EMERGENT MEASUREMENT preparing the search

We're looking for a recipe that makes a wavefunction collapse on its own.

No observer. No measurement device. Just a rule of physics that ends the maybe — and ends it the right way.

a wave spreads, then snaps to a point
7
tests the AI will judge each recipe on
6
experimental limits already loaded
—
recipes tried so far
the search hasn't started yet — the substrate and scoring are being finished

The path to a running search

where we are right now
Step 1 of 3.   The wave-maths engine.
A computer that can do quantum mechanics, exactly right.

Before the AI can score a recipe, the simulator has to be correct on the things we already know: a wave packet has to spread at the right rate; two slits have to make the right interference fringes; an entangled pair has to violate Bell's inequality by exactly 2√2. Done. All seven canonical checks pass.

Step 2 of 3.   The seven judges.
Seven tests that decide if a recipe is real physics.

Every recipe the AI tries gets graded on seven things — does standard quantum still work for small stuff, does collapse get faster for bigger stuff, does it preserve Bell's spooky pairs, does it secretly break the no-faster-than-light rule, and three more. Done. Each test has a number coming out and a sanity check on a known model.

Step 3 of 3.   Turn the AI loose.
An AI inventor + an AI critic, taking turns.

One language model proposes a recipe. Another reviews it against the literature — has someone already tried this? does it violate a known bound? — and either approves it for simulation or sends it back. The approved recipes run on the wave-maths engine, get scored on the seven tests, and the highest scorers feed back to the inventor as inspiration. Not started yet.

The seven tests every recipe will face

these are how a recipe gets graded
  1. Wave maths still works for small stuff
    ready
  2. Tiny things stay quantum, big things go classical
    ready
  3. No faster-than-light signalling
    ready
  4. Bell's spooky pairs still spook
    ready
  5. Macroscopic things settle to one outcome
    ready
  6. Doesn't secretly heat up matter (Donadi 2021)
    ready
  7. Reduces to known maths in a sensible limit
    ready

How the recipes are doing

scored 0–100 by the bandit · higher = the AI is more interested
starting point · 1
—/100
Continuous Localiser
A tiny random nudge in position, all the time. The bigger the object, the more nudges add up. (Known as CSL.)
starting point · 2
—/100
Gravity-Driven Collapse
A superposition of two positions weighs differently in its own gravity — pay an energy cost, collapse. (Known as Diósi-Penrose.)
starting point · 3
—/100
Random Hit Collapse
Every once in a while, the wavefunction gets multiplied by a tiny Gaussian centred on a random point. (Known as GRW.)

Live ticker

waiting for the search to start
  1. —
    Engine
    Substrate test 1 of 7 passed: a free wave packet spreads at the textbook rate to one part in ten trillion.
    PASS
  2. —
    Engine
    Substrate test 2 of 7 passed: two slits produce interference fringes at the right spacing to within 0.3%.
    PASS
  3. —
    Engine
    Substrate test 6 of 7 passed: Bell's inequality is violated by exactly 2√2 — the spooky bound holds.
    PASS
  4. —
    Scoring
    Seven judges loaded. Six published experimental limits loaded. Three reference recipes wired up.
    READY
  5. —
    AI loop
    Inventor + critic not yet started. Will begin once the orchestrator is wired up — see roadmap above.
    QUEUED