ProtocolLogic

independent research · est. 2026

A research program on how cognition-relevant behaviour arises in a biologically grounded neural substrate — and on what it takes to make claims about that honestly.

The work has two halves that are hard to separate. One is the instrument: a chemistry-driven, anatomically constrained substrate simulation, built to be observed rather than fitted to a target. The other is the discipline around it — what counts as a measurement, what counts as an artefact of the equations you chose, and how to tell the two apart before you publish either.

This page is a register, not a portfolio. Closed lines and falsified hypotheses appear alongside open ones, because in a simulation the negative results are the ones that carry information.

Run by one person. No institutional affiliation, no external funding, no staff, no peer review. Nothing here has been externally validated. Read it as a working notebook made public, and weigh it accordingly.

Inference ladder

  1. 1 Measured
  2. 2 Causally perturbed
  3. 3 Mechanistically consistent
  4. 4 Mechanistically identified
  5. 5 Theory-dependent

Every entry below is stamped with the highest rung its evidence actually supports. The scale runs from what was observed directly to what only holds if you accept the surrounding theory. A high rung is not a better result — it is a larger loan against interpretation.

Register

Substrate simulation

rung 3

A substrate built as an instrument, not a model

Most neural simulations are built to reproduce a target: fit the parameters until the output matches a recording. This one is built the other way round. Composition and connectivity are fixed from biology first, and whatever dynamics follow are treated as the readout — including when they are inconvenient.

The cost of that choice is that failure has to be diagnosed, not tuned away. When the substrate does not do what was expected, the working assumption is an error in the mathematics or a missing biological factor, and the search goes there rather than into the free parameters.

Status — running. Maturation phase separated from the observation phase, so that the pressure applied during setup cannot be mistaken for behaviour discovered later.

Dynamics · closed line

rung 2

A regime-separation hypothesis, closed on the route tested

One line of work asked whether a specific gating mechanism could separate two dynamic regimes in the substrate. Under direct perturbation the separation did occur, but it attributed to a simpler cause already present in the setup. The proposed mechanism was not doing the work.

The line is closed on the route that was tested — a narrower claim than "the hypothesis is false", and the distinction is kept deliberately. What survives is a demonstration that the target dynamics are realisable at all, which is retained separately from the hypothesis it failed to support.

Status — closed, tested route. Reopening requires a different mechanism, not a re-run.

Validation infrastructure

rung 5

Separating the object, the instrument and the criterion

A simulation that is simultaneously the thing being studied, the tool doing the studying, and the standard the results are judged against cannot validate itself. That circularity was present in this project and is now being unwound into an independent consistency layer with its own sources and its own verdicts.

Role separation on paper is not role separation in practice. The design therefore turns on two mechanical properties — extraction performed blind to the substrate's own documentation, and criteria registered before any verdict is computed — without which the split is decorative.

Status — design stage, scoped pilot. Expansion gated on completing at least one externally observable test.

Numerics & engineering

rung 1

Refactoring under a byte-exact equivalence gate

Scientific simulation code decays in a particular way: a cleanup lands, the numbers shift in the fourth decimal, nobody notices, and six months of results are quietly incomparable. The rule here is that any restructuring must reproduce the reference trajectory byte for byte in the static limit, checked by direct memory comparison rather than a tolerance.

It is a strict gate and it does reject otherwise reasonable changes. That is the point — it converts a class of silent numerical drift into a build failure, which is the only form in which it reliably gets fixed.

Status — standing policy. Applies to every refactor without exception.

Method

rung 4

Consequences of your own equations are not discoveries

The dominant failure mode in simulation science is not fabrication, it is category error: reporting as a finding something that follows necessarily from the model's construction. If a behaviour is entailed by the chosen equation forms, observing it in the output confirms the arithmetic and nothing else.

The countermeasure used here is a tag carried by every result separating what is entailed by construction from what could have come out otherwise, plus a discriminator committed in advance with at least two outcomes that would both be accepted. A test with only one acceptable outcome is not a test.

Status — in force across the project. Applied retroactively; several earlier results were reclassified downward.

Contact

contact@protocollogic.org

Correspondence welcome on any of the above — particularly disagreement, and particularly from people who work on substrate simulation, model validation, or the epistemics of computational neuroscience.

Not looking for funding or clients. If you think one of the entries above is wrong, that is the most useful message you can send.