Limitations & currency
Strength and boundary, stated together
The value of this evidence base depends on knowing which parts are live, which are frozen, which are proxies and which do not exist at all. Nothing here is presented as more recent or more certain than the source supports.
Currency spectrum
Where every dataset sits
Refreshed daily or on publication, with a short ingestion lag.
No time dimension, or refreshed per authority on an infrequent, non-simultaneous cycle.
A single point-in-time measurement; no trend analysis is possible.
Present but inconsistently populated, or scoped in ways that constrain querying.
Available in the catalogue but not exercised in the tested session, so depth is unverified.
Read the label, not the vibe
Confirmed data-quality issues
Nine traps found and verified in practice
Each of these was confirmed during analysis, not assumed. Serious issues could mislead a decision if unnoticed; technical issues are details worth carrying into every query.
An update stamp is not a content date
In appeals and several other datasets, the last-updated field records when the record was re-crawled, not when the event happened. The entire historic appeal archive is re-crawled nightly, so sorting by 'most recently updated' surfaced content from 2000 and 2005 at the top of the list. Content-specific date fields must be used to judge recency.
Placeholder and sentinel dates
Migrated records at some authorities carry default values instead of real dates: 1900-01-01, 1950-01-0X, and 2999, 2072 or 2099 in permitted development directions. Raw sorting surfaces these as if they were genuine history. Every date question must be range-filtered.
Static policy layers presented alongside live ones
Applications and appeals refresh daily. Policy designations, site allocations, permitted development directions and authority plan-making profiles refresh per authority and infrequently: one authority's designations were frozen at February 2024 and its profile at 2021/2024, missing a January 2026 plan adoption entirely.
No genuine vacancy indicator in rating data
There is no vacant or occupied field. A blank occupier name is inconsistently captured on large and composite buildings and is not a vacancy signal. An earlier suggestion that it could serve as one was explicitly withdrawn.
Geographic filters that fail silently
Consultation records hold no geometry or authority code field, so area and location filtering does not work at all and returns zero results rather than an error. Scoping works only by organisation name or an exact authority prefix.
References are unique only within an authority
The same application reference exists in other councils attached to unrelated buildings. During this work, a long heritage objection appeared under a matching reference but belonged to a different authority entirely, and was excluded rather than misattributed.
Incomplete area typology coverage
One of the area classifications is not populated for every Output Area. The example area queried returned nothing at all; neighbouring areas returned meaningful values.
Exact-match-only postcode fields
In several datasets, including energy certificates, the postcode field accepts only exact or list matching. Partial postcode searches fail, so location-based search is preferable.
Radius searches cross authority boundaries
A location search on policy designations returned layers registered by a neighbouring authority, including protected vista corridors. Without an explicit authority filter, a neighbouring borough's policy would be reported as governing the site.
Out of scope
What was sought and is not available
These were specifically pursued during analysis and could not be obtained. Knowing them in advance prevents questions being framed around evidence that cannot carry them.
Methodology notes
Six disciplines that protect the answer
Drawn from what actually went wrong and was corrected during the work.
Range-filter every date question
When looking for an oldest or newest record, always constrain to a realistic range. Raw sorting surfaces placeholder and corrupted dates: 1900, 1950, 2072, 2099, 2999: as if they were real.
Ask what each date field measures
A last-updated stamp records when a record was re-crawled. Content and event date fields record when something actually happened. Judge currency on the second, never the first.
Check for a geography field before filtering by area
Some datasets hold no geometry or authority code. Area and location filters on those datasets fail silently and return zero results rather than an error.
Treat references as authority-scoped
Application reference numbers are unique only within an authority. The same reference exists in other councils, attached to entirely unrelated buildings. Always verify the authority prefix.
Demand named evidence for attribution
An address match is a lead. An applicant field, a cover letter naming the client, or a registered proprietor is evidence.
Re-filter radius searches by authority
Location searches return designations registered by neighbouring authorities. Without an explicit authority filter, a neighbouring borough's policy layer will be reported as if it governed the site.