VoltDecision

• 

  Official lock 9 → 94 on first paint.

• 

  Official seed after repair: Coordination Score 94. e-wound still lies at −70 mV.

Try it (ten seconds)

https://moscraciunxxx.github.io/voltdecision/

1. The first number is the official lock 9 → 94.
2. Click Replay 9 → 94. Live score hits 9 (the lie stays labeled lock 94), then 94. Incident hash stays 7b11a7758c9b7cfc.
3. Or step it: Run incident only, then Apply repair & replay.

No install. The judged door is that URL, not localhost.

Held-out seed 20260912 is 11 → 89 on the same repaired policy. Ablation: no-fusion 9, no-ring 20, no-bridge 94. Stuck e-halo is 62 → 94. One G_GJ slider: at 0.5 nS repaired falls to 53.

A wet lab would next inject this repaired policy into five electrodes and keep the same replay hashes. This page is the twin. That dish is not here. Pattern class: wound current and gap-junction coupling (McCaig et al., Physiol. Rev. 2005; Levin, Cell 2021) — not a claim those experiments ran here.

Inspiration

Bioelectric networks — not just neurons — use membrane voltage to coordinate. A wound should recruit a hyperpolarized halo and close. A controller that trusts one lying electrode will decide the tissue is already at rest and do nothing. VoltHacks asked for systems that interact with the physical world. This lab is an honest digital twin of that loop: inject a voltage pattern, watch the tissue decide, repair the policy, prove the replay.

No dish was on this desk. No ESP32 is claimed.

What it does

VoltDecision simulates a 12×12 epithelial sheet as a Kirchhoff / gap-junction graph:

$$C \dot V_i = -g_\ell(V_i - E_\ell) + \sum_j g_{ij}(V_j - V_i) + I_i^{\mathrm{inj}} + I_i^{\pi}$$

Official compound incident (seed 20260905):

• Depolarizing injury current on a left-edge wound
• Gap junctions uncoupled on the col-3 / col-4 seam
• Electrode e-wound stuck at rest (−70 mV)
• Baseline policy: single-electrode PI toward rest → error is zero → no stim

Diagnosis uses electrode traces only. The repaired policy drops the flatline channel, fuses halo electrodes, and applies a hyperpolarizing ring. Replay keeps the same incident-input hash and a different policy hash. Coordination Score moves 9 → 94.

Held-out incident (seed 20260912): same failure class, mirrored geometry. The controller was not rewritten. Score moves 11 → 89.

After repair, dashed traces overlay the baseline plot — the lie stays on the stuck channel. Click a cell for (V_m), injury current, stim, and KCL residual.

How we built it

• Implicit Euler on the graph Laplacian, Jacobi-preconditioned CG
• Electrode-only controllers (hidden (V) is not an input)
• FNV-1a hashes over a canonical incident/policy payload
• Vitest locks: official 9 / 94, held-out 11 → 89, no-fusion 9, no-ring 20, no-bridge 94, stuck e-halo 62 / 94
• Static React/Vite lab on GitHub Pages

Challenges

The first repaired controller peeked at the hidden voltage field. That is leakage. It was removed; fusion is halo electrodes only. The seam-bridge ablation did not drop the official score — so the page does not pretend it did.

Scope

Hackathon prototype on simulated epithelium. Not a medical device, not a safety certification. The Coordination Score is scoped to four pattern checks on the named incident. A wet lab would next put this policy on five electrodes. That dish is not here.

Built With

• bioelectric
• digital-twin
• github
• kirchhoff
• react
• typescript
• vite
• vitest