Work that nobody can see may as well not exist

Five months of building, most of it invisible. This page is the whole of it, in one place — what is public and downloadable, and what is private and only described.

The public tier is on this site already, with download links and source. The private tier is named and explained but not linked, because those repositories are closed. That is deliberate: you can see the shape of what exists without me pretending it is available.

Every repository named on this page was checked against the live GitHub listing on 25 August 2026. If a name here does not resolve, that is a defect in this page.

Eight jurisdictions, free, MIT

One edition per jurisdiction, each mapped to that jurisdiction's own conduct rules. Described in full on the practice brains page.

Fourteen tools, fifteen repositories

Each does exactly one job and runs on your own machine. Described in full on the tools page.

Enterprise editions

Most public practice brains have a closed counterpart built for a firm rather than a solo practitioner: an in-tenant server instead of a single laptop, single sign-on, a self-lockdown mode, readiness work against ISO/IEC 27001 and 42001, and the legal pack a procurement team asks for. Seven jurisdictions — Hong Kong is public-only, and has no enterprise counterpart.

In-house and company builds

The same architecture aimed at a company's own legal function rather than a practice. The in-house brain is deliberately not a per-country product: the point of an in-house desk is that a matter arrives without first announcing which jurisdiction it belongs to.

The law-school family

Vidhi Likhit is Indian statute itself, turned into something you can teach and practise from: 11,715 Acts across the Union, the States and the Union Territories, and 90,715 sections, held as one searchable, citable corpus.

The point is what it replaces. A student ordinarily meets a section through a textbook's summary of it, and a practitioner meets it through whichever commentary is on the shelf. Here the text itself is in front of you — every Act, every section, searchable by what it says rather than by what someone wrote about it, and citable straight from the source.

A law school teaching from it sets exercises against real statutory text instead of extracts. Students search the whole corpus, trace a provision through its amendments, and cite the section itself. The institution gets its own seats over one shared corpus rather than every student assembling their own.

Three editions, each a separate repository over that one corpus:

Five engines, four of them with no language model in the decision path

These are working tools for a practitioner, not teaching material. Each does one narrow thing deterministically — the same facts give the same answer every time, and the answer shows its reasoning. Nothing here asks a model what it thinks.

All five are public and MIT licensed. pratipaksha is the one exception to the no-model rule — it is an AI agent by design, and its discipline is that it will not assert a citation it did not verify during the run.

Where the dissertation went next

My LL.M. dissertation at Queen's asked how much human oversight autonomous systems require before the law is satisfied, and proposed a formula for measuring it. Two problems surfaced once I tried to use it: it was a paper instrument, and it assumed an operator who was already paying attention.

chof-calc and chof-kernel — both public, both on this site — answer the first problem: the formula as running software, and an embedded governor that applies it in real time with a signed audit trail.

The second problem needed more than a calculator. A private architecture of eighteen repositories extends the same principle into full decision support — fusing what several sensing layers see into one picture, filtering it by the operator's role and current capacity, and refusing to drop a life-safety signal no matter how loaded that operator is. It is advisory throughout. It never decides; it tells a human what it can see, and how much of it they are actually in a position to hold.

It is not distributed, it is not for sale, and unlike every other tier on this page it is not named here. It is described because the research it came from is public, and because work nobody can see may as well not exist.

Where it stands

Checked against the live repository listing on 25 August 2026. This page describes the practice software; it is not an inventory of every repository I have published.

All of it was built alongside a practice — between boards, after court, and on the Sundays that were free. None of it has been sold. It exists because a solo practitioner has no back office, and building one turned out to be the only way to get one.

Start with a practice brain Or a single tool