Eighteenth check: scrambled brains at 30 spikes/s, and the browser demos re-checked

Scrambled brains at 30 spikes/s

Our own test · pre-registered · brain only (Shiu et al. LIF on FlyWire v783) · one model · not a real fly

What we did. Every LC4 and LPLC2 neuron of one eye fired at 30 spikes/s for 1 s, the weakest input at which the real brain's turn command appeared on 8 Oct (far-side DNa02 8 Hz for a shadow on the left eye, 12 Hz on the right). We ran two degree-preserving scrambled brains per side (shuffle seeds 4 and 5: each neuron keeps its number of connections), one 1-second trial each, with trial seed 2, the seed of the real-wiring body runs at 30 spikes/s. The readouts, the turn rule (far DNa02 more than 3 times near DNa02 plus 5 Hz), the escape rule and the drive maker are the ones of 5-10 Oct, unchanged.

Checks before reading (all passed). Code: the only change to the 9 Oct trial script is an option that sets the trial seed of a scrambled brain (before, every scrambled brain ran at trial seed 0). Drive maker: our 8 Oct drive maker, run on the saved 8 Oct brain files, reproduced the fingerprints of both real 30 spikes/s drives and of the no-brain drive. Environment: the 8 Oct real-brain trial at 30 spikes/s (left eye, seed 2) was repeated exactly: 10,073 whole-brain spikes, 379 active neurons, 4,807 input spikes and every descending-neuron rate. Graphs: each scrambled brain passed its property checks and was the same graph as its 80 spikes/s twin.

By the pre-registered rules both sides read the same: no scrambled brain passed the turn rule or the escape rule, and all 4 steering drives are byte-identical to the no-brain drive, so in the body they take the no-brain walk and no body run was needed. This matches what we wrote before the trials.

• Not matched for activity. The scrambled brains were not matched for overall activity: they fired 48% (left) and 44% (right) as many spikes as the real brain at this strength. Most of those spikes were the stimulated input neurons themselves; outside the input, the scrambled brains fired about 1% as many spikes as the real brain. At 30 spikes/s we did not run a turned-up scrambled brain (on 6 Oct, at 80 spikes/s, 3 turned-up scrambled brains also gave no turn command).
• Small and narrow. Two shuffles per side, one trial seed, one 1-second trial each, one model, brain only; scrambled brains were run at 30 and 80 spikes/s, not at 20, 40 or 60.
• The ledger. Our looming row (byo-flygym-loom) is not re-scored; its arms stay the 80 spikes/s test, and these sentences are added to its note.

The relays: a stricter control, counted

The relay headline of 13 Oct stays: The two sides disagree: on one side silencing the relays lowered the turn neuron's rate more than matched activity elsewhere, on the other it did not. We do not draw a conclusion.

This is a count from the map and the 15 Oct trial files, not a new trial. The relay panel with both charts: Is it real?

The browser demos of 3 Oct, re-checked

The same written procedure as on 3 Oct, written down again before the first load, and started only after the brain simulation had finished and its environment was deleted. Help the Fly Escape changed upstream since 3 Oct; its README still names the same action.

4 new entries and 2 new ledger rows

Project claims are the projects' own; ledger numbers are the authors', scored by our fixed rules and not recomputed by us.

• Flycraft A (thepuneeth/Flycraft): a frozen MaleCNS v1.0 nerve-cord graph (11,000 nodes) is simulated as leaky integrate-and-fire neurons, and its motor-neuron rates set the joints of an 18-joint MuJoCo hexapod through a fixed hand-made map; there is no rhythm generator in the files read. In the author's test the real nerve-cord graph moved the hexapod 0.85 mm forward in 8 s, fully rewired graphs 3.5 mm backward, and a time-shuffled real signal 0.72 mm (shuffled weights 2.8 mm backward, zero drive 3.3 mm backward; the author's p for real vs time-shuffled is 0.29). The author's results file is marked pending the author's own review; the paper's acceptance is the author's statement only. Ledger: helps (fair, strong). The grade describes what the code does, not the size of the effect.
• fly-connectome-reservoir B (cssius111): a fly-swatter game in which a frozen MaleCNS-derived network drives escape through a hand-designed decoder, with no wiring control on the game. Its separate reservoir test scores real wiring 100%, target-permuted wiring 83.9% and no connections 50% on a left/right task (3 seeds, trained linear readout); a raw-stimulus readout also reaches 100%. Ledger: helps (partly fair, strong); the no-brain baseline is not beaten.
• virtual-fly-brain D (bookseal): a PPO policy steers FlyGym's own turning controller toward food; no connectome is in the control loop. Its own run report says the learning improvement was not shown.
• Fruit Fly Connectome Explorer (n/a; chamodiHsenevirathna): ranks MaleCNS v1.0 neurons by connections and strength across weight thresholds; no neural dynamics.

Excluded on 17 Oct 2026 (no committed result file or wiring comparison in code at their current commit): wenlibo90/jumper-connectome, heyziq05/VirFly, arthen97/drawfly, snownmouse/cyber-fly-brain, undoaundo-ctrl/fly-brain-racing, Paradoxsmile/flybrain, Pratt2321/drosophila-carr-vision, marginli/brain_division and AloyMAitraz/fly-brain-visualizer.

The ledger now has 60 studies, 51 with a wiring null: helps 24, no difference 11, worse 4, mixed 10, not yet scored 2 (Does fly wiring help?).

Verdicts, tracking and freshness

• Roblox avatar claim: stays U. Referee check: no builder named in the NiftRex video description or on the Roblox profile (17 Oct). The BloxStrike game page names no fly brain or builder; a second recap (Roblox Universe) is recorded from search data only (verdict row).
• Repositories: 137 checked: 129 unchanged, 7 with new commits (among them help-the-fly-escape, fanc, flybrain-robot-bridge, flywire-connectome-scz, adfly-connectome, whose history was rewritten, and neuropest), 1 still gone (flydoom, record kept).
• Links: 223 checked: 195 answered, 12 redirected, 13 blocked our checker, 3 client errors (two demo pages of the former znatgost address and flydoom). fruitfly.live still has a TLS certificate error.
• Verdict rows: the repositories of 15 of the 22 verdict rows re-checked; none has new code since grading.
• Videos: 56 of 56 available on 17 Oct 2026; no video added or removed.
• Lead scan: 86 code repositories since 15 Oct, 31 of them new to us; they have not been opened yet and wait for the next check. No video met the bar for a new entry.

What we could not resolve

• Activity. The scrambled brains at 30 spikes/s were much quieter than the real brain outside the input (about 1%), and we did not turn them up. A turned-up arm at 30 spikes/s would need its own pre-registration.
• Fly Worker's "no activity". The on-page text we track may have been static labels after a page change; we did not check by eye.
• fly67. The 404 may be temporary, or the site may have moved again; we did not check its repository.
• Flycraft. It meets our grade A rule as written, but its controlled effect is under 1 mm; whether the ledger needs a note for very small effects is open.
• Leads. 31 new repositories wait for their first check.

Data behind this report: projects.json, controls.json, videos.json. The 30 spikes/s real-brain series: report of 8 Oct 2026; the body runs at each strength: report of 14 Oct 2026; the 3 Oct browser check: report of 3 Oct 2026.