Client identity from initialize.clientInfo (reliable for Claude, Cursor, most clients) or User-Agent fallback. Unauthenticated crawler probes show as 401 auth rows.
Funnels & path exploration — real visitors only (bots + owner IP filtered)UMAMI DASHBOARD ↗
CONVERSION FUNNELS
PATH EXPLORATION — TOP JOURNEYS
SESSIONS
STEPS
JOURNEY
EXIT PAGE
Sessions with identical click paths grouped. Owner traffic and bots excluded.
VISITOR MAP — WHERE YOUR USERS ARE (UK FOCUS)
circle size = unique visitors loading…
LIVE TRAFFIC (UMAMI)
EMAIL
STATUS
CREATED
LAST LOGIN
LAST IP
VIEWS
SCREENERS
ACTIONS
Non-UK IP addresses that have logged in or registered. Flag for review.
IP ADDRESS
COUNTRY
EVENTS
LAST SEEN
ACCOUNTS
TIME
EVENT
EMAIL
IP
COUNTRY
UK?
Filter
Loading…
← Select an event on the left to open its device dossier
SITE PATH EXPLORATION — how visitors flow through the site
Loading…
SHARED FINGERPRINTS ⚠ — one device, multiple accounts or countries
Loading…
LIVE DATA HEALTH—
…
STALE SCORE CACHE (target 0)
…
NO SCORE, ACTIVE + FILED
…
CS-OVERDUE FLAGS ACTIVE
…
CS SWEEP PROGRESS
SCORE vs CREDIT-VERDICT MISMATCH CHECK
Nightly count at 03:35 UTC · baseline 11,564 (6 Oct 2026)
Companies with a good score (50+) shown as credit-restricted because equity could not be read:
…
Verdict: … · scan took …s (full population, no sampling)
WORST OFFENDERS (highest score first)
Company
Score
Source
Filed
Every known error in our company data, in plain English. Red = fix first. Amber = fix later. Green = leave alone. Register below frozen 11–18 Aug; live counters above.
5,492,571
COMPANIES IN DATASET
351,619
FIXED 17–18 AUG (R2+A2)
0
R2/A2 RESIDUAL NOW
~681,000
OVERLOOK (R4 benign)
Green — fixed and verified (all counts final, 11–18 Aug)
Fixed-assets total missing
What’s wrong: stored the pieces (buildings, equipment, investments) but never summed the “fixed assets” total the score uses.
Who it hit:734,140 companies ≈ 1 in every 7
Why it mattered: the score lost its “assets” leg, so companies looked riskier (or safer) than they were.
Fix:FIXED 11 AUG — re-parsed from filings on our disk · verified residual: 0
What’s wrong: money set aside for future liabilities was also written into long-term debt, inflating how much the company “owes”.
Who it hit:720,798 filings ≈ 264,000 companies
Why it mattered: inflated debt, pushing scores down and making healthy companies look over-leveraged.
Fix:FIXED 17–18 AUG — identity derivation · pilot 10/10 vs raw filings · rescored · verified residual: 0
See a real example
ZODIAC EVENTS LTD (13193226) — £5,468 provision was also recorded as long-term debt.
Negative equity shown as positive
What’s wrong: some filings carry numbers with the sign flipped, so a company that owes more than it owns looked like it had positive worth.
Who it hit: sign-flipped filings across the dataset
Why it mattered: hid the single worst red flag (negative equity) and inflated scores by ~34 points on average.
Fix:FIXED 11 AUG — sign-flip correction · verified residual: 0
See a real example
PLAINVIEW VENTURES LTD (09687391) — the raw CH filing itself is sign-flipped (+£301K); we corrected it and the page now shows the true −£301K.
Working-capital warning printed the wrong deficit
What’s wrong: when current assets were blank, the page’s “short-term obligations at risk” banner did the maths wrong.
Who it hit:461,458 companies ≈ 1 in every 12
Why it mattered: a visible warning told users a false number.
Fix:FIXED 17 AUG — display fix, no re-parsing
See a real example
AUSTIN ANDREW HOLDINGS LTD (10090304) — banner said “deficit of £1,449,235”; real stored deficit £1,448,229.
Blank “net assets” on abridged accounts
What’s wrong: abridged filers put equity under a tag we didn’t read, so net assets showed blank even though we held the number.
Who it hit:269,641 companies ≈ 1 in every 20
Why it mattered: showed a blank cell — looked bad, didn’t misstate a score.
Fix:FIXED 13 AUG — tag mapping · verified residual: 0
See a real example
SUMO MOVE LTD (12876080) — equity held but net assets showed blank.
Long-term creditors dropped
What’s wrong: a rarer filing style splits creditors “within a year” / “after a year” using a label we missed, so the after-a-year figure was dropped.
Who it hit:173,076 filings ≈ 87,000 companies
Why it mattered: hidden long-term debt made companies look safer than they were.
Fix:FIXED 17–18 AUG — restored via balance-sheet identity · pilot 10/10 vs raw filings · rescored · verified residual: 0
See a real example
Company 11680195 — showed £13,911 of long-term debt (a duplicated provision); raw filing proved £367,675. Score corrected 8.45.
Amber — open (deferred or needs diagnosis)
Equity total replaced by share capital — DEFERRED (mostly benign)
What’s wrong: when a filing lists equity in parts we sometimes kept only the first part (ordinary share capital) instead of the whole.
Who it hit:~14,000 companies ≈ 1 in 400 — a 300-company classification against raw filings found 98% of the flagged population is benign (share capital genuinely is their whole equity, often £1–£2)
Why it mattered: small amounts on few companies; net assets drives the funding leg of the score, so affected firms look slightly asset-poor.
Fix:OVERLOOK for now — fold into any future legacy re-parse (£0 when we get there)
See a real example
AUSTIN ANDREW HOLDINGS LTD (10090304) — showed net assets £611, true figure £4,801. (The case that started this audit.)
What’s wrong: on these filings the balance-sheet identity (fixed assets + current assets − short-term creditors − net assets − provisions) returns a negative figure — impossible for debt, so a DIFFERENT field in the row is wrong.
Who it hit:845,157 filings ≈ 457,000 companies
Why it mattered: unknown until diagnosed — deliberately NOT touched by the 17–18 Aug fix; they sit outside the corrected population.
Fix:needs its own diagnosis pass — DIAGNOSED 18 Aug: 240-filing raw check found 58% parser extraction errors, 17% genuinely unbalanced source filings, 21% abridged gaps — no algebraic patch exists; deferred to the re-parse queue (Quaid, 18 Aug). Evidence: migration/quarantine_phase2_results.json
The verdict
Fix the five RED items first — all £0, parseable from data already on our own disk, one afternoon of compute, no new spend. Negative equity moves scores DOWN, so it waits for explicit go. The two AMBER items are cosmetic or rare — later. GREEN stays empty until the battery classifies what’s truly unfixable. All counts final — identity battery complete.
Are we up to date with Companies House accounts data? Two tracks: iXBRL bulk (one zip per month → parse → load) and PDF filers (per-company nightly truth-check → Qwen-VL extraction — they never appear in the monthly zips). Live against CH’s download page + our DB.
Checking…
TRACK 1 — iXBRL BULK (monthly CH zips)
Loading…
TRACK 2 — PDF FILERS (nightly truth-check + Qwen-VL extraction)
—
PDFs IN EXTRACTION QUEUE
—
DETECTED LAST 7 DAYS
—
PDF FILINGS IN DB
Companies that file scanned PDFs never appear in the monthly zips. Our nightly truth-check re-asks CH whether each company has filed since we last looked; new PDF-only filings are extracted with Qwen-VL and scored within the nightly sweep. So “are we current?” covers PDFs too — typical lag is the truth-check rotation (~2–4 weeks) plus one night of extraction. No monthly action needed.