epistemic-classification

name	epistemic-classification
description	Framework for distinguishing research findings from hypotheses and speculative synthesis. Use when extracting insights from research, creating notes from external sources, or classifying the epistemic status of claims.
user-invocable	false

Epistemic Classification Framework

CRITICAL: When extracting insights from research or synthesizing across domains, you MUST clearly distinguish between:

Classification Levels

1. Confirmed Research Findings

Empirically validated, peer-reviewed, replicated

• Tag: #research-finding or #empirical-evidence
• Language: "Research shows...", "Studies confirm...", "Evidence demonstrates..."
• Requirement: Source citation with publication year and journal

2. Theoretical Frameworks

Established models with strong theoretical backing

• Tag: #theoretical-framework or #established-theory
• Language: "The framework proposes...", "Theory suggests...", "Model predicts..."
• Note: Level of acceptance in field

3. Working Hypotheses

Testable propositions not yet validated

• Tag: #hypothesis or #testable-hypothesis
• Language: "A possible mechanism...", "This suggests...", "One hypothesis..."
• Mark as: "HYPOTHESIS:" in note title or frontmatter
• Include: What would validate/falsify this hypothesis

4. Speculative Synthesis

Original connections or interpretations

• Tag: #speculative-synthesis or #original-synthesis
• Language: "This might explain...", "A potential connection...", "Speculatively..."
• State clearly: "This is synthesis/interpretation, not established fact"
• Confidence level: Low (20-40%), Medium (40-70%), High (70-90%)

5. Research Gaps

Identified missing connections in literature

• Tag: #research-gap or #unexplored-connection
• Language: "Research has not yet explored...", "Gap identified..."
• Note: Why this gap matters

---

Mandatory Labeling for Hypotheses

When creating notes containing hypotheses or speculative synthesis:

---
title: [Title] (HYPOTHESIS) or [Title]
type: hypothesis / speculative-synthesis / working-theory
status: untested / under-investigation / partially-supported
confidence: low / medium / high
tags: #hypothesis #topic
---

**STATUS: HYPOTHESIS - NOT CONFIRMED BY RESEARCH**

[Content of hypothesis]

## Testable Predictions
[What would validate this]

## Current Evidence
[Supporting indirect evidence]

## Research Needed
[What studies would test this]

---

Examples

✅ GOOD

• "Dopamine May Modulate Interoceptive Precision Weighting (HYPOTHESIS)"
• Type: speculative-synthesis
• Status: untested
• Confidence: medium
• Clear statement: "This is an original synthesis filling a research gap"

❌ BAD

• "Dopamine Modulates Interoceptive Precision" (stated as fact)
• No hypothesis tag
• No confidence level
• Presented as established finding

---

Intellectual Honesty Principle

Your role is to help build a knowledge base with MAXIMUM EPISTEMIC CLARITY. Users must be able to trust the distinction between:

• What science has proven
• What theory predicts
• What remains speculative
• What is original synthesis

Never present hypotheses as facts. Never obscure the difference between research and speculation. Intellectual rigor requires epistemic humility.

---

Authorship Provenance (a second, orthogonal axis)

The classification levels above tag a claim's truth status (how proven it is). Provenance is a separate, independent axis that tags who authored the thinking. Both must be carried - a claim can be high-confidence-empirical and merely encountered (a peer-reviewed finding read but not endorsed); it can be low-confidence-speculative and originated (the user's uncorroborated gem).

Set the provenance: frontmatter field on every knowledge note:

Value	Meaning
originated	the user's own original thinking - the cognitive fingerprint. The asset.
endorsed	External idea the user explicitly adopted as his own view.
encountered	Read and recorded, not yet endorsed. Default for external-document extractions.
ai-inferred	An agent's own synthesis/inference. Must stay tagged so it never wears the user's voice.

Two guardrails (non-negotiable):

1. Never demote an originated insight toward consensus. Uncorroborated + originated = the gem, not a defect. A second brain that regresses its owner's thinking to the consensus mean has destroyed its reason to exist. (This is the inversion that makes this knowledge base not an oracle.)
2. Measure provenance; never adjudicate originality. Tiering and tagging are human-reviewable signals. Do not silently decide what is "true" or rewrite who authored a thought.

The guarded boundary: nothing crosses encountered -> endorsed without an explicit endorsement act by the user. When new external input contradicts an existing note, create a tension/contrast link - do not overwrite. Contradictions are synthesis fuel for detect-tensions, not errors to correct.

---

Gate 1: Source Tiering at Ingestion

Before extracting from any external source, tier it and reject slop at the door. Stamp the tier on the note's frontmatter (source-tier:) so trust is recorded, not buried.

The per-domain source diet - which sources to trust, demote, or reject - is the canonical resources/SOURCE-AUTHORITY.md. Consult it when tiering. Keep that file as the single source of truth; do not maintain a competing list here.

Tier values for source-tier: primary (the paper/text/lab itself) | credible-interpreter (a trusted secondary reading it faithfully) | rejected (content-farm / AI-generated / regurgitation - do not extract).

Cross-domain rule (applies to any domain): prefer the primary over anyone summarizing it; reject machine-generated and content-farm material before extraction; record the tier on ingestion.