4.1 KiB
QPI Case Preprocessing Workflow
Date: 2026-06-17
Status: reusable draft workflow for owner review
1. Purpose
Convert owner-provided raw materials into Markdown QPI case drafts that the owner can confirm or correct.
This workflow exists because the owner should not need to write JSON or fill a heavy form before case extraction begins.
2. Inputs
Input is a raw Markdown or text source supplied by the owner.
Examples:
- long reflective analysis;
- process redesign notes;
- GPT discussion records;
- QPI case analysis documents;
- model extraction or workflow failure records.
Do not copy the full source into this repository unless the owner explicitly asks. Record the source path and extract only concise evidence snippets or summaries needed for case review.
3. Output Location
Write draft cases under:
reports/model_case_preprocessing/qpi/<round-id>/
Use one output file per raw source:
<short-source-slug>.cases.md
Each file must be marked:
status: draft_owner_review_needed
4. Case Draft Format
Each source file may produce multiple cases.
Use this structure for every case:
## Case qpi-draft-XXX: <short title>
status: draft_owner_review_needed
source_path: `<absolute source path>`
### 1. Surface Problem
What the problem appears to be at first glance.
### 2. Subject Position
Who is framing the problem, including role, experience level, responsibility scope, and whether the subject is the present owner, future owner, an agent, a team, or another proxy stakeholder.
### 3. Scenario Context
Where the problem appears, including workflow, repo, article, relationship, tool, organization, or life context.
### 4. Expectation-Reality Gap
- Expected:
- Reality:
- Gap summary:
### 5. Attempted Paths
What has already been tried or assumed.
### 6. Dynamic Shift
How the problem frame changes over time, such as Q -> P, P -> I, Q/P -> P/I mixed, or insufficient context -> multi-perspective.
### 7. Scarcity Profile
- data_scarcity:
- path_or_resource_scarcity:
- consensus_or_order_scarcity:
Use `high | medium | low | unknown`.
### 8. Mixed Or Multi-Perspective
Choose one:
- `intra_frame_mixed`
- `inter_viewpoint_divergence`
- `not_mixed`
- `unknown`
Explain why.
### 9. Governance Load
Does the case require ongoing tradeoff, role boundary management, audit authenticity, downstream reuse, state continuity, or human confirmation?
### 10. Misframing Risks
Check any that apply:
- violent_reduction
- malicious_inflation
- tool_solutionism
- premature_classification
- single-cause_reduction
- over_pathologizing
- other
### 11. Candidate QPI Judgment
- classification_scope:
- is_provisional:
- classification:
- dominant_scarcity:
- classification_confidence:
- recommended_next_step:
### 12. Owner Review Questions
List concrete questions the owner should confirm or correct.
5. Extraction Rules
Preserve subject-context before classification.
Do not classify a sentence in isolation if the source provides broader context.
Separate:
intra_frame_mixed: same subject, same scene, same stage, multiple scarcities.inter_viewpoint_divergence: different subjects or responsibility positions would classify the same surface issue differently.
Do not equate issue with "many people". A one-person workflow can be an Issue if it has ongoing governance load, proxy stakeholders, state continuity, audit boundaries, or unstable success criteria.
Do not turn psychological, relational, or workflow material into clinical claims. Use QPI terms: subject, context, gap, scarcity, frame shift, misframing risk.
6. What Not To Do
Do not:
- write selector JSON directly;
- update regression cases directly;
- upgrade model status;
- create a third model;
- summarize the whole source instead of extracting cases;
- quote long source passages;
- pretend draft cases are owner-approved.
7. Promotion Path
The promotion path is:
raw source
-> Markdown case drafts
-> owner review and correction
-> case digest JSON
-> selector calibration inputs or regression cases
Only owner-reviewed cases may become machine-readable calibration or regression assets.