Tearline: the accounts your desk needs, verified and cited.
A company name goes in. A decision-grade tear sheet, capital structure, supporting model, and hyperlinked references come out, every figure one click from the filing it came from. We supply the data ourselves, from public UK registers, so no vendor licence is needed to use Tearline.
Illustrative · synthetic data · no figure below is real
Key financials · LTM
| Revenue | c. £420m |
| EBITDA (adj.) | c. £58m |
| EBITDA margin | 13.8% |
| Free cash flow | c. £6m |
| Net debt | c. £340m |
| Net leverage | c. 5.9x |
| Liquidity | c. £45m |
Capital structure · value break
What just happenedFigures were sourced and cross-checked, and the capital structure resolved to its fulcrum security (the Senior Unsecured Notes, where value breaks under a stress case). The full model, references, and the four artifacts sit behind this one-pager.
It learns what good looks like
Point it at the tear sheets your desk has already produced. It reads them as examples, not templates: your ordering, your defined terms, which figures you lead with.
Every example sheet becomes a fact the system can reason from. The typed page dissolves into the nodes it will connect.
The document becomes the system
The typed page is the surface. Underneath, a six-phase engine reads the same filings and market data an analyst would, cross-checks itself, and reconciles every figure to a source before a line reaches the page.
The tear sheet is a data structure first and a document second. The language layer may reference a fact, never originate or alter one, so fabrication has no path in.
Two engines, one instrument
One reads a name into a finished deliverable. The other reads a mandate into the names that fit it. The same graph of facts drives both.
What it learns, and what it does not
The shared record is the filings, ratings and register data everyone can see. What the platform learns from it is how desks work, never what any one desk knows.
Two engines, one instrument
A name in, a finished deliverable out.
The six-phase engine extracts, enriches from live sources, fact-checks itself, then builds the one-pager, capital structure, supporting model, and hyperlinked references.
Your mandate in, the names that fit it out.
Describe the box the fund invests in, and the origination engine surfaces companies that fit, ranked and evidence-backed, refreshed as the market moves. It also watches the names you already hold, so deterioration reaches you before the market prices it.
Decision-quality, by construction
Trust and checkability
Every figure is one click from its source, read as a card rather than reconstructed at view time. Closure checks confirm each number resolves to a document before release.
No AI language
A pure-code style linter bans empty intensifiers, hedging, process narration, and em-dashes. Prose that reads as machine-written throws rather than shipping.
Labelled substitution, never blanks
A resolution ladder: R1 reported verbatim, R2 a declared formula, R3 a bias-flagged proxy, R4 abstain with an attempt log. “Reported” is reserved for R1 alone.
Present the data, do not spin it
Prose is a gated exception: a typed sentence schema with no outlook or summary type, and directional verbs sign-checked. The FactManifest comes first, the sentence second.
The wall is the feature
What the platform learns is how desks work, never what any one desk knows.
We are building our own store of UK public-register data: Companies House accounts, charges, officers and PSCs, used under the Open Government Licence. We never resell premium vendor data, and no vendor licence is needed to use Tearline.
- No training on client data.
- Per-tenant isolation: on the roadmap, ahead of the first production tenant.
- Public Companies House filings only go to the Anthropic API to be read. Client documents are never sent to any model provider.
See it run on a name you know.
We are taking a small number of design partners from UK and European private-credit and special-situations desks. Bring a name, and we will bring the one-pager.
Request a design-partner slotOr see the working engine on a live Companies House name: the demo. Demo data is synthetic or illustrative unless a live source key is configured.