Quantium AI Lab · Objective C.3
Introducing Cavendish.
A lab that keeps score.
A research instrument that watches the field continuously, turns noise into a small number of tracked positions, tests the promising ones, and publishes what it finds — with the evidence attached and an expiry date on everything.
Three questions
What is coming. What works now. What doesn't.
The third is not a by-product. A firm that only says yes is indistinguishable from a vendor.
The map
30 fields elected. 10 candidates in the pool.
Deep on few, thin on many, and the two kept visibly separate. Candidates carry a one-line thesis and a signal tier, and cannot be cited as findings.
Tested this quarter
Only tested work becomes a recommendation.
Signal and tried will vastly outnumber it — which is fine, and is exactly why the tier label has to be loud enough to survive being screenshotted into someone's slides.
Proof
The track record, marked to a market that does not exist.
Killed
What we got wrong, with names and causes.
Nothing goes to the graveyard. It is the library filtered to killed outcomes, with lineage intact — the reasoning that killed something is the reusable asset.
Built
Tested, with a live artifact.
The demonstrations page is the library filtered to tested items with something you can run.
In flight
Running now, with the kill condition on top.
The preregistration sits at the top of every experiment and does not move. A hypothesis that scrolls out of sight is one nobody is held to.
Geiger · latest signal
Watching the field continuously.
Sources organised by lead time, not platform. Bands 1–2 are supply; band 3 is demand. The gap between them is where advisory value lives.
Routes in
Anyone in the firm can put something in. The lab decides what goes deep.
The lab is a bottleneck on validation, never on the firm's awareness.