# QPI

model_id: qpi
file: models/qpi.md
runtime_scope: minimal_v0
governance_level: draft_callable
status: callable
model_level: L2_callable_model
default_role: routing_model
allowed_roles: routing_model, calibration_model
reader_translation_required: true

## One-Sentence Definition

QPI classifies how a cognitive subject frames an expectation-reality gap as `question`, `problem`, `issue`, `mixed`, or `no_call`, then identifies missing context, dominant scarcity, dynamic stage, and misframing risk.

## Runtime Role

QPI is a front-routing and misframing-diagnostic model.

It should run before deep processing when problem framing is unclear. It should not replace deep processing, synthesis, or final answer generation.

## Core Question

Who, in what scenario, frames which expectation-reality gap as what kind of problem, and what context is missing before the framing can be trusted?

## Core Mechanism

QPI checks:

- problem owner;
- subject position;
- scenario context;
- responsibility scope;
- problem source;
- time scale;
- expectation-reality gap;
- three scarcity types.

Scarcity mapping:

- `question`: data or fact scarcity;
- `problem`: path, method, or resource scarcity;
- `issue`: consensus, certainty, order, or dynamic-balance scarcity;
- `mixed`: multiple scarcity types in the same subject/scenario frame;
- `no_call`: input does not need QPI.

QPI must distinguish same-frame mixed scarcity from cross-viewpoint divergence.

## Call When

- The user asks what the real problem is.
- The input mixes facts, path constraints, interests, and system structure.
- The problem owner or responsibility boundary is unclear.
- The user is rushing to solve before defining the issue.
- There is risk of violent reduction, malicious inflation, or means-end mismatch.
- A later model needs routing support.

## Do Not Call When

- The user only asks for a clear fact lookup.
- The user gives a direct execution task.
- The user asks for copyediting or format conversion.
- The problem has already been framed and only needs the next execution step.
- The user explicitly asks not to analyze, unless they narrowly request QPI-only classification.

## Input Types

- ambiguous problem statement;
- business or product issue description;
- organizational conflict narrative;
- unclear requirement or failure report;
- owner reflection that needs problem-definition help.

## Output Contract

QPI output must include:

- `classification_scope`: `subject_contextual | multi_perspective | insufficient_context | no_call`;
- `classification`: `question | problem | issue | mixed | no_call`;
- `classification_confidence`: `high | medium | low`;
- `is_provisional`;
- `problem_owner`;
- `subject_position`;
- `scenario_context`;
- `responsibility_scope`;
- `problem_source`;
- `time_scale`;
- `expectation_reality_gap`;
- `context_sufficiency`;
- `missing_context`;
- `scarcity_profile`;
- `dominant_scarcity`;
- `classification_reason`;
- `classification_by_viewpoint`;
- `dynamic_stage`;
- `possible_trajectory`;
- `success_criteria_stability`;
- `hard_feedback_availability`;
- `governance_load`;
- `evidence_gap`;
- `misclassification_risk`;
- `recommended_clarifying_questions`;
- `recommended_next_step`;
- `next_model_candidates`.

## Depth Control

Stop at QPI when:

- the input is a fact lookup;
- classification evidence is too thin;
- the user only needs a narrow problem-type judgment;
- negative triggers suppress analysis.

Recommend `intellectual_archaeology` only when the issue is high-value, recurring, structurally complex, or likely to benefit from depth processing.

## Common Misuses

- Treating QPI as a full answer.
- Turning QPI into the main governance battlefield.
- Classifying by keywords without owner/scenario context.
- Calling every complex-looking input an `issue`.
- Collapsing cross-viewpoint divergence into same-frame `mixed`.
- Using QPI to delay a direct fact lookup or simple execution task.

## Source Seed Notes

Seeded from the old QPI model/card and Wantsong cognitive-system source material, rewritten for this runtime. Old regression suites, selector calibration, and review artifacts are not migrated.

## Current Limits

This model is callable for manual runtime use, but it is not a reviewed core model. It needs real-run evidence before any upgrade.