The screening funnel

The protocol begins with brute force: 294,912 MLP neurons, each up and down at dose ±1. That makes 589,824 single-neuron configurations. Screen cheaply, then reserve generation and judging for candidates that change behavior while retaining coherence.

1. T0 · KL screen. A cheap KL-divergence screen compares steered and unsteered next-token distributions on fixed text, without generating replies. The goal is to screen every single neuron in both directions, using KL and perplexity to separate negligible changes, promising shifts, and severe disruption.
2. T1 · Short probes. Four to six short prompts, capped at 96 tokens, test coherence, degeneration, and behavior changes with Rune. Promotion calls for a change beyond the unsteered null spread while retaining coherence.
3. T2 · Mini-suite. A 14-prompt mini-suite uses Rune and GLM judgments. The planned interest score weighs behavior shift against retained coherence; the top slice advances.
4. T3 · Full examination. Run all 67 prompts with Rune and GLM judgments, then describe and verify candidates for a working neurotype label. A name is an annotation, not a finding by itself.

Protocol status: these stages describe the plan; the records below establish progress. Thresholds await calibration. The pair stage would combine promoted singles to look for effects neither single produces. Stored decisions can be reconsidered under a new policy without rerunning the specimen.

T0 · procedure log t0-v0.1

Started 2026-10-27 04:39:16 UTC · dose 1.0 · 32 reference tokens · seed 1. Configurations are screened in random order. These figures describe the rows recorded so far; they do not establish completed coverage.

Distribution of disturbance

Aggregates cover 47,936 screened rows (refreshed about every 30 seconds). KL is the mean KL(unsteered || steered) over the reference answer tokens. KL measures a distribution shift, not a behavioral diagnosis.

KL · frequency plate

Log-scale KL bins (10 per decade); bar height is log(1 + count). Hover a bar for its exact count.

Layer plate · median KL

Each cell is a layer: layer number, median KL (brighter is higher, scaled across layers), and rows screened so far. Medians are approximate (10 bins per decade).

Triage bands provisional

“Quiet” is relative to each layer’s non-broken rows, counted in whole histogram bins; it is not an absolute KL floor. “Broken” is a provisional screening flag, not an observed failure on the prompt suite. These first cutoffs need calibration against T1 results.

Steer lens · candidate plates

Shared token stems are leads for inspection. They do not establish a concept or a behavioral trait. Filters: KL 0.3 to 2.5, top-1 agreement at least 0.45, and at least 3 of the top 8 tokens sharing a 4-letter stem with the first. Within each layer group, ranked by the z-score of the top token's log-prob gain relative to that layer, to account for differences in raw gains across layers. Tokens here are distribution measurements, not generated replies. provisional heuristic

early (L0-7)

middle (L8-23)

late (L24-31)

Severe disturbance · highest KL

The highest-KL rows retained by the screen. Inspect before promotion: a large distribution shift alone does not establish incoherence.

Recent ledger · 300 screened rows

Promotion ledger · recorded decisions