$ masli approach --four-rules
How the work is done.
Four rules that hold regardless of what is being built, and regardless of which of the three engagements you buy.
-
01 ACCEPTANCE
Acceptance is a number, agreed before the work starts.
A feature list can be delivered in full and still leave the original problem exactly where it was. So scope is written as a measurable acceptance condition: what has to be true at the end, how it gets measured, and who does the measuring.
Sometimes a result cannot be stated that way. That is worth discovering before the contract rather than at the review, and finding it out is part of what the consulting engagement is for.
-
02 INTEGRITY
The record is checked by machine, not by discipline.
Documentation kept by hand goes stale quietly: references break, an index stops matching the pages under it, a decision is reversed in one place and left standing in another. Documentation kept by a language model goes stale faster, because it is produced faster.
So integrity is enforced by a checker that runs over the whole corpus and fails the build on broken references, index drift, structural violations, and claims whose source no longer exists. Twenty-two rule classes at present. A check that runs is worth more than an intention to be careful.
-
03 METHOD
The method does not depend on the subject.
The same pattern — a written schema, an agent that maintains the corpus under it, and a checker that enforces the schema — runs across eight unrelated subject domains of our own. The largest carries 318 commits and 429 ingested sources since 16 May 2026.
The usual objection to a methodology is that it works for the person who invented it and for nobody else. Eight unrelated domains is the answer to that: the method holds when the subject changes completely, which is the property that makes it worth applying to yours.
-
04 HANDOVER
What is handed over runs without us.
The schema, the corpus and the checks are plain files in your own repository. No proprietary format, no hosted service of ours in the middle, no dependency on our being reachable next quarter.
This is the part most AI engagements skip, and it is the part that decides whether anything survives the handover. A system nobody can tell has broken is indistinguishable from a system that works, right up until it matters.