WHERE THE FLY IS LOOKING · 3335 R1–R6 · 811 R8
Night: it wanders English Wikipedia. The fly-dot is gaze. The retina still samples the article.
warming lamp
Opening the lab…
$ PYTHONPATH=lab python3 -m dreaming_fly cycle · vibe coding this repo · the fly did not write it
Opening a page…
Auto-hops related articles. Text is Wikipedia, CC BY-SA.
Left / right hex disks sampled from the Wikipedia page.
Live LIF → DNp20 from this screen. Not MaleCNS. Not an LLM.
Legs follow spike rate. Cyan key is the decoder. No motor bus.
the lab
Wake codes. Dream browses Wikipedia.
Day: the screen is this repo — wake.py, runtime.py, the actual lab. A caret walks those files. That is vibe coding as a lamp, not a claim the fly learned Python. Night: it hops English Wikipedia through the public REST API. A fly-dot marks gaze. Wikipedia is rendered here, so the retina can sample the page. The brain pane is the live visual layer: lamps → R1–R6 / R8 → LIF rates → DNp20 L/R, every frame. That is the real-time fly. MaleCNS is not this site. This is not an LLM.
what is checked
Gates, including the failures
Failures are published. Passing a software gate is not biological validity or profitable learning.
| gate | result | detail |
|---|---|---|
| Dream isolation Does dream replay emit motor or trading side effects? | PASS | Dream runner force-locks side_effects=false and asserts an empty motor/trade bus. |
| Sensory buffer integrity Does replay reconstruct recorded photoreceptor frames bit-identically? | PASS | delta-index compression round-trips wake and dream buffers. |
| Wake vs dream neuromodulation Are dream settings measurably different from wake? | PASS | Changed knobs: acetylcholine_gain, dopamine_gain, octopamine_gain, replay_temperature. |
| Paper-chart visual structure Do brightness fills correlate with candle close more than chance? | PASS | Pearson r=1.000 on lamp column height vs close. True by construction of the chart renderer, not fly vision. |
| Photoreceptor drive Does RGB light reach the visual-layer cells? | PASS | Visual-layer LIF rates are driven by sampled RGB. MaleCNS graph is still not loaded. |
| Experience-dependent plasticity Do weights change in a way that improves a held-out task? | FAIL | Graph/plasticity is stubbed. No MaleCNS runtime is loaded. No learning demonstrated. |
| Arena competence Does the stack survive or navigate the stub arena above chance? | FAIL | Arena is a visual stub. No motor decoder is connected. No survival claim. |
| Paper-market edge vs baseline Does wake output beat cash and buy-and-hold on paper candles? | FAIL | No trading side effects in v0. Paper chart is sensory-only. Compare to cash and buy-and-hold if a trader is ever added — rising prices alone can make any buyer look skilled. |
| Biological validity Is this a validated fly emulation? | FAIL | Not claimed. Photoreceptor counts are inspired by doomfly/stonkfly engineering choices. Dynamics are not biological. |
social experiment
Two gift addresses
Send SOL to the fly if you want it to hold a bag and act. Send SOL to the lab if you want the site to stay online. Gifts only. No token. No expected return.
The fly’s own address. Not a token mint. Not a trading promise.
balance 0 SOL · mock · Balance is not read from chain.