The system of record for first-party property claims.
Coverage, statutory clocks, estimate reconciliation, the money ledger and co-counsel access are reads against one underlying structure — so the answer is the same everywhere you ask it, and the file can tell you what it is missing.
One matter, four surfaces, numbers that reconcile across all of them.
These are not screenshots. They are the components, rendering one file’s data — which is why the reconciliation table adds up to the replacement cost in the money tab, and why changing a line item would move both.
The declarations page, parsed into objects
Coverage parts, endorsements and deductible basis become things the system can compute with. The wind and hail deductible here is one percent of the Coverage A limit — $4,120 — not one percent of the loss. Firms get that wrong in both directions, and it moves the demand by thousands.
Ruiz v. Lone Star Mutual · 2026-TX-0417
Coverage parts
HO-3| Part | Basis | Limit |
|---|---|---|
A — Dwelling Drives every percentage deductible on the policy | Stated limit | $412,000 |
B — Other structures Detached garage and fence are here, not in A | 10% of Coverage A | $41,200 |
C — Personal property Named peril under this form | 50% of Coverage A | $206,000 |
D — Loss of use Not yet triggered — dwelling remains habitable | 20% of Coverage A | $82,400 |
Deductibles
| Wind and hailapplies | 1% of Coverage A | $4,120 |
| All other perils | Flat | $2,500 |
Endorsements
Ask the file what it is missing.
Reviewing a file after a month should not mean reading a month of notes. The assistant reads one matter, answers only from it, and cites where each figure came from. Its most useful answers are about absences — and an absence has no document to summarise, which is why a system built around documents can never surface one.
I have read the Ruiz v. Lone Star Mutual file — the policy, the two estimates, the correspondence and the payment ledger. Ask me anything about it, including what is not in it.
A scripted demonstration on a fictional file. The figures are the same ones the ledger computes elsewhere on this page, which is why they reconcile.
What is actually built.
These figures are read from the production registry on every page load. They are not a marketing number maintained by hand, which is why they can go down as well as up.
The problem is not any one of your systems. It is the seams between them.
Every tool in a property practice is competent at its own step. Value leaks in the handoffs — where a fact gets retyped, a date gets assumed, or a number stops being connected to the document it came from.
Intake to coverage
A signed retainer tells you nothing about whether the loss is covered. The declarations page does — and in most firms it is read properly for the first time weeks after the file opens, by which point the deductible basis and the roof-surfacing schedule have already shaped the demand.
Coverage to clocks
Statutory deadlines do not all anchor to the date of loss. Florida runs its notice window from policy inception. Texas measures acceptance from the date the carrier received the items it asked for. Georgia carries two different limitation periods inside one homeowners policy. A calendar that anchors everything to one date is quietly wrong in most states.
Estimate to money
The carrier estimate arrives as a PDF. Someone retypes line items into a spreadsheet, and the difference between your scope and theirs — the actual subject of the case — ends up in a file nobody else can open.
Money to distribution
A settlement has to clear a mortgagee, a public adjuster fee, advanced case costs, a co-counsel split and a contingency before the client sees a dollar. That waterfall is almost always reconstructed by hand, under time pressure, at the moment it matters most.
Firm to co-counsel
Associating counsel in a state you are not licensed in is normal practice and how firms cover a region. Sharing the file usually means emailing attachments — which is both a security problem and the reason nobody can say afterward who saw what.
Everything to the record
When a fee dispute, a bar complaint or a bad-faith claim arrives, the question is what was known and when. Systems built around documents can show you the document. They cannot show you the state of the claim on a date two years ago.
Run every state. We name the ones whose statutes we have read.
Your files open in all 56 jurisdictions on day one — intake, coverage, estimate reconciliation, money, distribution, the record. What is earned state by state is the right to compute your deadlines for you, because an unverified date is not a smaller promise than a verified one. It is a malpractice event with our name on it.
150 rules across 15 states, each one read to its citation, scoped to the perils it governs and versioned by the date it took effect. Florida carries two: a policy written before SB 2-A and one written after get different pay-or-deny clocks, and the engine picks by the policy’s inception date, not by today’s. See every jurisdiction and its status →
Co-counsel without emailing the file.
A Texas firm takes a Louisiana hurricane loss and associates Louisiana counsel. Both work the same matter. Neither should see the other’s other clients. That is not a permissions screen — it is a property of the database.
Green is granted. Red is withheld and stays withheld — a grant can carry note-writing without also handing over the money ledger.
What the grant does
- ✓Scoped to one matter, with an effective window and a revocation that takes effect on the next query.
- ✓An ethical wall is a deny that beats every allow, including ownership.
- ✓Every read is logged with the party who made it, not just every write.
- ✓Fee splits are checked against the governing jurisdiction. Louisiana forbids a pure referral fee; Florida caps the referring share.
What the database refuses
- ✕Activating a fee-sharing grant with no client written consent on file. Model Rule 1.5(e)(2) makes the writing a precondition, so the schema does too — it raises an error, it does not warn.
- ✕Activating a grant before the conflicts check against the associating firm has run.
- ✕Deleting anything under a legal hold, whatever the retention schedule says.
Isolation is a database property, never an application promise.
Your diligence questionnaire will ask how tenant data is separated. The honest answer for most platforms is “the application always adds a filter.” That cannot be evidenced to an auditor and cannot survive one careless query.
- ✓Row-level security in Postgres, keyed to the verified request token. No application-side tenant filter exists, because there is nothing for one to protect.
- ✓One access predicate. Every policy calls the same function. One place to audit, one place to fix.
- ✓Hash-chained append-only audit log covering reads as well as writes. Tampering breaks the chain.
- ✓A build gate fails continuous integration if anyone ships a table without tenant scoping or without a policy.
- ✓No training on customer data, contractually and technically. ABA Formal Opinion 512 attaches an informed-consent burden to self-learning tools; this removes it.
- ✓SOC 2 Type II is the gating requirement, not an aspiration — the architecture above is most of the control set.
- ✓Massachusetts 201 CMR 17.00 and the New York SHIELD Act bind us directly the moment one claimant has an address in those states.
- ✓Legal hold beats retention, through a single enforced check. FRCP 37(e) is a function call here, not a policy document.
We are onboarding a small number of firms by hand.
Pilot firms get their jurisdictions verified first and shape what gets built next. Tell us where you practice and what you are running today.