A Neuralhue publication · Vol. I
EST. MAY 2026 · LONDON

Field notes.

What we're learning about Firm Memory, written down — in the voice we'd use at the table, not the voice we'd use on a deck.

№ 01 · 21 May 2026№ 02–03№ 04 · 22 August 2026
Latest · field note №04Framework · five questions
№ 04 · Latest

Five questions between the business case and the build.

Every AI programme starts with an objective and ends with a system. These are the five questions in between.

22 August 20264 minutes
Read the note →

Also published

№ 01 – 03
№ 03 · ArchiveIllustration · cross-matter evidence
№ 03 · Published

Your archive knows things no single matter file knows.

Why the hardest knowledge questions are not document questions — and what a system has to do when the answer lives across the archive.

13 August 202612 minutes
Read the note →
№ 02 · RefusalIllustration · three gap states
№ 02 · Published

The most useful answer a legal AI can give is “the firm hasn't done this.”

Why refusal and gap detection matter more than fluent answers over a firm's own archive.

25 June 202613 minutes
Read the note →
№ 01 · LeadIllustration · the corpus, stacked
№ 01 · Published

What we learned generating a 700-document synthetic legal corpus.

The trust-architecture argument, with the receipts.

21 May 202614 minutes
Read the note →

In preparation.

Notes drafted and scheduled. Windows shift only if the writing isn't ready — not if the calendar is.

Notes 05 – 07 · upcoming
№ 05
The gap between RAG demos and production legal knowledge systems.
Citation enforcement, gap detection, refusal discipline — the engineering substance behind a demo that doesn't lie.
StatusUpcoming
№ 06
30 queries, 29 passes, one honest fail.
What the evaluation pack revealed about the system — and what we shipped anyway.
StatusUpcoming
№ 07
What Firm Memory will not do.
The negative claims as a manifesto.
StatusUpcoming