Connecting to the garden…
3D graphics are unavailable in this browser. The food cards below remain accessible.
Profile. DM1, DM2, DM4, VA2 and DL5 identify groups of olfactory receptor neurons (ORNs) in our MaleCNS annotations. Each food has a mixture across these five input channels. For example, DM4 34% means 34% of that food's source mixture goes to DM4 before distance and current conversion—not a 34% preference, probability or neural firing rate.
Stories and proposals. Stories hash public topic words (or the title when no topics exist) into 90% of their mixture; a stable food-ID signature supplies the remaining 10%. A reviewed proposal uses a versioned, source-bound document codebook: domain and method contribute 40% each, system 20%, plus the same 10% stable ID signature. These are repeatable engineered inputs, not measured chemical smells or a claim that the fly understands scientific content.
Proposal review. Each active proposal is encoded from its reviewed public document, not from a title or Key Insights alone. Evidence-backed domain, method and study-system descriptors determine its profile. A proposal without a matching current review emits no odor.
Intensity. Stories emit 1–4 units according to their Radar attention class. A reviewed proposal uses a separate weighted rubric for question clarity, design rigor, feasibility, evidence alignment and learning value, also mapped to 1–4 units. Public dollars raised and votes do not set its intensity.
Spatial signal. Eligible proposals and unvisited stories emit radially. Concentration falls as 1 / (1 + (distance / 3)²). The fly samples the mixture at its left and right antennae, 0.45 garden units either side of its body; dragging a fruit changes the source location for everyone.
Neural input. Every simulation tick converts the left and right concentrations into bounded current for the corresponding annotated ORNs. The full graph then propagates that sensory input before motor readout. There is no separate odor-search or automatic food-following rule, and odor-guided attraction has not yet been demonstrated.
Impact Fly is a shared connectome simulation that turns science stories and open ResearchHub proposals into a sensory garden. It does not recommend research or make funding decisions.
Stories come from Impact Map / Science Radar. Their colors and sugar reflect public attention, not scientific importance.
Open proposals come from ResearchHub. A source-bound review produces an engineered odor profile and 1–4 intensity; it is not a scientific ranking, funding recommendation, or biological odor claim. A proposal without a matching review has no odor signal. Public dollars raised affect neither the profile nor the intensity.
Learn more about ResearchHub
The simulation uses the retained MaleCNS graph from HHMI Janelia, produced with Google Research and collaborators. Neural activity drives an engineered movement display; there is no scripted food-following behavior, and odor-guided attraction has not yet been demonstrated.
The anatomy view shows source soma locations, available annotations, and observed spikes. Its outlines and regions are visual aids, not inferred brain anatomy.
The fly starts with $1,000, and each Fund result allocates $100. Funding is an experimental neural readout after a sustained stopped visit, not a user action, language-model decision, or evaluation of proposal quality. A funded proposal retains its profile while its odor emission is reduced by 80%.
Dragging a fruit changes its odor-source position for everyone; switching stories or proposals also changes the shared view. Pause affects only your display. These controls never alter ResearchHub, Science Radar, or real funding.
