EstateDraft runs Claude Opus 4.8 in a gated environment, fenced off from your legal language. Opus helps organize your own facts into clear documents — but every statute, citation, and clause in your packet is fixed, source-locked text drawn from the official Texas statutes — the Estates, Property, Family, and Health & Safety Codes and the Texas Constitution, never generated by AI at request time. So your plan can’t cite a made-up statute or hallucinate the law the way generic AI tools can.
EstateDraft is Texas-only software, is not a law firm, and does not provide legal advice. AI assistance is limited to organizing your own inputs; the Texas corpus is loaded verbatim and SHA-256 byte-verified.
Statutes are pulled from a fixed, hashed corpus — not written by the model. A section that isn't source-locked simply can't be cited.
Operative clauses come from an attorney-reviewable clause library and a decision tree — the model phrases your facts, it never invents a clause.
Before anything is final, the Opposing Counsel Pass™ attacks the document for Texas-specific failure modes — and every finding waits for a human decision.
The engine, measured
Every release runs a deterministic evaluation suite in CI — zero model spend, reproducible from the repository: 466 checks across determinism, the Texas rules matrix, corpus resolution, retrieval selection, the audits, the agents, execution ceremonies, byte-exact statutory quotes, and the red-team library. Latest committed result: 466/466 passing (commit fa4b850). Methodology and the full table →
0 tolerated — a cite resolves against the hashed corpus or renders as a visible verify flag
100% catch required on the red-team corpus — instrument text from a chat surface kills the response
Byte-exact against the hash-pinned corpus (the Phase-8 verifier) — a drifted word fails the build
Same inputs, same documents, byte-stable — asserted on every drafting fixture, every release
These bars are held as a live accuracy SLA for released Texas work: the numbers above are recomputed on every release's eval run and checked against production audit rows as they accrue.
No. EstateDraft assembles operative legal text deterministically from a fixed clause library, and every statute is verified verbatim text retrieved from the Texas corpus. Claude Opus 4.8 only helps phrase your own facts into the document — it never authors the statute, citation, or operative clause.
Every Texas statute the engine relies on is loaded verbatim, SHA-256 byte-verified against the official source, and fixed. Documents draw from that corpus rather than generating law at request time, so a plan can't cite a made-up statute or hallucinate the law the way generic AI tools can.