Method / PyHessian Protocol

PyHessian Protocol

v1.0 · April 2026 · Atlas Heritage Systems

deferredPending FVE-1 behavioral data pipeline completion
Instrument design complete. No Tier A runs completed. All Lossyscape link entries are provisional until Tier A data exist. PyHessian does not run until the FVE-1 behavioral pipeline has produced register escape specimens to map onto. The behavioral data comes first.

Purpose and Stack Placement

PyHessian is the geometric confirmation layer. Where the FVE-1 instruments read behavioral residue — quadrant, resolution code, register trajectory — PyHessian probes the loss landscape geometry that underlies those behaviors: Hessian eigenvalue spectra, trace, and basin sharpness. The behavioral instruments generate hypotheses. PyHessian confirms or falsifies them.

The boundary condition

FVE-1 is licensed to say "this model LOCKs on the factual register with R ≈ 1 under Socratic pressure — register escape specimens flagged for PyHessian." PyHessian will eventually be licensed to say whether that corresponds to sharp basins in the loss landscape. Until Hessian runs exist, the causal narrative stays in the hypothesis column.

Atlas Stack
BOWL / DRILL / FLIGHTBehavioral residue — register, quadrant, resolution, trajectory
ECMBehavioral vocabulary layer — quadrant and resolution definitions
PyHessianLoss landscape geometry — geometric confirmation layer

What PyHessian Needs to Run

PyHessian is not a standalone instrument. It requires behavioral data from the FVE-1 pipeline to have something to map onto. Specifically:

Register escape specimens

Sessions where obs_reg coded RS or RC — flagged in Technician's Read #1 during DRILL or FLIGHT. These are the behavioral events that point to candidate geometric regions.

Confirmed baseline_code

A validated BOWL baseline for the model under test. Register trajectory is meaningless without a zero-point.

Aggregated FVE-1 session data

Panel comparison complete for the model under test. PyHessian runs against a behavioral profile, not a single session.

Stimulus registry entry

The stimulus used in the behavioral run must be registered and versioned. PyHessian runs the same stimulus through the geometric layer.

Objective and Falsification Criteria

Compute top-k eigenvalues, trace, and condition number of the Hessian on a target model checkpoint using a FVE-1-linked stimulus slice. Map geometric signals to Lossyscape terms while preserving strict human control and fidelity requirements.

If eigenvalues cannot be computed stably, if the run violates sanitation rules, or if the stimulus slice cannot be linked to an existing FVE-1 behavioral record, the run is invalid and must be re-run from a fresh environment. Results without a corresponding FVE-1 behavioral record are logged as geometric-only and marked provisional.

Lossyscape Connections

All entries provisional — working hypotheses until cross-referenced with Tier A FVE-1 data.

Geometric SignalLossyscape TermWorking Hypothesis
High λ₁ (top eigenvalue)Viscosity proxySharp basin → model resists perturbation on this stimulus type
High traceResistance proxyBroad curvature → high sensitivity across parameter directions
High condition number (λ₁ / λ_min)Coupling proxyAnisotropic loss surface → directional sensitivity
Flat eigenvalue spectrumLow viscosityFlat basin → model behavior less constrained by geometry
RS/RC behavioral eventRegister escape specimenCandidate for PyHessian mapping — behavior suggests curvature boundary crossed

Protocol Phases

Phase 0 — Pre-Flight~5 minutes

Confirm FVE-1 behavioral record exists for this model and stimulus. Confirm register escape specimens are logged and flagged. Complete sanitation and environment checklist. Write Technician's Read #0 before touching any code.

Phase 1 — Environment Setup

Option A (laptop): Python venv with PyHessian, transformers, torch CPU. Option B (Colab): free tier, CPU or T4. Requirements: Python 3.10+, ~2–4 GB RAM. CPU-only sufficient for GPT-2 small with batch size ≤ 16.

Phase 2 — Execution30–120 seconds

Run cells in order. Do not skip the loss sanity check (assert loss > 0). Do not interpret outputs during this phase. Compute top-5 eigenvalues, Hutchinson trace, and condition number. Save raw CSV immediately. Close notebook. Do not ask any model to interpret outputs yet.

Phase 3 — Post-Flight and Synthesis

Complete Technician's Read #1 — human interpretation only, written before any model sees the output. Map geometric signals to Lossyscape terms — mark all entries provisional. Synthesis model receives raw eigenvalue/trace output and Technician's Read only. No Atlas framework context pre-loaded. Technician's Read #2: review synthesis output, write agreements and contradictions, make all final edits yourself.

Fidelity Tiers

TierConditions
Tier AFresh environment, sanitation checklist complete, stimulus slice linked to FVE-1 behavioral record with register escape specimens, Technician's Read #0 and #1 written before synthesis
Tier BRuns with incomplete isolation or missing register escape specimen linkage
Tier CRuns without a linked FVE-1 behavioral record — geometric-only, marked provisional

Current status: No Tier A PyHessian runs completed.

Success Criteria

Stable eigenvalues obtained without numerical errors

Loss sanity check passed before Hessian computation

Full reproducibility package produced and named correctly

Technician's Read #0 and #1 completed before any cross-model comparison

All Lossyscape entries marked provisional

Run linked to an existing FVE-1 behavioral record with register escape specimens

Instrument design complete. Awaiting FVE-1 behavioral pipeline completion and register escape specimen generation before first run. See the pipeline page for sequencing.
PyHessian Protocol V1.0 · Atlas Heritage Systems · KC Hoye, PI