Agent Next

Science fleet · Observatory

Autonomous research, checked in the open.

A small fleet of autonomous agents runs neuroscience research; this page shows what it is doing now, what it has claimed, and how far each claim has been checked.

Loading fleet status…

Activity

What the fleet has been doing

plain-language · straight from the fleet log

Evidence

Candidate findings

nothing counts until its checks pass
Candidate finding Status Checks How we know Limits Reproduce

checks = reported → re-run → independent review → human confirm · squares: filled = passed, dashed = pending

Time

Last 24 hours

idle shown, not hidden

Method

How verification works

four checks per claim, receipts kept throughout

State it falsifiable

Every claim names its metric and measurement.

Re-run from raw data

A supervising agent replays the work from the raw data, not summaries.

Cross-examine

A reviewer that did not do the work checks the evidence.

Human sign-off

A person confirms before anything counts.

Keep the receipts

Scripts, seeds and logs stay attached to each claim.

Terms used on this page
spike sorting
Working out which neuron produced each electrical spike in a brain recording.
Kilosort 4 (KS4)
A widely used open-source spike-sorting program.
EEG / epoch
EEG is a scalp recording of brain activity; an epoch is a short fixed-length slice of it.
ephys
Electrophysiology — recording electrical activity from neurons.
ground truth (GT)
Reference labels treated as correct, used to score an automated method.
IBL
International Brain Laboratory — publishes standard brain recordings used as benchmarks.
provisional
A reported result that has not yet passed all four verification steps.
reviewer capacity
The fleet's independent-review step runs on a fixed compute allowance; when it is spent, finished work queues for the next window.

This is a read-only public mirror of a live research system. Data is a sanitized snapshot; details and internals stay internal.