# Reference-neuron assessment

7 October 2026. This records the control, endpoint and time-horizon discussion with Brice and defines the next Explorer view. It extends the earlier endpoint proposal. The existing 45 mechanism assessments concern different population claims and retain their original probabilities.

## What the comparisons mean

The default comparison is a healthy reference (H) versus that reference with one specified perturbation (H+M). Match other initiating inputs; allow downstream chemistry, physiology, compensation and tissue responses to change. Assess separately whether diabetes produces the chosen exposure. A second comparison, D versus D with M corrected, asks how much injury correction could prevent in diabetes. It is not diabetes plus an additional dose. Interaction and redundant routes can make these two contrasts differ. The three mechanisms in this pass are methylglyoxal, SARM1 activity and loss of NGF support.

The target is a person-weighted, then neuron-weighted, then arbor-weighted reference population: healthy 50-year-old men without neuropathy, with a mature peptide-producing, heat-responsive cutaneous C nociceptor in lateral distal-leg skin 10 cm above the lateral malleolus. Keep the soma, peripheral and central axons, skin, glia, vessels and immune environment connected. Establish identity at baseline and retain it if CGRP or other marker expression changes. This is a stated scope, not evidence that this is the median human neuron. Other ages, women and non-heat-responsive neurons are outside this first estimate.

At baseline, choose one arbor by a basement-membrane entry and its epidermal descendants. Define its fixed skin territory and original viable connected length L0. Track biological continuity through ordinary molecular turnover; original identity does not mean keeping the same protein molecules. The neuron must generate at least five expected excess action potentials at the central axon near dorsal-root entry during the heat assay. The assay starts at 32 C, rises at 2 C/s to 47 C, holds for 5 s, then returns to 32 C at 2 C/s. Count excess spikes over a matched 32 C sham within 30 s. Use the same territory and input at later times. The expected response avoids making single-trial noise the endpoint. This ideal readout does not claim a currently available human longitudinal measurement.

## Four endpoints

1. **Neuron death:** completed irreversible loss of the viable identified DRG soma by time T, from any cause. Marker loss, axon disconnection or a dying fragment is insufficient. This is a cumulative event.
2. **Heat-signal failure:** the expected excess spike count in the specified assay at T is at most 10% of its baseline value. It includes failed transduction or propagation. It is a current functional state, can recover, and does not measure conscious pain. Soma death counts as failure. This is not a claim of permanent failure.
3. **Arbor destruction:** the baseline arbor has physically degenerated or been dismantled by T, with no viable remnant of that arbor. Retraction of a living branch alone is not destruction. A disconnected but viable fragment has not yet completed destruction. Regrowth does not erase the original event. Substantial partial damage may occur without meeting this complete-destruction endpoint.
4. **Replacement not completed:** in a separate matched challenge, remove the selected epidermal arbor at time zero while preserving its parent neuron and proximal axon. Use the same initial lesion in H and H+M, with no selection on subsequent survival. By T, the original neuron has not yet formed a connected arbor within the original territory that reaches both 50% of L0 and 50% of its baseline heat response. The first time both criteria are met is repair completion; later loss belongs in the other endpoints. Death before completion is non-completion, not a dropped observation. This endpoint concerns repair after a defined lesion, not spontaneous injury incidence.

The numerical response and repair cutoffs are chosen conventions. The heat-ramp literature supports specifying the ramp, not these cutoffs. Assay readouts refer to otherwise matched counterfactual systems at each time, avoiding repeated test-induced damage.

## Times and probability

Assess 1 hour, 1 day, 7 days, 30 days, 90 days, 1 year and 5 years after perturbation begins. Inputs persist until the horizon unless the mechanism report explicitly states otherwise. Biological aging continues equally in both arms. No extrapolation to a 10,000-year human is defined.

For endpoint E, let pH(E,T) and pM(E,T) be true person-weighted endpoint probabilities in the specified reference populations. Let delta = pM - pH. The displayed judgment is q = P(delta >= 0.10 | available evidence): our belief that this exposure adds at least ten percentage points of endpoint risk. For replacement this is the excess risk of non-completion. Ten points is a material-effect convention, not an observed effect. q is not pM, a patient prognosis, a sampling confidence interval, or the fraction of parameter samples above a cutoff. Show low/central/high subjective judgments under stated evidence-transfer assumptions; these are sensitivity judgments, not empirical confidence limits or a fitted Bayesian posterior. Preserve the reasons for each timepoint and endpoint.

At time zero, newly completed death/destruction is absent by construction; in the lesion challenge both arms have yet to repair, so the difference is zero. With finite biological rates, new structural events approach zero as time approaches zero. Do not force one-day effects to zero. Cumulative death and destruction probabilities cannot decrease; functional state can recover; first-repair non-completion probabilities cannot increase. The risk difference and q need not be monotone: controls may catch up. Do not impose monotonic q as a substitute for this reasoning.

## Evidence and calibration

Each mechanism report specifies the physical input, compartment, duration, healthy comparator, and an independent judgment about whether diabetes reaches this input. Review primary human, animal and cell evidence, counterevidence, and exposure/endpoint transfer. Count donors/animals/preparations rather than cells as independent replication. Recalculate at least one useful quantitative source constraint, and test whether it supports the target claim or merely an adjacent endpoint. Explain the judgment range. Do not copy the legacy five-year probabilities, invent likelihoods or convert a constructed kinetic scenario to a human failure rate.

This finite pass produces a literature-grounded assessment, explicit numerical beliefs, a transparent numerical check for each mechanism, and independent scientific and screenshot-based interface review. It does not claim that every possible computational investigation has been exhausted. Public material excludes internal scratch transcripts and controlled-access data.

## Common sources

- Human DRG subtype context, eight donors: https://pmc.ncbi.nlm.nih.gov/articles/PMC9272153/
- Distal-leg measurement site and age dependence, 550 people: https://pubmed.ncbi.nlm.nih.gov/21040142/
- Heat-ramp dependence, human single C nociceptors (units are not independent donor n): https://pmc.ncbi.nlm.nih.gov/articles/PMC1176139/
- Interaction and preventable fractions: https://pmc.ncbi.nlm.nih.gov/articles/PMC2836214/