Good design
goes unnoticed
Most of what goes wrong with AI work goes wrong quietly. A number appears in a report and nobody can say where it came from. A summary reads well and is subtly untrue. A model is confidently wrong on the one occasion it matters.
The discipline underneath everything here is the opposite of confident: every claim carries its source, evidence stays separate from argument, and a system that cannot stand a statement up is built to withhold it rather than publish it.
Provenance travels with the datum
Every figure keeps a link back to where it came from, through every stage of processing. Not a bibliography at the end, provenance attached to the individual number, so any claim can be followed back to a record.
ProofThe Monitor · Rebeccanomics
Checks at both ends
The subject is resolved against a public register before a model is asked anything, and every source the model cites is verified to be real and live afterwards. Those two checks are what let an output be argued with rather than simply read.
ProofSpotlites · The Monitor
The system can abstain
A mechanical tripwire fails any figure absent from the source data; a second pass reads the finished piece back. If integrity cannot be met, nothing publishes. An automated daily that always publishes has no way to tell you the morning it was unsure.
ProofThe Monitor
How an engagement starts
No forms, no discovery phase you pay for twice. Four steps, and you can stop after any of them.
A sentence
Describe the problem in a sentence or two. You get a real reply from a person, usually with a first read on the problem, sometimes with the observation that you don't need us.
A conversation
Half an hour, no charge, no deck. The purpose is to establish whether the thing you want is the thing you need, and whether it is worth doing at all.
A defined first piece
Scoped, fixed and small enough to finish in weeks. A diagnostic, a prototype, a first publication. You end it holding something usable whether or not there is a second piece.
Then, if it's working
Longer engagements are built out of pieces that already landed. Nothing is committed to on the strength of a proposal.
The person you talk to is the person who builds it. W'A.I is deliberately small. That is a real constraint, there is a limit to how much runs at once, and it is the reason nothing is handed down to someone junior after the pitch.
Thirty years of it sits behind the work: brand and campaigns for FTSE 100 firms and regulators, digital and data for trade finance, and now intelligence systems that run unattended. The work wall is the whole record, including the parts that were delivered under someone else's name.