Corrections log

A pre-broadcast cross-check on 2026-04-29 found the OG share card used Labour's 305k political pledge as the percentage denominator, while the homepage delivery-commitment card and the housing_trajectory indicator score the same numerator (146,880 annualised Q4 2025 completions) against the OBR 300k working assumption. The two denominators answer different questions, but on a single share card we prefer arithmetic coherence with the indicator the card visualises. Labour's 305k pledge remains surfaced on the homepage delivery-commitment card prose, where both denominators are explained side by side (live indicator 49% vs OBR 300k; annual NAD 73% vs Labour 305k). The numerator (147k annualised completions) is unchanged.

A pre-broadcast audit on 2026-04-29 found the editorial copy could attribute a pillar-level move to the headline when the dominant pillar moved against the headline. The fix changes the sentence subject so the magnitude can never be misread as the headline's movement; the headline's own deltas remain on the response unchanged. Methodology and pillar/headline values are unaffected.

The live EIA Open Data v2 path had been silently returning empty rows since 2026-04-17, so every five-minute cron tick was re-writing the editorial fixture rather than ingesting a fresh print. The latest-observation selector previously preferred the most-recently-ingested live row, so the stale-dated fixture row kept winning over fresher backfill data. Fixed by a two-tier selector that prefers fresher observed_at when backfill is more recent than a fixture-fall-through, plus an editorial fixture refresh to match. Live-path root cause (EIA returning empty rows for the EPCBRENT facet) is still under investigation; diagnostic logging widened in the same change.

The live EODHD path had been returning HTTP 404 for FTMC.LSE, so the adapter was silently falling through to the editorial fixture every cron tick. Same root cause and same fix as the Brent correction above: two-tier latest-observation selector now surfaces fresher backfill rows when a live adapter is silently failing through to an older fixture, plus an editorial fixture refresh. EODHD ticker-namespace investigation still open.

Methodology change to align the published Tightrope Score with ordinary reader intuition: a falling score now means conditions are worsening. Historical headline, pillar, and event-delta rows were converted by complementing score values and flipping score-delta signs. NULL and out-of-range rows were left untouched.

A snapshot-selector bug caused the API to serve a stale seed-era row that outranked the current fixture under MAX(observed_at). The selector now picks live observations by MAX(ingested_at) over rows whose payload_hash is not historical-backfill or seed; the homepage now displays the fixture-driven Q4 2025 completions × 4 / 300k = 49.0%. The 72.4% figure was an annual outturn ratio (net additional dwellings 221,400 / 305k) — a different measure from the indicator's methodology and never the value the fixture intended to publish. Both measures remain documented in the delivery commitment card alongside the indicator.

Q4 2025 large-site consents re-derived from the MHCLG release (major dwellings 900 + minor dwellings 6,300 = 7,200). The earlier 7,300 figure was a provisional rounding superseded by the published breakdown.