Research

How do you produce a finding that can change a decision and improve the knowledge system?

Research is not complete when sources have been collected. It is complete when a claim has a provenance trail, a confidence boundary, a relationship to existing knowledge, and a named implication for action.

Output Contract

Start with one decision question. Finish with one research record:

Field	Required answer
Decision	What decision, belief, or action could this research change?
Claim	What is the one smallest decision-relevant finding?
Sources	Which primary and contrary sources support or challenge it?
Confidence	What is demonstrated, inferred, disputed, or unknown?
Canonical owner	Which page or governed research surface owns the finding?
Retrieval cue	Which future situation should retrieve this finding?
Model update	Which belief strengthens, narrows, dies, or remains unresolved?
Review trigger	What new evidence or date should reopen the finding?
Next action	What should a person or agent do differently now?

The Core Problem

Most research systems re-discover. Every time you return to a topic, you re-read, re-find, re-derive. Information exists but knowledge does not accumulate. The time cost is constant — no matter how many times you have visited the domain.

The alternative: design your research system so each new source integrates into what you already know — updating summaries, flagging contradictions, strengthening or challenging existing synthesis. Knowledge compiled once is available instantly on the next question. Knowledge re-discovered is expensive forever.

1. Bound the Question

Write the decision before the search. A broad topic such as “knowledge management” has no stop condition. A bounded question such as “Which metadata fields help an agent distinguish current guidance from historical evidence?” tells you what evidence matters and when to stop.

2. Seek Multi-angle Coverage

A single research pass produces a single perspective. Investigate any topic from at least four angles:

Angle	Question
Academic	What does evidence say?
Practitioner	What works in practice?
Contrarian	What is the strongest case against?
Historical	What has been tried before?

The contrarian angle is the most skipped and most valuable. The strongest objection to your thesis is the one most likely to be true. Finding it early prevents compounding on a false foundation.

3. Test the Finding

Three questions before a note becomes permanent:

1. Does this change what you would do, believe, or decide?
2. Does it contradict something you already hold? Resolve the contradiction — do not file both.
3. Would it survive being explained to someone unfamiliar with the domain?

Material that fails all three can remain source information. It has not yet earned a place as accepted knowledge. The goal is a smaller collection of higher-quality findings, not a larger archive of everything read.

4. Bound Source Trust

Publication speed is not credibility. For any source:

• What are the author's incentives?
• What would need to be true for this to be wrong?
• Has the claim survived contact with contrary evidence?

Primary sources — original research, first-hand accounts — degrade slowly. Secondary sources — summaries, analyses — degrade fast. When in doubt, trace back to the primary source and check when the secondary account last engaged with contradictory evidence.

5. Integrate, Do Not Merely File

Research ends when you can teach the finding and show where it changes the existing map. Until you can explain it to someone unfamiliar with the domain, distinguish it from neighboring claims, and name its next use, you have stored information—not synthesized insight.

Use the Knowledge Schema gates before integrating it:

1. Ontology: Which entity, state, or relationship does the finding describe?
2. Epistemology: What evidence permits the claim, at what confidence and freshness?
3. Taxonomy: Which domain, reader job, artifact type, and lifecycle own it?
4. Nomenclature: Which canonical term and aliases let people and agents retrieve it?
5. Axiology: Which decision or action becomes clearer because it exists?

If a gate is unresolved, keep the material as research with its uncertainty visible. Do not promote it into doctrine to make the map look complete.

Proof of Done

• A cold reader can restate the finding without opening every source.
• The strongest contrary case is visible beside the claim it qualifies.
• The record points to one canonical owner and carries a retrieval cue rather than creating a parallel truth.
• The model update says what the evidence changes; a source collection without an update remains research input.
• The confidence and review trigger prevent a stale finding from masquerading as current knowledge.
• The next action is specific enough to observe.

Failure Modes

• Collection without decision: sources accumulate but no claim changes an action.
• Parallel truth: a finding creates another owner instead of updating the canonical one.
• False confidence: contrary evidence or freshness limits disappear from the record.
• Retrieval failure: the finding exists, but no future situation can recall it.

Context

• Meta-Learning Pipeline — Where research fits in the Awareness→Teach cycle
• Knowledge Schema — Map new domains fast using a reference anchor
• Note Taking — Capture and protect what research produces
• Matrix Thinking — Empty cells as research agenda
• Decision Frameworks — Convert research into decisions

Links

• Feynman Technique — Teaching as the final research filter
• How to Read a Paper — Multi-pass reading for depth

Questions

What is the most valuable insight you have synthesized recently — and where does it live now?

• Which of your research cycles has compounded over time — and which keeps restarting from scratch?
• At what point does research become a substitute for action rather than a foundation for it?
• What does your research agenda look like if the goal is synthesis, not collection?
• Which information source do you trust most — and have you validated why you trust it?

Changes my mind: a research record without one canonical owner, retrieval cue, confidence boundary, or model update repeatedly improves later decisions more reliably than this contract.

Next question: which current decision needs one finding rather than another collection of sources?