@dualcondition/scorer

Dual Condition T1/T2 prompt scorer — the core scoring engine for the SNP framework.

Decomposes any prompt into Template Specificity (T1) and Generative Headroom (T2) using four independent signal channels. Runs entirely client-side. Zero API calls. Zero dependencies.

Install

npm install @dualcondition/scorer

Usage

const { score } = require('@dualcondition/scorer');

const result = score(
  "You are a theoretical physicist with a background in continental philosophy. " +
  "Examine the concept of 'negative space' across sculpture, jazz, and gravitational lensing. " +
  "Do not summarize — find the structural invariant."
);

console.log(result.t1);           // 0.72 — Template Specificity
console.log(result.t2);           // 0.81 — Generative Headroom
console.log(result.interaction);  // 0.583 — T1 × T2
console.log(result.zone);         // "Optimal zone"
console.log(result.recommendation);

What It Scores

T1 (Template Specificity): How precisely the prompt activates deep patterns. Composed from three channels:

• Lexical specificity (35%): Domain vocabulary density across 8 fields
• Structural constraints (35%): Persona, format, audience, style specifications
• Meta-cognitive operators (30%): Reasoning directives, depth signals, epistemic constraints

T2 (Generative Headroom): How much creative freedom remains open.

• Verb classification (mechanical vs. generative)
• Output predetermination detection
• Open-ended framing indicators

Zones

| Zone | T1 | T2 | Meaning | |------|----|----|---------| | Deep activation | >0.8 | >0.8 | Multi-layer activation, maximum freedom | | Optimal zone | >0.6 | >0.6 | Both conditions met | | Overconstrained | >0.6 | <0.4 | Specific but no room to generate | | Underspecified | <0.4 | >0.6 | Freedom but no direction | | Dead zone | <0.4 | <0.4 | Neither condition met |

API

score(text: string): DualConditionResult

Returns:

• t1 — Template Specificity (0–1)
• t2 — Generative Headroom (0–1)
• interaction — T1 × T2
• zone — Classification string
• signals.t1 — Array of T1 signal descriptions
• signals.t2 — Array of T2 signal descriptions
• signals.warnings — Array of warning strings
• recommendation — Actionable recommendation string
• channels — Detailed per-channel scores (for debugging)

Architecture

This package is the shared engine across three Dual Condition products:

• Free web tool (dualcondition.com) — imported directly
• Enterprise API (api.dualcondition.com) — used as the heuristic tier
• Agent SDK (@dualcondition/sdk) — local scoring tier

One engine, three products. Changes here propagate everywhere.

Testing

npm test

Runs the five calibration prompts from the launch plan. All must pass before deployment.

License

MIT

Reference

Schlegel, S. (2026). Spontaneous Narrative Projection: A Formal Model. OSF Preprints. https://osf.io/nj4vf