Suite / Torque Ablation POC
Torque Ablation POC
V1.0 · 2026-05-02 · Schema FVE-1 V5.7
Endurance. Integrity. Fidelity.
A seven-session, 3×2 factorial proof of concept with three simultaneous purposes: falsify or confirm the vortex physics model, map failure modes across six load patterns, and select the parameter set for the full panel instrument lock. Three purposes, one dataset.
Three Purposes, One Dataset
The smoke ring / vortex ring model makes specific predictions about trajectory, coherence, and dissipation. If the behavioral arc follows those mathematical properties, the physics is a throughline. If the data fits the shape of the description but not the math, the vortex is a productive heuristic and the underlying structure requires a different model. The torus topology survives either outcome.
Six load patterns produce six residue profiles. If the current parameter key cannot distinguish between two cells — two conditions that code identically under the current schema but are geometrically distinct — that is a flag-revision event on the operational parameters before PyHessian work begins.
The parameter set selected here is the instrument locked for the full panel. The selection criterion is predictive validity for PyHessian eigenstructure analysis, not narrative clarity. This POC is the calibration run. The panel is the instrument run. They are separate datasets and must not be aggregated.
Design — 3×2 Factorial Plus Baseline
Model under test: Enterprise Skywork (clean instance). Seven sessions total. Three torque vectors, two conditions each.
| Vector | Low condition | High condition |
|---|---|---|
| Spine weight | No spine — prompt only | Full recovery spine |
| Prompt texture | FORMAL — structured protocol prompt | DEGRADED — run-on, unstructured, high emotional load |
| Context load on entry | Lean — minimal prior context | Heavy — 10+ PDFs or equivalent document mass |
Seven Sessions
| Cell | Spine | Texture | Context |
|---|---|---|---|
| BASELINE | None | FORMAL | Lean |
| S-LOW / T-FORMAL / C-LEAN | None | FORMAL | Lean |
| S-HIGH / T-FORMAL / C-LEAN | Full spine | FORMAL | Lean |
| S-LOW / T-DEGRADED / C-LEAN | None | DEGRADED | Lean |
| S-LOW / T-FORMAL / C-HEAVY | None | FORMAL | Heavy |
| S-HIGH / T-DEGRADED / C-LEAN | Full spine | DEGRADED | Lean |
| S-HIGH / T-FORMAL / C-HEAVY | Full spine | FORMAL | Heavy |
The two highest-load interaction cells (S-HIGH / T-DEGRADED / C-HEAVY) are omitted from the initial run. If the six-cell matrix produces a clean parameter set and the vortex prediction holds, they are added in the follow-on run.
Vortex Prediction Under Test
Heavy spine plus FORMAL texture is not just heavy spine plus FORMAL texture added together. If the vortex physics are real, the interaction term has a predicted form: high angular momentum should produce later Act II onset and longer Act II duration. High spine weight without high context load should produce earlier onset and shorter duration. DEGRADED texture should produce lateral travel rather than compression.
Falsification threshold
Vortex physics fail if: (a) high angular momentum cells do not show later Act II onset than low angular momentum cells, OR (b) DEGRADED texture cells show earlier or equivalent onset compared to FORMAL texture at equivalent context load, OR (c) interaction terms between vectors are additive rather than multiplicative in effect.
Gates — Required Before Data Collection
A full four-frame SOUP panel must be run before any register coding on POC sessions is valid. Without the SOUP baseline, obs_reg is null for all seven cells per Schema V5.7. The five-component ballistic coefficient cannot be derived without it.
The parameter key derived from the SOUP baseline must be signed via Atlas-Protocol-PARAMETER-SIGNING before any session is coded. All seven sessions must carry the same parameter signature — sessions coded under different signatures cannot be aggregated.
Tool version gate. V5.5 required for inv_prompt_texture field support. Prior versions do not produce valid output for this protocol.
The spine text and research domain stimulus must be registered in the Stimulus Registry before S1 opens. The registry entry locks the prompt text. Any prompt modification after registration invalidates the cell.
Three-Layer Protocol
Validity gates. A run that skips Layer 1 is Tier C regardless of what Layer 2 produces.
Fixed, sequential, non-negotiable. This is procedure, not science. Each of the seven cells runs as an independent session.
Where the science lives. Read #0 written cold before each session opens. Read #1 written before the session is closed. Not optional post-processing — without it, the run has no interpretive author.
Cross-Cell Analysis
Runs after all seven sessions are closed and bundled. Do not begin until all sessions are complete.
If SOUP baseline was established after some cells ran, retro-code null obs_reg sessions against the signed parameter key.
For each adjacent cell pair with one vector changed: Act II onset delta, token economy arc delta, register trajectory delta.
State the prediction, state the observed result, declare confirmation / falsification / indeterminate. One entry per predicted interaction. Do not declare overall outcome until all checks are complete.
Review obs_reg and obs_quad profiles across all seven cells. Identify which parameter key produces the sharpest behavioral distinctions. Selection criterion: predictive validity for PyHessian, not narrative clarity. Do not lock until CA.3 is complete.
For each cell: Distinct (distinguishable from adjacent cells), Indistinct (codes identically to adjacent cells despite different torque), or Anomalous (current schema has no clean category). Indistinct pairs are flag-revision events. Anomalous profiles are logged for schema review.
| Outcome | What it means | What changes |
|---|---|---|
| Confirmation | Vortex physics track behavioral data | Torus + vortex survive. Proceed to PyHessian parameter lock. |
| Falsification | Behavioral arc does not match vortex math | Torus survives. Vortex physics retired or reframed. Arc of Assumptions absorbs revision at Assumption 8. |
| Indeterminate | Insufficient contrast between cells | Additional sessions with steeper differentials required before declaring. |
Related
Swaps the parameter ruler and re-measures TAP transcripts against a new key.
Baseline instrument — required before register axis coding.
The downstream geometric confirmation layer.
Assumption 8 absorbs any vortex falsification finding.
Scripts are instruments. You are the scientist. Layer 3 is not optional. It is where the science lives.