Research Framework
A controlled process for developing evidence that people can review.
What Is the Research Framework?
The DCPF Research Framework turns an information need into a deliberate investigation. It defines what must be learned, which evidence may be used, how sources and uncertainty should be handled, and how findings must be documented for human review.
Research produces a Research Output. That output is evidence for review, not automatically Accepted Knowledge. A person with the appropriate authority decides what to accept, revise, exclude, or return for additional investigation.
DCPF is experimental.
DCPF has demonstrated promise through development work, practical applications, and training practicums, but it has not been broadly or independently validated. Begin with controlled, low-risk Research, validate the approach for your domain and AI environment, independently verify consequential findings, and retain human judgment and accountability.
Research and Runtime Have Different Jobs
Research asks, “What does the project need to learn?” Runtime asks, “What must the system do now using approved Knowledge?”
Keeping these responsibilities distinct prevents an execution request from quietly searching for new facts, choosing sources, resolving conflicting evidence, or treating unreviewed information as though the project had already accepted it. When Runtime exposes a Knowledge gap, route that need back to Research rather than filling it silently during execution.
The Ten Standard Research Responsibilities
Every Research artifact should address these ten responsibilities. The answers may be brief for low-risk work and more formal for consequential, regulated, or technically demanding work.
1. Research Purpose
Define why the investigation is needed and how its findings are expected to support the project.
Teaching question: What decision, Knowledge need, or project objective will this Research support?
2. Research Questions
State the questions the investigation must answer with enough precision to guide evidence gathering.
Teaching question: What must be learned, clarified, compared, or verified?
3. Scope and Boundaries
Define included topics, excluded topics, time periods, populations, jurisdictions, domains, and stopping boundaries.
Teaching question: What belongs inside this investigation, and what does not?
4. Source Priority Hierarchy
Identify which source types should be sought first and how authority, proximity, expertise, and relevance affect priority.
Teaching question: Which sources should control when available, and what may be used when they are not?
5. Source Qualification and Exclusion
Define source-quality requirements and any prohibited, unsuitable, inaccessible, stale, promotional, or unverifiable sources.
Teaching question: What makes a source acceptable for this Research?
6. Evidence Collection and Documentation
Specify what source details, links, dates, quotations, data, notes, or provenance records must accompany each finding.
Teaching question: What must be recorded so another person can inspect the basis of the finding?
7. Conflict and Uncertainty Handling
Require the Research process to expose disagreements, gaps, limitations, ambiguity, and confidence rather than force false certainty.
Teaching question: How should conflicting or incomplete evidence be represented and investigated?
8. Analysis and Synthesis
Define how evidence should be compared, organized, interpreted, and separated from unsupported inference.
Teaching question: What reasoning is required, and how must evidence be distinguished from interpretation?
9. Research Output Requirements
Specify the required structure, coverage, citations, findings, limitations, unresolved questions, and presentation format.
Teaching question: What must the Research Output contain for a reviewer to evaluate it efficiently?
10. Review and Acceptance Handoff
Identify who reviews the output, which criteria apply, and how accepted portions move into the Knowledge process.
Teaching question: Who decides what happens to the findings after Research ends?
The Research Request Is the Operational Artifact
A Research Request is the artifact supplied to the AI to authorize and direct a specific investigation. It should contain the relevant answers to all ten responsibilities and the applicable Governance controls needed for that work.
The AI must have actual access to the complete Research Request and every source, file, or resource it is expected to use. Naming a file, linking to an inaccessible location, or referring to instructions from another conversation does not give the AI access. If required material is unavailable, the AI should stop, identify what is missing, and request it rather than claim to have used it.
A separate Governance artifact may be used for convenience, but it is not a required attachment or dependency. Put the controls the Research needs directly in the Research Request so the executable artifact is complete.
Evidence Does Not Become Knowledge Automatically
Completion of a Research Output does not change the status of its contents. The material remains evidence until an authorized person reviews it under the applicable Governance.
- Accept: approve all or part of the evidence for preservation as Accepted Knowledge.
- Revise: correct or qualify the material before acceptance.
- Research further: return gaps, conflicts, or weak support for additional investigation.
- Exclude: reject material that does not meet the project’s requirements.
- Preserve as unresolved: record uncertainty without treating it as settled Knowledge.
When a Research Responsibility Is Left Undefined
The AI or user may fill the gap with an unstated assumption. Undefined questions can produce irrelevant coverage. Undefined source rules can lead to weak or promotional evidence. Missing documentation can make findings impossible to verify. Unclear conflict rules can flatten disagreement into false certainty. An undefined handoff can allow unreviewed claims or hallucinations to enter later work as though they were approved.
DCPF makes these responsibilities explicit so omissions can be found and assigned before their consequences spread into Knowledge or Runtime.
Use Research Proportionately
A small factual question may need a short Research Request with a few authoritative sources. A high-impact decision may require formal search methods, source logs, specialist review, validation, and documented acceptance. DCPF requires the responsibilities to be addressed, not every project to use the same amount of process.