UK installed-base intelligence
Public data.
Cracked open.
The plant your firm services is already documented — in grid registers, environmental permits, compliance reports, planning files, contract notices and company filings. Nobody joins it up. We do, one sector at a time.
You get a register: who operates the asset, exactly what it is, when it next needs work, and what that work is worth — with the source document cited on every single field.
THREE REGISTERS LIVE · NATIONAL COVERAGE · REFRESHED WEEKLY
LICENSED TO ONE FIRM PER TERRITORY, PER REGISTER
- Operator
- ███████ Chilled Foods Ltd[1]
- Site
- Melmerby, HG4 ███[2]
- Capacity
- 812 kWp · roof-mount[2,4]
- Connected
- Aug 2013 · FiT cohort[3]
- Inverter window
- OVERDUE — yr 13[3]
- Incumbent O&M
- None located[5]
↓ this is the summary. The full record carries 34 more fields.
What sits behind a territory sample
Figures from the July 2026 build. Each register states its own universe, coverage and known gaps on the sample cover — including what we couldn't place.
The difference
A list tells you who exists.
A register tells you what to do on Monday.
Every service firm has bought a list. Names, addresses, maybe a phone number, sold to nine of your competitors, no idea where any of it came from. That is not what this is.
The list you bought last year
- A company name and a postcode. No asset detail at all.
- No provenance — you cannot check a single field, so you cannot trust any of it.
- No timing. Everyone on it is equally cold, forever.
- Scraped once, decaying since. Nobody tells you when it went stale.
- Sold to whoever pays. Your competitor is calling the same name this morning.
- Duplicates, dissolved companies, the wrong trade entirely.
A Cracking Data register
- Operator, group ownership, site, asset specification, capacity, vintage.
- Every field carries a citation and an evidence tier. Check our working yourself.
- A dated service clock: what is due, when, and on what authority.
- Refreshed weekly. Changes land in your territory feed, flagged for action.
- One licensed firm per territory, per register. Contractual, not a promise.
- Filtered at ingest and again at build — wrong-trade records never reach you.
One record, opened up
This is what you're buying.
Below is a single record from the Commercial Solar Register, shown in full. Values are illustrative — the structure, the evidence tiers, the clock logic and the scoring are exactly what ships. Every register uses the same anatomy; only the sources and the clock change.
Official stated in a statutory source Matched joined across two sources, confidence stated Inferred derived from rules, marked as such Modelled estimated, bounds shown
01 · Operator & site
- Operator
- Off███████ Chilled Foods LtdCompanies House 0█████7 · active · accounts current
- Ownership
- OffAcquired by ██████ GroupPSC change filed Nov 2025 — supplier review window
- Trade
- OffSIC 10130 · meat processingCold store — high self-consumption profile
- Site
- OffMelmerby, HG4 ███Street address held in full on licensed records
- From your base
- Mod21 mi · 34 minSorted by drive distance from your depot postcode
02 · Asset
- Capacity
- Off812 kWpDNO Embedded Capacity Register, Jun 2026 refresh
- Declared cap.
- Mat780 kWp · Δ −3.9%High-confidence match, not confirmed — district + capacity ±10%
- Mounting
- OffRoof · 3 buildingsPlanning ref █████████, decided Feb 2013
- Energised
- OffAug 2013FiT accreditation date
- Inverter class
- InfCentral · 2 unitsRule-derived from vintage + capacity band. Verify on site.
03 · Service clock
- Asset age
- Off12 yr 11 mo
- Inverter window
- InfOVERDUE · yr 13Signal: PREDICTED — design life exceeded, no replacement filing found
- Missing evidence
- OffNo replacement on recordA post-2023 swap would leave no public trace. Ask on the call.
- Tariff term ends
- OffAug 203320-year term · post-Aug-2012 accreditation
- Incumbent O&M
- OffNone locatedSearched provider references + award notices, Jul 2026
04 · Commercial
- Annual output
- Mod731,000 kWh/yr (est.)Regional irradiance · 0.4%/yr degradation applied
- Value at risk
- Mod£71,300 – £128,000/yrLow/base/high stack — self-consumption 25–65%, import 22–30p
- Downtime
- Mod£264/day (est.)The number that ends the "we'll think about it" call
- O&M contract band
- Mod£4,100 – £8,700/yr (est.)
- Replacement job
- ModFive figuresTwo central units at this capacity
PRIORITY FALLS IF — the inverters were replaced privately after 2023 (no public filing would show it), or the site is under a group-level framework agreement following the November ownership change. Both are first-call questions, and the record tells you to ask them.
Illustrative composite. Field structure, evidence tiers, clock logic and scoring are as shipped.
Each field is tagged with how we know it. A modelled revenue figure never sits on the page looking like a measured one — and you can tell the difference at a glance, in a meeting, out loud.
"Missing evidence" is a field, not an omission. Where the public record simply cannot answer a question, the record says so and hands you the question to ask instead.
Trigger, value, entity quality and contact quality are scored separately and shown. You can disagree with our weighting — because you can see it.
Every high-priority record carries its own disqualifier. A register that only argues for itself is a brochure.
How it's built
Every source is public.
The join is the product.
None of the inputs are secret, and we don't pretend otherwise. You could assemble one record yourself in about an hour. The register does it for every asset in your territory, cross-referenced, graded and kept current — which is a different thing entirely.
Walk it all
Statutory registers are ingested end to end, not sampled. When a national register turned out to be 5,189 entries rather than the 4,000 our first pass assumed, 23% of the market was invisible. Completeness is a build assertion now.
Go a layer deeper
Behind a register entry usually sits a shelf of documents the register's own search will never show you: compliance reports, assessments, plant schedules, acoustic surveys. That is where the model numbers and the dates live. We open them.
Read and verify
Documents are machine-read at scale, then every extracted date and figure must appear verbatim in the source before it is allowed onto a record. Anything that fails validation is flagged for a human, not published.
Resolve the entity
Site to operator, operator to registered company, company to parent, parent to the people who sign. Address deduplication, dissolved-company checks, wrong-trade filters at ingest and again at build.
Start the clock
An asset with no date is a name. We attach a service clock from statutory cycles, accreditation vintages and contract expiries — then grade it, gate it, and say plainly which rung of evidence it stands on.
Grid connection registers, environmental permits, accreditation records, contract-award publications. The anchor for every register we build.
Ingested in full · refreshed monthlyPer-asset document shelves behind those registers — compliance reports, monitoring returns, permit conditions. Most of the value nobody extracts is here.
286 reports parsed · 133 assets with verified evidenceAcoustic reports name generator models. Plant schedules name boilers. Solar applications state kWp and mounting. It is all filed in public, buried in PDFs.
Earliest signal an asset exists — before it's builtAsset-finance charges name borrower, lender and often the equipment. Ownership changes flag a supplier-review moment nine months before anyone rings.
Entity-resolved to registered numbersEvidence discipline
The hard part isn't finding data.
It's knowing what it's worth.
Anyone can produce a spreadsheet with a "due date" column. The question that decides whether you waste a morning is: due according to whom? Every register answers it on every field, on two axes.
Axis one — evidence tier: where the value came from
Stated in a statutory source. Capacity on the grid register. Permit status on the regulator's file. Directors at Companies House. Quotable in a meeting without a caveat.
Joined across two independent sources, with the join rule and its tolerance printed. A high-confidence match is never dressed up as a confirmation — and where the two sources disagree, we print the delta.
Derived from a published rule. A statutory testing cycle counted forward from a registration date. An inverter class from vintage and capacity band. The rule is named on the record.
Estimated, with bounds. Yield, revenue, contract values. Always three-significant-figure, always labelled (est.), always low/base/high. Never a suspiciously precise £7,926,117.
Axis two — signal level: how firm the timing is
A deadline stated in a document we hold. The regulator wrote the next-due date; we quote it, with the report reference. This is the only rung that produces the words "due" without a qualifier.
Last known event plus a published cycle. The evidence for the last event is real and cited; the forward date is arithmetic. Records say so, in those words.
A credible hypothesis worth one phone call. Registration date only, no compliance shelf. Graded lower on purpose, and it can't be promoted without evidence — the gates are enforced in the build, not by judgement.
Quality control
Checked before it reaches you.
Every build runs a fixed battery of automated checks before a single page is rendered. These are real failures we have caught and now test for permanently.
Coordinates come from the register's own address and grid-reference fields. Regional labels are never geocoded to a centroid — that is how a plant 40 miles outside your patch ends up in your territory pack. 100% of permitted sites in the latest build carry a real address.
Records with implausible dates, impossible capacities or broken joins go to a repair queue and are never sampled into a pack. The count is printed on your cover sheet, so you know exactly what was held back.
A regulatory threshold of "50MW or more" is not a plant rated at 50MW. Confusing the two makes a register look authoritative and be wrong. It is a build-breaking assertion here.
"Maintenance" contracts that are actually software support. Medical scanners inside an industrial keyword. Grounds, fleet and catering inside an estates feed. Filtered at ingest, filtered again at build, regression-tested after every miss.
After a complete walk of a source, anything no longer present is deleted rather than left to rot. One recent pass removed 717 stale rows. Registers that only ever grow are lying to you.
Each register has its own language and its own units. Solar vocabulary cannot appear on a combustion record; estate terms cannot appear on either. Cross-contamination is how a pack stops sounding like it was written by someone who knows the trade.
Registers
Three live. One sector at a time.
Each register is standalone: its own statutory sources, its own service clock, its own trade language, its own territory licences. Depth beats breadth — a register that covers everything covers nothing usefully.
Commercial
Solar
Grid-connected arrays from 50kWp up — operator, capacity, connection vintage, tariff term, inverter-replacement window, and whether anyone holds the O&M.
Open the register → DIV. 02 OpenPermitted
Combustion
Permitted generators, boilers and standby plant — thermal rating, permit class, statutory emissions-test cycle, live compliance evidence and emission-limit deadlines.
Open the register → DIV. 03 OpenPublic
Estate
Plant and maintenance contracts across councils, schools, housing providers and NHS trusts — incumbent, value, expiry, recompete likelihood and live tender windows.
Open the register →Further registers are built to order where a sector has a statutory paper trail and a service trade that needs it. Tell us the trade.
On the registers this week
A feed, not a file.
Registers move. New connections appear, permits change status, compliance reports are published, contract notices land, owners change. Subscribers get their territory's changes weekly — each one flagged with what to do about it.
DIV 02Compliance report published against a 12MWth permitted site — regulator-stated next test date extracted and dated. Record promoted to confirmed.
DIV 03Live tender opened, closes in 26 days — building-services term contract, four-year value band. Bid-window override applied.
DIV 01Monthly grid-register refresh ingested: new connections added, 14 arrays entered the year-12 inverter window this month.
DIV 01Ownership change filed against an operator holding two registered arrays — supplier-review signal raised on both records.
DIV 02717 delisted rows pruned after a complete source walk. Territory counts corrected downward, and we said so.
Territory licence
One firm per patch.
That's the whole point.
Intelligence sold to everyone is worth nothing to anyone. One licensed firm per territory, per register — held contractually. When a territory goes, it's closed, and we tell the next enquirer it's closed.
Your territory, your register, nobody else's. Renewal rights sit with you for as long as the licence runs.
The register is national. Your records arrive sorted by drive distance from your depot postcode, or filtered to the exact areas you serve.
Dispute a record and it's corrected within one working day, with the correction and its reason logged on the record's evidence trail.
Take the middle of the three registers. A single emissions-monitoring visit is a four-figure job. An annual service position on a permitted site is four to five figures a year, and it renews. On the solar side, one commercial O&M contract runs into the thousands per year on a multi-year term, and an inverter replacement on the same array is a five-figure job on its own.
One won contract > twelve months of the licence. Everything after that is the actual return.
Who builds this
A small team that takes data seriously.
Cracking Data is deliberately small: a founder who sells and specifies, and a data science and engineering team who build, validate and maintain the registers. No sales floor, no resellers, no white-labelling.
Builds the registers and runs every territory conversation personally. Specifies which sources go into a sector, where the evidence line falls, and what a register is not allowed to claim. If you take a licence, you deal with him.
The data team
A small group of data scientists and data engineers, working across all three registers rather than one each. What they're responsible for:
- Ingest & entity resolution — walking each source completely, resolving sites to operators to registered companies to groups.
- Document extraction — reading the report and permit layer at scale, with verbatim-quote validation on every date and figure that reaches a record.
- Clock & grading logic — statutory cycles, evidence gates, priority scoring, and the disqualifier on every high-priority record.
- Quality assurance — the automated check battery that has to pass before a build ships, and the repair queue for everything that doesn't.
Specialist trade input is bought in per sector to calibrate cost and cycle assumptions. Where a figure hasn't been calibrated yet, the register says so instead of guessing confidently.
Straight answers
The questions you're already asking.
Where does the data actually come from?
Public sources, all of them — statutory registers, regulator document shelves, planning portals, public contract notices, Companies House. We name them on every record, because a citation you can't check is just a claim.
What isn't public is the join: the extraction, the entity resolution, the validation rules, the clock logic and the grading. That's the part we've spent the time on, and it's the part you're paying for.
How accurate is it — and what happens when it's wrong?
"How accurate" is checkable rather than claimed: every field carries its source and its evidence tier, so you can audit any record against its documents yourself. Where sources conflict, we print both and the delta. Where we've inferred, the record names the rule.
When a record is wrong, tell us: it's corrected within one working day and the correction, with its reason, is logged on the record's evidence trail. We'd rather ship a smaller register that survives audit than a bigger one that doesn't.
Is this just scraped data with a nice wrapper?
No, and the difference shows up on the second phone call. A scrape gives you the surface of a register — the row you'd see if you searched it yourself. The value sits a layer deeper, in the documents attached to each entry: the compliance report with the regulator's own next-due date, the planning file with the plant schedule, the notice attachment with the asset list.
Those documents have to be located, retrieved, read, validated against their own text, and tied back to a resolved company. That's the work.
How much of this is done by a machine?
The reading is automated — thousands of documents is not a human job. The trust rules are not. Any date or figure extracted from a document must appear verbatim in that document before it's allowed onto a record; anything that fails is flagged for review rather than published, and a machine-read value can never be promoted to "confirmed" on its own.
The result is a register where the automation makes it possible and the constraints make it usable.
What does a licence cost?
Territory licences start at £499 a month on a six-month minimum term. The price for your patch depends on its size — how many assets sit on the register and how many live service windows are in it — and is quoted alongside your sample, so you see what you're buying before you're asked for anything.
National and multi-register licences by enquiry.
What about GDPR?
The registers hold site- and company-level data: operators, addresses, assets, ratings, dates and contract information drawn from public statutory sources. Where a published notice names a procurement mailbox, that's carried as published and scored as a low-quality contact — it never drives a call angle.
Who to ring inside the operator is your craft. Which operators to ring, and when, and with what opening line, is ours.
What's in the free sample, exactly?
A territory pack for your patch within two working days. One record fully open — every field, every citation, check our working against the sources yourself — plus the surrounding records with operators sealed, and a cover sheet stating your territory's universe, the qualified count, what we're monitoring, and what's sitting in the repair queue.
No call required to receive it. If the pack doesn't stand up, don't buy the licence.
Your territory is either taken or it isn't.
Find out — free sampleFree territory sample
See your patch first.
Tell us the register and the postcode you work from. We'll pull your territory pack by hand and send it within two working days — one record fully open, the rest sealed, and an honest count of what's in your patch. Licences go to one firm per territory: availability is first come, first served.