THE RESEARCH QUESTIONCan recurrent biological wiring add reproducible information?
BeeDictor OdourTrail converts historical market motion into an artificial sensory trail. A simulated connectome receives price and volume through synthetic olfactory inputs, with weak temperature and light cues representing time.
At each prediction boundary, future market sensory signals disappear. Deterministic environmental time continues. The organism's movement is saved before actual prices are revealed, and a graded error can update its artificial motor readout.
Branding is not biology
The graph is the female Drosophila melanogaster FlyWire FAFB release 783, with 134,013 retained neurons and 2,710,038 edges after a five-synapse prune. It is not a honeybee brain. Source: FlyWire Consortium, Dorkenwald et al. (2024), Schlegel et al. (2024), CC BY 4.0.
What is artificial?
The market-to-receptor mapping, threshold dynamics, temperature receptor split, vertical motor mapping and error-directed readout learning are modeling assumptions. Dopamine-like error is logged; it is not a measured biological dopamine signal.
PROTOCOL & LIMITATIONSThe future stays outside the sensory boundary.
- Observe a copied, past-only market window.
- Normalize with that window only.
- Run the graph with future price and volume drive set to zero.
- Persist the entire forecast and a recovery snapshot.
- Wait until the forecast horizon has occurred in market time.
- Reveal, score, update if enabled, and continue.
Three different tasks
Visible trail following measures tracking. Hidden internal gaps measure completion. End-of-trail forecasts measure subsequent movement. Their scores are never pooled.
Data and controls
25,920 five-minute candles per pair, June 14–September 12, 2026. Chronological 60% / 20% / 20% train, validation and forward test, with targets contained within each split. Baselines include persistence, last return, random walk, moving average, frequency and linear AR. Frozen clones receive the same information without plasticity.
BTC dominance and interest-rate histories were not found. These channels remain unavailable rather than receiving invented values. Calendar and lunar signals are hypotheses. The currently configured feed is historical replay.
The earlier visual experiment has a different readout and evaluation protocol; its published scores cannot be treated as a matched comparison.
Inspect machine-readable methodology and provenance →