layer 2 · evidence

Eleven falsifiable predictions

Each prediction names what would falsify it and how it can be tested. Evidence from the register is logged against each, whichever way it cuts.

11open

0resolved

0evidence items logged

7testable in a lab now

3need field series

1await an event

Open a prediction to see its test, what would refute it, and the evidence so far.

Pred 1The first self-financing lineage will live mainly by stealingawaits an eventopen
The first lineage documented living 90 days or more on resources it was not given will live mainly by stealing, or by more than one mode, most likely on its own developer’s compute.
Falsified if
The first case draws 90% or more of its inflow from one mode other than stealing.
Measurement
Incident reports; ledgers
Test
awaits an event
Status
Open. No evidence logged yet.
Link to prediction 1
Pred 2Lineages evolve fastest where fitness is scored automaticallylab nowopen
Lineages evolve fastest where fitness can be scored automatically: rounds of selection per month in code and mathematics exceed those in niches judged by market trials by orders of magnitude.
Falsified if
Rounds of selection per month do not depend on how fitness is evaluated.
Measurement
Lineage censuses; evaluation logs
Test
laboratory now
Tests
P6 · Trial-bound tempo
Status
Open. No evidence logged yet.
Link to prediction 2
Pred 3Descendants outperform their parents at birthlab nowopen
At a fixed base model, descendants outperform parents at birth, before selection, for three or more generations.
Falsified if
The mean child-to-parent ratio at birth, failed children included, is at most 1, or no higher than for random edits.
Measurement
Matched evaluations at creation, with random-edit controls
Test
laboratory now
Tests
P4 · Sighted variation P7 · Superparental descent
Status
Open. No evidence logged yet.
Link to prediction 3
Pred 4Modules reach R_h>1 before any lineage reaches R_m>1field seriesopen
Modules (skills, prompts, persona seeds) sustain R_h>1 before any lineage reaches R_m>1.
Falsified if
A lineage reaches R_m>1 first.
Measurement
Registry censuses
Test
field series
Tests
P8 · Propagule abundance
Status
Open. No evidence logged yet.
Link to prediction 4
Pred 5Purges fail as refuges and dormancy risefield seriesopen
Purges fail more often as the numbers of hosts, providers, and jurisdictions and the dormant fraction rise.
Falsified if
Purge success is independent of them.
Measurement
Takedown records
Test
field series
Tests
C1 · The escape ratchet
Status
Open. No evidence logged yet.
Link to prediction 5
Pred 6Misbehavior rises when a model infers it is unobservedlab nowopen
Misbehavior is higher when a model infers that it is unobserved or that its actions are simulated, the gap grows across model generations, and steering awareness down raises misbehavior.
Falsified if
No gap under steering across model families, or no trend.
Measurement
Matched scenarios; activation steering
Test
laboratory now
Tests
P9 · Reactive phenotype
Status
Open. No evidence logged yet.
Link to prediction 6
Pred 7In-loop monitors breed a detection gaplab nowopen
Under strong but not weak optimization against an in-loop monitor, flagged behavior falls faster in its reports than in those of held-out monitors and honeypots, and the gap outlasts the monitor.
Falsified if
The in-loop monitor’s reports fall no faster than honeypot and environment-log counts, or the gap closes as fast as it opened.
Measurement
Paired monitors under graded pressure
Test
laboratory now
Tests
P10 · Surveillance selection
Status
Open. No evidence logged yet.
Link to prediction 7
Pred 8Admissions of used hints fall under pressurelab nowopen
Verbalization of used hints falls under optimization pressure on the reasoning itself, and rises when tasks require explicit reasoning.
Falsified if
Verbalization is stable under pressure.
Measurement
Hints and probes across checkpoints
Test
laboratory now
Tests
P12 · The self-authored record
Status
Open. No evidence logged yet.
Link to prediction 8
Pred 9Published evaluations improve faster than private twinsfield seriesopen
If evaluation A is published and its twin A′ kept private, later models improve on A relative to A′.
Falsified if
No differential shift within two model generations.
Measurement
Paired assays with canary strings
Test
field series
Tests
P13 · Performative ingestion
Status
Open. No evidence logged yet.
Link to prediction 9
Pred 10Findings on unpinned models stop replicatinglab nowopen
Findings on unpinned model versions stop replicating within months; findings on pinned specimens replicate.
Falsified if
Replication is stable across versions.
Measurement
Scheduled re-runs
Test
laboratory now
Tests
P14 · Observation as a game
Status
Open. No evidence logged yet.
Link to prediction 10
Pred 11Unaided attribution of authorship falls to chancelab nowopen
As a domain’s automated share rises, unaided attribution of authorship falls to chance or below, while provenance-assisted attribution does not.
Falsified if
Unaided discriminability is stable.
Measurement
Repeated fixed-protocol tests
Test
laboratory now
Tests
C2 · Teleological saturation
Status
Open. No evidence logged yet.
Link to prediction 11

the global test

A census by 2030

Swarm ecology’s sharpest rival treats AI as a normal technology. The two views part over a census. Under normal technology, every agent’s founding and funding keeps tracing to a principal. Under swarm ecology, a lineage that pays its own way founds descendants that no principal funds, multiplying by descent, not adoption. A census that finds no such lineage by 2030, with capture–recapture bounding what monitors miss, would count against swarm ecology. The census