The 2055 and 2105 sheets and the drawing itself have not changed.
Who was affected.
The drawing on the Tideline page, from about 09:30 GMT on 11 November 2026 until the fix. No report or free check used it: the Tideline report reads the Agency’s zones for the frontage nearest the postcode, not the drawing.
Since then.
A test requires the upper-end key and caption to give only the figures of the stretches that add ground to the strip. The build now stops if any stretch adds no ground to the 2055 or 2105 strip.

11 November 2026 · Sinkcheck UK page and FAQ page

Two answers called gypsum the fastest-dissolving rock, and one left out the condition on its hundredfold rate

What happened.
On the Sinkcheck UK page, the answer to “What actually causes sinkholes in Britain?” said “Gypsum is the fastest” and that it “dissolves about a hundred times quicker than limestone”. The British Geological Survey’s geologists give that rate only for gypsum in flowing water next to rivers. On the FAQ page, the answer to “Could a sinkhole open up under my house? (UK)” said “Ripon’s gypsum is Britain’s fastest”. Both answers list salt among the rocks that dissolve, and British Geological Survey work puts salt at about a thousand times faster than gypsum under similar conditions.
Cause.
The Sinkcheck answer gave the figure without the condition its source attaches to it. Both answers ranked gypsum first with no source for the ranking, though each names salt. No check tied either claim to a source.
Fix.
The Sinkcheck answer now says “Gypsum is far more soluble than limestone, and in flowing water beside a river it can dissolve about a hundred times faster.” Its sentence on Ripon now says it comes from the British Geological Survey. The FAQ answer now lists the rocks as “chalk, limestone, gypsum or salt” and ranks none of them. The sources, in their own words, are recorded in the Sinkcheck page’s source beside the answer.
Who was affected.
The Sinkcheck UK page and the FAQ page, and the questions data each publishes for search engines, from the launch on 9 September 2026 until the fix. The free check and the reports do not use either answer.
Since then.
A test requires the Sinkcheck answer to give the rate only for flowing water beside a river and forbids “Gypsum is the fastest”. It also requires the sources, with their words, in the page’s source beside the answer. Another test requires the FAQ answer to list the four rocks and forbids it to call any of them the fastest.

11 November 2026 · Landcheck page, Warham Camp drawing

The Warham Camp drawing put a separate monument’s figure where it read as part of the fort, and the page gave one source for every register the drawing names

What happened.
The scheduled-monument note on the Warham Camp drawing said the fort is scheduled in two parts, one either side of the channel of the River Stiffkey. It then said “The second area shows next to nothing, 0.4 m by the same measure, and it is scheduled all the same.” The 0.4 m was measured on neither part of the fort. It was measured on the Iron Age ditched enclosure 270m north east of Warham Camp, which Historic England schedules as a monument of its own, so the note read as if the figure were one of the fort’s two parts. Beside the drawing the page also said “The free check reads the same registers for the spot you give it, through the national planning data platform.” The drawing names four registers. The free check reads two of them, scheduled monuments and SSSIs, through that platform. It reads the other two, open-access land and registered common land, from Natural England.
Cause.
The recipe measured each list entry on its own and kept each figure with the right one. The note then called the enclosure “the second area”, after saying the fort is scheduled in two parts. The fact-check before it shipped confirmed both figures, not which monument the words put them on. The page’s sentence was checked against the scheduled monument and SSSI lookups only, not against the lookups for the other two registers the drawing names.
Fix.
The note now names the enclosure as Historic England lists it before it gives the figure: “The Iron Age ditched enclosure 270m north east of Warham Camp is a scheduled monument of its own. The survey shows next to nothing of it, 0.4 m by the same measure, and it is scheduled all the same.” The page now says “For the spot you give it, the free check reads scheduled monuments and SSSIs through the national planning data platform. It reads open-access land and registered common land from Natural England.” The drawing, its figures and the free check have not changed.
Who was affected.
The Landcheck page and its drawing, from about 09:30 GMT on 11 November 2026 until the fix. Both monuments were hatched as scheduled throughout and the note said both are scheduled, so the advice on where not to detect did not change. No report or free check reads the note or the page’s sentence.
Since then.
A test requires the note to name the enclosure as a scheduled monument of its own before its figure, and forbids “second area” in it. Another pins the page’s sentence. Tests that do not go online check that the free check asks Natural England for open-access and common land, and the national planning data platform for scheduled monuments and SSSIs.

11 November 2026 · Extendcheck page, Bathwick drawing

The Bathwick drawing said more than the Environment Agency says of its vegetation model, and called the blank ground roof and wall without checking

What happened.
The tree crowns note on the Bathwick drawing said “The model is automated and not checked by hand, so a few objects are classed wrongly.” The Agency’s page for its Vegetation Object Model says production is fully automated, with no manual checking or editing other than visual checks, and that some objects not in Ordnance Survey’s mapping could be misclassified. It does not say objects are classed wrongly, and it does say the output is checked by eye. The caption also said “What still stands blank is roof and wall”, which nothing in the recipe checked.
Cause.
The note paraphrased the Agency’s page past what it says, and nothing held the note to the page. The caption’s words were an assumption about what raised ground with no crown on it must be.
Fix.
The note now says “The Agency says the model is made automatically, with no checking or editing by hand other than visual checks. It also says some objects not on Ordnance Survey’s maps could be classed wrongly.” The recipe now compares the blank ground with the buildings on Ordnance Survey’s OS OpenMap - Local. Of the blank ground, 94.1% lies on or within 2 m of a building, so the caption now says “What still stands blank is mostly buildings”, the note gives that share, and the sheet carries Ordnance Survey’s credit. The crowns, the sun sheets and their figures have not changed.
Who was affected.
The drawing on the Extendcheck page, from about 23:20 GMT on 10 November 2026 until the fix. The Extendcheck report and free check do not use the drawing: it is read only by the page.
Since then.
The build stops if the Agency’s page no longer gives the 2.5 m threshold or no longer says either thing the note repeats, or if half or less of the blank ground lies on or beside a building. Tests pin both sentences, the 94.1% share and the Ordnance Survey credit.

11 November 2026 · SearchCheck page, Kirkstall drawing

The Kirkstall drawing counted pieces of the flood map as floods on record, and left out a credit

What happened.
The flood note on the Kirkstall drawing said “Two of the Zone 2 outlines here are floods on record whose source the Agency gives as unknown.” The Environment Agency’s map service cuts its zones into pieces along the kilometre lines of the National Grid, and one piece can hold several separate outlines. The two were two such pieces, holding 12 separate outlines on the sheet, so the sentence counted pieces of the map, not outlines or floods. The drawing also named Kirkstall and Leeds from the Office for National Statistics’ ward boundaries without crediting them.
Cause.
The recipe that builds the drawing counted the Agency’s features and printed the count as a number of outlines. The fact-check before it shipped confirmed the count, not what one of the Agency’s features is. The ward boundaries were read only to name and check the place, and the credit for them was missed because they are not drawn.
Fix.
The note now gives the share of the sheet those outlines cover: “Outlines of floods on record whose source the Agency gives as unknown cover 0.2% of the sheet.” The grid line now says the place name comes from the Office for National Statistics’ Wards (December 2024) Boundaries UK BGC, with the two credit lines the Office for National Statistics asks for. The drawing itself and its other figures have not changed.
Who was affected.
The drawing on the SearchCheck page, from about 15:00 GMT on 11 November 2026 until the fix the same day. No report or free check used the note: SearchCheck reads the Agency’s flood map live around the address, not from the drawing.
Since then.
A test requires the note to give a share of the sheet and forbids a count of outlines, floods or pieces in it, and another requires the Office for National Statistics credit on the grid line.

11 November 2026 · Tideline coastal report

The Agency’s upper-end climate layer called a worst case, and recession counted from today instead of from 2020

What happened.
The paid coastal report and the coastal town pages labelled the Environment Agency’s 95th-percentile climate layer “By 2105, worst-case climate, no defences”. The Agency calls that layer the upper end of its climate allowances, says erosion can in extreme cases go beyond its zones, and uses “worst case” for something else. The town pages also said the distances were cumulative from today; the Agency counts them from 2020, its baseline year.
Cause.
The label was our own shorthand, written without the Agency’s scenario names beside it, and the baseline year was not checked against the Agency’s help page on what the erosion map shows.
Fix.
The row now reads “By 2105, upper-end climate, defences not maintained”, and the town pages say the distances are counted from 2020, the Agency’s baseline year. The figures were right and have not changed.
Who was affected.
No Tideline report has been sold. The 16 paid renders archived since 4 September 2026 are all of our own Norfolk sample postcode; the five free checks archived since 3 September 2026 do not show the erosion rows. The label was also on the 30 coastal town pages that carry the figure.
Since then.
A test pins the upper-end wording in the report, the town pages and the Happisburgh drawing, and forbids “worst-case climate” and “cumulative from today” in them.

26 October 2026 · HomeGround coal mining report

A single sale printed as a median, schools ranked in degree space, “Open right now” for stale applications, and a hazard a kilometre away called none

What happened.
The “median recent sale” used every sale since 1 January of the year ten years back rather than a ten-year window, and for an even count took the upper middle value, so one postcode’s figure was a single 2016 sale printed as a median. The sales chart was captioned as every recorded sale since 1995 even when the read was cut at its 200-row limit, and the England average beside it carried no date. The nearest two schools per phase were chosen in latitude–longitude degree space, where a degree of longitude is about 60% of a degree of latitude, so a nearer school could lose its place to a farther one. Any planning application the council had not marked decided was shown as “Open right now”, including ones lodged years earlier. When the BGS GeoSure 5 km cell held no rated hazard the report said no hazards were mapped for the area without looking at the neighbouring cells. The flood-zone line said “at this point” for a read of a ~100 m box around it, and for Scottish postcodes the SEPA extent, which includes surface water, sat under a “rivers & sea” label. A coal consultation layer that did not answer rendered as not in a consultation area, every point was called “the postcode point” whether it was a house, a postcode centre or a dropped pin, and the fallback summary could call the area “not a concern”.
Cause.
The window and median arithmetic were approximate; the schools query ranked with the PostGIS nearest-neighbour operator on unprojected geometry; PlanIt’s “Undecided” state was read as currently open; the GeoSure read stopped at the containing cell; the flood read is a bounding box by design but the copy did not say so; the Scottish flood label was the English one; the consultation lookup folded “did not answer” into “no”; the point description ignored how the point was located; the template summary carried a reassurance the data did not support.
Fix.
The median now covers exactly ten years to the day with a true middle value, and the count label says when the read was capped. Schools are over-fetched and ranked by geography distance. Undecided applications are labelled “Undecided (within the five-year window)”. A cell with no rated hazard reports the nearest rated cell within 2 km, with its hazard, severity and distance. The flood line says “mapped within ~100 m of this point”, Scottish extents are labelled as SEPA river, coastal and surface-water extents, a silent consultation layer is named as not answering, the point is described by how it was located (address point, postcode centre, dropped pin), the FAQ and free check say “the 5 km grid cell containing the point”, and the template summary no longer says “not a concern”.
Who was affected.
Two reports have been sold, both for England addresses on 2 September 2026, and both are corrected at their existing links. Re-running their lookups today: each listed a second school that was not the second nearest (one secondary about 500 m farther than the school omitted, one primary about 330 m farther); each showed applications lodged a year or more earlier as “Open right now” (four at one address, at least four at the other); one printed a “median” that was a single 2016 sale now outside the ten-year window, and the other’s median moves by £500 under the exact window; one sits in a cell with no rated hazard and was told none was mapped while a cell rated Significant for shrink–swell clay lies about 1 km away. Neither flood-zone answer changes at the point itself, neither postcode’s sales read was capped, and both were located at the postcode centre as the report said. The thirteen paid renders archived since 4 September 2026 are all of our own Welsh sample postcode; 23 free checks are archived since 3 September 2026. Updated 11 November 2026: the flood read this entry calls a bounding box by design was itself at fault and has been replaced, and one of these two reports did overstate the zone at its point. See the 11 November 2026 entry above.
Since then.
Tests pin the exact ten-year window and true median, the capped-count label, the geography ranking of schools, the undecided label, the nearest-rated-cell fallback, the flood and SEPA wording, the named silent layer and the point description; the new schools and GeoSure queries were run against the live database at both customers’ postcodes before this entry was written.

26 October 2026 · Tideline coastal report

An erosion rate printed as metres, a median over a cut-off sample, “inland” when the layer had not answered, and a flood zone read from a box

What happened.
For Scottish postcodes the Dynamic Coast historical shoreline-change rate, which is metres per year, was printed with the same “m” unit as the English NCERM projections, which are cumulative metres; the “typical” figure was the upper middle of at most 20 transects when the layer holds up to about 130 within 1.5 km; and an ArcGIS error returned inside a successful response was read as “no frontage”. In every nation a frontage lookup that failed, or that no erosion layer covers (Wales, Northern Ireland), rendered as inland: the free teaser said “inland”, and the paid report drew no distinction between “none within 1.5 km”, “not assessed” and “the lookup failed”. A SEPA flood layer that did not answer was silently treated as outside the flood extent. The flood-zone line said “at this point” for a read of a ~100 m box, and the sales and schools faults recorded today against the coal mining report applied to this report too.
Cause.
The erosion rows carried no unit or source; the Dynamic Coast read asked for 20 records and took the upper middle; the fetch helper returned ArcGIS error bodies as data; the coastal verdict was a single yes/no; the SEPA read dropped a layer that failed; the flood copy did not describe the box.
Fix.
Every erosion row now carries its unit and source and the page prints both. Dynamic Coast is read in full (up to 2,000 transects) with a true median and the transect count stated, and a cut answer says so. An error body counts as the lookup failing. The report carries a coastal status of frontage, none within 1.5 km, not assessed or unknown, and only “none within 1.5 km” may read as inland, on the teaser and the report. A silent SEPA layer is named and “outside” is never claimed over it. The flood line says “mapped within ~100 m of this point”. The coastal status is part of the render health check. Updated 11 November 2026: the box read behind the flood line was itself at fault and has been replaced by a test of each zone’s outline. See the 11 November 2026 entry above.
Who was affected.
No Tideline report has been sold. The nine paid renders archived since 4 September 2026 are all of our own Norfolk sample postcode; two free checks (one Leicestershire, one Norfolk, both England) are archived since 3 September 2026, and the report page has one recorded view. Any Scottish render would have carried the unit and median faults; any render during an erosion-layer outage would have read as inland.
Since then.
Tests pin the unit on every erosion row, the full-read median and transect count, the error-body handling, the four coastal states and which of them may read as inland, the SEPA named-silent-layer rule and the flood wording; an unknown coastal status now counts as a degraded render in the health alert.

26 October 2026 · HomeGround US address report

Lookups that did not answer rendered as clear, and figures labelled as something they were not

What happened.
When FEMA, EPA, USACE or USGS did not answer, the flood, shoreline, dam, wildfire, Superfund, earthquake, petroleum-release and National Risk Index sections rendered exactly as they do for a clean result, on the page, in the PDF and in the written summary. An address the geocoder could not place was told there was “No reliable satellite measurement at this exact location”, as if the ground had been checked. Where the nearest measured InSAR cell sat 150–250 m away, the assessed figure was labelled “worst within ~150 m” although that radius held nothing and the figure was that single cell. A point measured as rising or stable was still ranked “sinking faster than X% of measured locations”. At the few NOAA tide gauges with a negative relative trend (the land rising faster than the sea, as in south-east Alaska) the sea-level line read “rising −18 mm/yr”, and for an unmeasured address the section printed a “combined” figure that contained no ground motion at all. The PDF omitted the sampling and selection caveats the web page carried.
Cause.
Every paid lookup folded “the service did not answer” and “nothing here” into the same null; the not-located case reused the no-coverage copy; the fall-back-to-nearest query kept the radius label; the percentile was printed without checking the sign of the figure; the sea-level sentence assumed every trend was positive; the PDF kept its own copy of the section text.
Fix.
Each lookup now reports whether it answered. A failed one is listed in its section as “could not be checked this time (the source did not answer) — treat it as unchecked, not clear”, on the page, in the PDF and in the summary prompt, and counts as a degraded render in the health alert. An unlocated address says the address could not be located and nothing below was checked. A 150–250 m figure is labelled “Nearest measurement (N m away)” with a note. A non-sinking point gets no “faster than” rank. A negative gauge trend reads “falling N mm/yr (the land is rising faster than the sea)”, and an unmeasured address gets the gauge trend alone, labelled as such. The PDF shares its caveat and sampling text with the page.
Who was affected.
No US HomeGround report has been sold. The eight paid renders in the findings archive since 5 September 2026 are all of our own sample address, 350 5th Ave, New York. Twelve free checks are archived since 21 September 2026; any whose upstream lookups failed, or whose nearest cell was beyond 150 m, would have carried the fault without our being able to tell, because a failed lookup left no trace.
Since then.
Tests pin the failed-versus-empty distinction for every lookup, the unchecked wording on page, PDF and prompt, the not-located copy, the nearest-measurement label, the no-rank rule for non-sinking points, the signed sea-level sentence and the shared PDF caveats; a failed lookup now reaches the health telemetry and the degraded-render alert.

26 October 2026 · HomeGround US address report

Sites called “nearest” that were not, a “Significant” that implied likelihood, and benchmarks left unsaid

What happened.
The report described the dam and the Superfund site it printed as the nearest, when the code chose the most significant within 15 km and 5 km (poor-condition and high-hazard dams first; National Priorities List sites first). A dam’s “Significant” hazard class was shown without saying it describes the consequences of failure, not its likelihood. Petroleum-release sites were waved through as “most are historic and remediated” without their status being read. Beside a landfill that our own satellite sweep measured as settling, the text said “settlement is normal landfill physics” as if that closed the matter. Counties were named “Orleans County” for a Louisiana parish, “District of Columbia County”, and “X city County” for Virginia’s independent cities. The earthquake figure never said it is the USGS 2%-in-50-years benchmark on firm rock (Vs30 760 m/s), and the wildfire figure never said it is the worst of 9 points sampled across a ~1 km ring.
Cause.
The selection copy was written before the selection rule changed; the caveats lived in the author’s head rather than in the output; the county label came from a field that assumes every US county is called a county.
Fix.
The page, PDF, FAQ and homepage describe the dam and Superfund site as “the most significant within 15 km / within 5 km” with the priority order stated. The dam class carries “consequences, not likelihood”. Petroleum-release counts are grouped by the status read from the EPA layer instead of a stock reassurance. The landfill line reports the measured rate as a measurement, not a verdict. The county label uses the National Risk Index county-type field (parish, city, district). The seismic and wildfire figures carry their benchmark and sampling notes on every surface.
Who was affected.
No US HomeGround report has been sold; the eight paid renders archived since 5 September 2026 are of our own sample address. Every render since launch carried the “nearest” wording and the undisclosed benchmarks, so every reader of the sample report saw them.
Since then.
Tests pin the shared selection constants and caveats on page, PDF, FAQ and homepage, the status-grouped petroleum query, the county label for parishes and independent cities, and the benchmark and sampling notes.

26 October 2026 · Sinkcheck US

A “no incident” answer for points outside Florida, and glacial cover claimed for unclassified rock

What happened.
The Florida sinkhole incident register was read for any point inside a rectangle drawn around Florida, so an address in Mobile or Dothan, Alabama, or Valdosta, Georgia, was told there was no recorded incident within 1 km, from a register that does not cover those states. Where the USGS karst map gave a carbonate unit without classifying its cover, the verdict borrowed the deep-cover sentence, “buried under thick glacial deposits … collapses are rarer here”.
Cause.
A bounding box stood in for the state outline, and the unclassified-cover case fell through to the deep-cover copy.
Fix.
The Florida gate is now the geocoder’s state, falling back to a point-in-polygon test against the state outline; outside Florida the register is not read and the page says so. An unclassified cover says the map does not classify it, and the glacial sentence is used only when the map itself says deep cover.
Who was affected.
Of the eight free Sinkcheck US renders archived since 9 September 2026, none lay inside the rectangle but outside Florida and none carried the glacial wording, so no recorded render showed either fault. The same two lookups feed the US HomeGround sinkhole section, where every paid render is of our own sample address.
Since then.
Tests pin the out-of-state answer for points in Alabama and Georgia inside the old rectangle, the Tampa positive, and the unclassified-cover wording.

26 October 2026 · Fieldwork farm report

A soil-service outage reported as “no soil data”, a silent gap in ground motion, and a crop history a year short

What happened.
When the USDA Soil Data Access service was down (it has scheduled maintenance on Tuesday and Thursday evenings US Eastern and a nightly restart), the free checker and the paid report said there was no USDA soil survey data at this exact point, as if the survey had been read and found empty. Outside the metro areas our InSAR grid covers, the paid report’s ground-motion line was simply absent rather than saying the parcel is not measured. The crop history asked the Cropland Data Layer for the current year, which is never published yet, so the newest row was always missing and the ten-year table was a year short; years the service failed on were dropped without a trace. The upsell promised that the paid report “adds” a drainage class the free view already showed.
Cause.
A failed lookup and an empty one returned the same null; the ground-motion line rendered only when a cell existed; the year range started from today; the upsell copy predated the free view’s drainage line.
Fix.
A soil-service outage reads “did not answer just now … the soil could not be checked — try again shortly” on both surfaces. The ground-motion line is always rendered and says “Outside our current InSAR ground-motion coverage — not measured” where there is no cell. Crop history requests the ten years ending last year and prints “not returned by the CDL service” for any gap, with a whole-table notice when every year failed. The upsell no longer claims drainage as paid-only.
Who was affected.
No Fieldwork report has been sold. The seven paid renders archived since 4 September 2026 are of our own sample parcel in Iowa. Every render before today had the year-short crop table; any rendered during a Soil Data Access outage would have shown the false “no soil data” line.
Since then.
Tests pin the outage wording on checker and report, the always-rendered ground-motion line, the year range and gap rows, and the upsell copy.

26 October 2026 · GroundTruth feasibility check

Scene counts inflated by non-interferometric acquisitions, and a search outage shown as “Limited feasibility”

What happened.
The free feasibility check on the GroundTruth order page counted every Sentinel-1 scene over the point, including acquisitions in modes that cannot be used for interferometry, so the scene count and the feasible verdict could run higher than the stack that would actually be processed. When the ASF search service did not answer, the check showed “Limited feasibility”, which reads as a finding about the site rather than a lookup that failed.
Cause.
The ASF query had no beam-mode filter, and a failed search returned the same shape as an empty one.
Fix.
The search is restricted to interferometric wide-swath (IW) scenes, and the response carries a checked flag: an unanswered search shows “Could not check coverage” and never a feasibility verdict.
Who was affected.
The only GroundTruth order placed is our own test order of 26 October 2026, which was processed and then deleted. Four views of the order page are recorded in our traffic log since 25 July 2026.
Since then.
Tests pin the IW filter in the ASF query and the unchecked state for a failed search.

26 October 2026 · BNG exemption check

A verdict outside England, and mapped habitat labelled “Not exempt”

What happened.
The free biodiversity net gain check issued an exempt-or-not verdict for postcodes in Wales, Scotland and Northern Ireland, where mandatory BNG does not apply and the Priority Habitat Inventory we read does not reach. Where the inventory mapped priority habitat at the point, the badge said “Not exempt”, although presence on an indicative map is not the statutory test. The page said the habitat limb was tested “at your site location” when it is tested at a single point — the postcode centre, or the house point when a house number resolves — and the landing page FAQ answered “No” to “does it apply to an application I make today?”, which has been wrong since 6 August 2026.
Cause.
The verdict function had no idea which country the point was in; a finding was given the same status as a conclusion; the copy was written before the exemption came into force and never revisited.
Fix.
Outside England the check returns “Not applicable — England only” and runs neither limb. Mapped habitat is its own status, “Priority habitat mapped here — exemption may not apply”, with the ecologist caveat. The result page prints which point was tested and that it is not the red line, and the FAQ explains that the date the application is made decides which rules apply.
Who was affected.
No view of a BNG result page is recorded in our own traffic log since the check launched on 27 July 2026, and the check joined the findings archive only today, so earlier renders cannot be listed. Any result for a non-English postcode, or any “Not exempt” badge, would have carried the fault.
Since then.
Tests pin the not-applicable answer for each non-English country, the habitat-mapped status, the point-tested wording and the FAQ copy; BNG results are archived from today.

26 October 2026 · Sinkcheck UK

A “Low” for points the layers do not cover, and a rating described as a rock map

What happened.
A pin outside Great Britain, or a Northern Ireland address, could be rated Low, because the BGS GeoSure grid and the Mining Remediation Authority mine-entry layer were read beyond their coverage and their empty answers were treated as nothing mapped. The Low headline said “No soluble rock mapped in this 5 km area”, which overclaims: GeoSure is a hazard rating, and a hex rated below Moderate can still sit on chalk, limestone, gypsum or salt. The mine-entry section was still read for Northern Ireland, where the layer does not extend.
Cause.
Coverage was not checked before the Great Britain-only layers were queried, so an absence of data read as an absence of hazard; the headline was written as if the grid were a geological map.
Fix.
Points in Northern Ireland or with no Great Britain postcode within about 10 km now get an explicit “Not covered” answer, never Low, and neither layer is read for them. The Low headline is “The BGS 5 km grid does not rate soluble-rock hazard here”, with a line saying that is a rating, not a rock map.
Who was affected.
Of the Sinkcheck renders archived since launch on 9 September 2026, one Scottish point was rated Low under the old headline and one Northern Ireland point had the mine-entry layer read for it. No render outside Great Britain and Northern Ireland is archived. Sinkcheck feeds the UK HomeGround sinkhole section too; no sold HomeGround report was rendered after Sinkcheck launched.
Since then.
Tests pin the not-covered answer for Northern Ireland and for an unlocated pin, that the layers are not read outside coverage, and the new headline wording.

26 October 2026 · SearchCheck

A nil coal-mining result for Northern Ireland, and no word on what was located

What happened.
For a Northern Ireland address the coal-mining search item read as a clear result, when the Mining Remediation Authority reporting area covers Great Britain and the address was simply outside its coverage. The page also never said whether the verdicts were taken at a house point or at the postcode centre, which for a boundary-edge property can sit in a different flood zone or coal zone.
Cause.
The coal lookup was run outside the layer it reads, and an empty answer from it was presented as a nil result; the resolver knew what it had located but the page did not print it.
Fix.
A Northern Ireland address gets “not covered” for the coal search, with the reason, and the layer is not read for it. The result page says which point every verdict was taken at and, for a postcode centre, that a boundary-edge property can differ.
Who was affected.
Four SearchCheck result renders are recorded in our traffic log since launch on 26 July 2026. SearchCheck joined the findings archive only today, so we cannot tell whether any was a Northern Ireland address.
Since then.
Tests pin the Northern Ireland coal verdict and the located-as line; SearchCheck results are archived from today, and the daily coal-zone and flood-zone checks now name SearchCheck among the pages they feed.

26 October 2026 · Swimcheck bathing-water check

Overflow counts capped at 40, failed reads shown as zero, and “Nobody tests the water here”

What happened.
The point check counted storm overflows within 5 km from the first 40 rows of the query, so the “within 5 km”, “discharging now” and “last 7 days” figures were understated wherever more than 40 monitored overflows lie within 5 km. When the overflow archive could not be read, the four figures rendered as 0. A spot with no designated bathing water nearby was described as a place where “nobody tests the water”, although we hold Environment Agency results for England only and a spot in Wales, Scotland or Northern Ireland may be sampled by NRW, SEPA or DAERA.
Cause.
A row limit written for the list was applied before the counting; a failed read was caught as an empty list; the copy mistook the edge of our data for the edge of testing.
Fix.
The statistics count every overflow within 5 km and only the list is capped, with “Showing the nearest 12 of N” beside it. A failed read renders an explicit “unavailable” panel and no figure. The untested-spot line now says no EA (England) bathing water is designated here and names the other agencies.
Who was affected.
Point checks are made in the browser and were not archived until today, so we cannot count them. The understated figures would have affected dense urban and estuary catchments only; the zeros, any check made while the archive was unreachable.
Since then.
Tests pin the uncapped count, the unavailable state and the England-only wording; point checks are archived from today.

26 October 2026 · Consent Clock

“0 MW (0%, 0 schemes)” when the large-scheme query failed

What happened.
The sentence giving the share of past-deadline capacity consented under a Development Consent Order or an Electricity Act consent printed “0 MW (0%, across 0 schemes)” whenever the query behind it failed, as if no large scheme were past its deadline. The councils table did not say that its counts cover every technology while each link opens the authority’s solar page.
Cause.
A failed database read was caught and returned as zero; a label was left implicit.
Fix.
A failed query is carried as unknown and the page says the share could not be computed at this refresh, with no figure. The councils table carries a note on what it counts and what the links open.
Who was affected.
The page has been viewed 176 times in our traffic log since launch on 26 July 2026. We cannot tell on how many of those the query had failed; whenever it had, the DCO sentence read as zero.
Since then.
A test pins that a failed large-scheme query renders as unknown, not zero.

26 October 2026 · Published pages

Typed counts and figures that had drifted from the data

What happened.
llms.txt gave the paid price range as £19–£24 when the catalogue reaches £39, said Swimcheck scores “all 464” English bathing waters when the count comes from the live EA record, and quoted Consent Clock totals typed in from one quarterly extract. The US homepage said the UK had “five more” products, a typed figure: the catalogue lists ten UK products, eight beyond the two the sentence names. The Compare tool described a postcode outside the coalfield reporting area as “clear of the coalfield”, which the reporting-area boundary does not establish.
Cause.
Numbers typed into prose do not update when the catalogue or the data does; a boundary of coverage was described as a finding.
Fix.
The price range and the product count are derived from the catalogue in code and tested against it; llms.txt describes what the Consent Clock computes rather than one extract’s figures and drops the 464; Compare says “outside the coalfield reporting area”.
Who was affected.
Every reader of those pages while the copy stood. No address-level answer was wrong except the Compare label, which one recorded comparison could have shown.
Since then.
Tests read llms.txt and the homepage and fail when a typed count disagrees with the catalogue.

26 October 2026 · Wellcheck water-table report

Eight faults in how borehole readings were described

What happened.
A review of every report found eight ways the water-table page could mislead. The source protection zone line printed “(in a zone)” instead of the zone number. A borehole whose datum the Environment Agency lists as 0.00 — a placeholder, not a survey — was treated as ground level, so its water level could read as artesian. When the EA station lookup itself failed, the page said no borehole sat within 25 km rather than that nothing had been checked. The “rising or falling” trend compared the first and last readings of the year window, not the recent direction. The page and PDF said the three nearest boreholes were read, when the code picks the most useful three (surveyed datum, active status, nearest first). “Today is near its 12-month low” could be said of a reading months old, and a silent best borehole was not replaced by the next one. Northern Ireland was pointed to SEPA and the BGS ground-hazard layer was called UK-wide. A negative level at a datum-less borehole was phrased as above sea level.
Cause.
Each was a gap between what the code does and what the copy claims, or an unchecked assumption about the EA feed — a placeholder datum, or a failed lookup that returned nothing and was read as nothing there.
Fix.
The zone number is read from the EA layer; a zero datum is treated as unknown and no artesian head above 2 m is derived from one; a failed lookup renders as unchecked; the trend is the last 42 days of readings; a reading older than 45 days is never called today, and the next-ranked borehole is read when the best is silent; Northern Ireland names DfI Rivers and GSNI and the ground-hazard layer is described as Great Britain only; negative levels read “below sea level”; the page, PDF, FAQ and llms.txt describe the boreholes as the most useful nearby.
Who was affected.
No customer report: no Wellcheck report has been sold, and the only paid renderings in the findings archive since 3 September 2026 are of the public sample for a Petersfield postcode. Free checks since launch on 20 August 2026 could have shown the datum, lookup and trend faults; free views before 3 September were not recorded, so they cannot be ruled out.
Since then.
Sixteen new tests pin each behaviour, including a zero datum, a failed station lookup, a stale latest reading and the Northern Ireland copy, and a test reads the FAQ and llms.txt to check the public copy matches.

26 October 2026 · Redline renewable site screening

A favourable verdict with an unread flood zone, and six other faults

What happened.
When the flood-zone service returned nothing, the screening could still read as favourable, as if Zone 1 had been confirmed. The “mixed” headline gave a fixed reason that was often not the real one. The comparables section was titled as every scheme within 25 km when it shows the nearest twelve. The report never said whether it had located the site from a house-level address or only a postcode centre, which in open country can sit a kilometre from the plot. Outside England the notice said flood risk was unavailable, although the flood zone is read from SEPA and Natural Resources Wales, and any paid section whose data did not arrive — yield, topography, grid, comparables, flood — simply disappeared rather than saying so. On the council planning pages the FAQ “does this council approve…?” always began “Yes”, whatever the approval rate. A purchase receipt for a battery or wind screening would have linked to the solar version.
Cause.
The verdict chain treated a missing flood answer as a clear one; headlines, titles and notices were written for the common case and not re-checked against the data they describe; the checkout flow dropped the technology from its links.
Fix.
A favourable verdict now requires a read flood zone; the headline is composed from the actual causes; the comparables are described as the nearest twelve; the report states how the point was located and warns on a bare postcode; each missing section renders a line saying what failed; the non-England notice names the flood source and lists what is England-only; the council FAQ answers from the rate; receipts and QC links carry the technology.
Who was affected.
No customer report: no Redline screening has been sold, and the only paid renderings in the findings archive are of the public sample for a Norfolk postcode. Free screenings and the FAQ on every council planning page could have shown these faults since launch.
Since then.
Tests pin an unread flood zone blocking favourable, the headline causes, the located-as line on page and PDF, the fallback line for each missing section, the council FAQ lead word and the technology in receipt links.

26 October 2026 · Extendcheck extension report

Eleven faults in the paid constraints, ground, party-wall and sun sections

What happened.
“Outside the influence zone” and a foundation band were shown when no tree height had been measured. When OpenStreetMap buildings could not be fetched, roofs were counted as trees and garden. Listed-building and protected-tree rows said “within 15 m” and “covers this ground” for features up to 50 m away, and the listed-building effect was misdescribed. A shed or garage could be taken as the house without saying so. The sun map said “m² of garden” when it had used a 40 m box around the point. The party-wall 3 m and 6 m distances were measured from the existing house without saying that an extension moves the excavation closer. A roof 60 to 100 m away could be reported as the nearest tree. The neighbours’ decisions teaser counted listed-building consents, used the wrong denominator, did not sum, and silently stopped at the latest 50 applications. A main river 8 to 30 m from the point could be missed although the permit zone runs from the bank, not the centreline. The mine-entry row named the Coal Authority, now the Mining Remediation Authority. The free-tier verdict bodies claimed checks that had not been made.
Cause.
Sections written for the case where every lookup succeeds, and distances and counts described in words that did not follow the measured values.
Fix.
Each section now withholds a figure it cannot support and says why; row copy follows the measured distance; outbuildings are never taken as the house and any snap distance is disclosed; the sun caption names the area it used; party-wall copy says to re-measure from the planned footprint; the tree search is limited to the radius where buildings were masked; the neighbours count says what it counts and when it is truncated; main rivers are checked to 30 m; the verdict bodies claim only what was checked.
Who was affected.
No customer report: no Extendcheck report has been sold, and the only paid renderings in the findings archive are of the public sample for a Bath postcode. Free checks since launch on 10 September 2026 could have shown the teaser and verdict faults.
Since then.
The paid section builders are now pure functions with a test for each fault, including a missing building mask, an unmeasured canopy, a shed beside the point and a river 20 m away.

26 October 2026 · Landcheck detectorist land report

LIDAR hillshade lit from the wrong side, and six other faults

What happened.
The 1 m LIDAR hillshade was lit from the south-east with inverted relief, so ridges read as hollows and hollows as ridges. A missing terrain render was always blamed on no Environment Agency coverage, even when the terrain service had merely timed out, and squares outside coverage could render as flat grey ground. Sentinel-2 cropmark frames from 2022 onwards were drawn with flattened contrast because of a change in how the imagery is encoded, so drought-year and recent frames were not like for like. From January to May the “most recent summer” frame was silently absent, any promised satellite window that returned nothing was dropped without notice, and a failed heritage lookup was invisible in the PDF and on the free view. The £19 report was offered for spots in Wales and Scotland where it cannot be delivered. The cloud figure implied cloud over the square rather than the whole scene, and the landing page promised a verdict per register.
Cause.
An aspect formula that produced a compass bearing where a mathematical angle was expected; failures and absences handled as if they were negative results; a Sentinel-2 processing change not accounted for.
Fix.
The hillshade is lit from the north-west; a missing render says whether coverage or the service is the reason, with a retry; the 2022-onwards encoding offset is removed; the last complete summer is used and labelled in the early months; missing windows and failed lookups are named on the page and in the PDF; the report is not offered outside England; the cloud caption and landing copy are corrected.
Who was affected.
No customer report: no Landcheck report has been sold since launch on 13 August 2026. Free checks could have shown the heritage-lookup fault and the out-of-England offer.
Since then.
Tests pin the hillshade orientation on synthetic slopes, the terrain route’s 404 and 503 paths, the encoding offset by processing baseline, the summer fallback and the PDF’s missing-section lines.

26 October 2026 · Wellcheck water-table report

Boreholes counted as “within 25 km” that were not

What happened.
For a Market Harborough postcode the page said “19 monitoring boreholes within 25 km” and read the nearest of them as the local water table. Every one of those boreholes is 33 to 40 km away. The honest answer the page was built to give — no Environment Agency groundwater borehole within 25 km, so no measured level to show — never appeared.
Cause.
The Environment Agency hydrology API takes a distance parameter that is not a true radius: it returns stations well beyond it. Our code trusted the parameter and never measured the distance itself before counting.
Fix.
Every station is now measured from the searched point, and anything beyond the stated radius is dropped before the count, the ranking and the “also within range” list are built.
Who was affected.
No customer report: the only paid renderings on record are of the public sample report for a Petersfield postcode, whose nearest borehole is under 3 km away. The free check could have shown a wrong count anywhere the nearest monitored borehole is more than 25 km away, as it is around Market Harborough. The findings archive holds no such view since it began on 3 September 2026; free views before that date were not recorded, so we cannot rule them out.
Since then.
A test now feeds the report a station 33 km out and asserts that it is excluded and the count is zero.

2 September 2026 · HomeGround coal mining report

A report was computed for the wrong place

What happened.
A paid report for a Cheshire address was computed for a point about 100 km away in Leicestershire, so it said the property was outside the coalfield. It is in fact inside a Mining Remediation Authority Development Low Risk Area and the statutory coal mining consultation area.
Cause.
The address resolver appended ", UK" to house-level lookups, and the geocoder matched a business name elsewhere in the country instead of the address.
Fix.
The suffix was removed, and any house-level match more than 2 km from the postcode centroid is now rejected in favour of the postcode itself.
Who was affected.
One customer. We told them the same evening, the corrected report was live at their existing link, and the purchase was refunded in full.
Since then.
That postcode’s answer is one of the golden checks run every morning, and the distance guard is pinned by a test.

2 September 2026 · HomeGround PDF

The mine-entries map was missing from a PDF

What happened.
The PDF’s map of recorded mine entries showed the search circle and markers with no map underneath. The government map service it drew from had been switched off without notice, and every layer came back empty.
Cause.
A web map service formerly hosted for the Coal Authority was retired in the move to the Mining Remediation Authority. Our map code treated a dead upstream as an empty layer.
Fix.
The map was rebuilt on the Mining Remediation Authority’s current tile service with an OpenStreetMap street basemap underneath, and tile alignment is pinned by a test against a hand-verified shaft.
Who was affected.
The same customer’s report. A corrected PDF was available from the same link the same evening.
Since then.
The composited map is one of the golden checks run every morning, and a paid PDF that renders without its map raises an alert at once.

If yours is wrong

Email james@novarly.com with the link from your receipt. You will hear back from a person, and a wrong report is refunded before it is argued about.

How the measurements work · FAQ · Check an address