Working Theory · Rachel McBride
Building Mappa, and what its limits taught me about the Verifiable Core.
I built an instrument called Mappa. It's a living map of a firm: the picture a good leader carries in their head, made shared and kept current. It organises the firm around what it's actually working on, who it knows, where it sits in its market and how it really holds together, not around its org chart. A raw, human-owned record at the core. Intelligence as a layer over it. A human owning both ends. Then I did the thing most case studies skip: I audited it honestly. The interesting part wasn't where it worked. It was where it stopped.
"Intelligence at the centre" gets heard as: let the AI generate everything, then build the firm on what it produces. The serious version refuses that. What goes at the centre is the raw, human-owned record; the intelligence sits on top, never underneath. I set that principle out separately, in The Verifiable Core. Mappa was built to test it.
Most write-ups of a tool show you where it shines. I'm going to show you where mine stops, because every place it breaks names a discipline the principle demands. The weaknesses are the proof.
Where it stops: the intelligence layer can suggest things the data hasn't proven yet. So Mappa marks every machine-proposed value as visibly distinct from a verified one, and a human has to promote it before it counts as fact. That seam between what's known and what's guessed isn't an unfinished edge. It is the gate working. Ship nothing as fact that hasn't passed it.
Where it stops: only a few people can decide what's true in the core, and that looks like a cap on growth. It isn't. Curation is judgement-arbitration, not data entry. An uncurated core scales infinitely and is worthless. The bottleneck is the feature, not the failure.
Where it stops: the map covers the relational layer of a firm. Every firm is relational at its edges (customers, team, suppliers), so that part is universal. But it is not the whole of any firm: a maker still has a production core the map can't see. The map doesn't generalise. The method does, raw core, intelligence on top, a human at both ends, applied one substrate at a time. Apply the method; don't copy the map.
Where it stops: any instrument quietly pulls a firm toward the things it can measure. The counter is the human who owns the outer end, plus a rule I build into the instrument itself: every map states, on its face, what it can't see. Without that, the tool slowly colonises the judgement it was meant to serve.
Where it stops: how do you prove a record is current, sourced, unaltered, and fit to act on? That's the open problem, and I won't pretend it's solved. Nothing ships as "act on this now" without carried, checkable reasoning a human can inspect. Naming the frontier honestly is part of the work, not a gap in it.
These five aren't really about relationship maps. They're what any firm has to do to put intelligence at its centre without it rotting. The same shape shows up one level up, in how a single person holds their own practice, and one level down, in the data itself.
The test of whether you've built an intelligence-centred firm, or just bolted AI onto the one you had, isn't what your tool can show you. It's whether you can say exactly where it stops, and whether a human still owns both ends of that. Can you say where yours stops?
The principle behind this: The Verifiable Core. The instrument itself: mappa.co.nz.
v1.1 · Added a plain "what Mappa is" sentence up front (a living map of a firm) so a cold reader is oriented before the five disciplines; "live endeavour" reworded to "what it's actually working on." v1.0 · The Verifiable Core in practice, derived from Mappa's honest limits (five disciplines). The meaning is public; the relationship data, lens logic and operator specifics stay gated.