Research-to-Artifact System

A repeatable process for turning ambiguous questions into decision-ready evidence.

Reusable framework, implemented and refined across multiple research efforts. A methodical research approach for business problems and personal questions alike, built to surface curated evidence from credible sources.
View the operating model Representative studies
At a glance

A 10-stage research lifecycle, from problem definition through quality audit to final delivery. Skipping a stage requires documenting why.

A 7-level decision authority hierarchy resolves every judgment call, from explicit instructions down to a bounded, documented assumption.

The convergence loop that produced this page ran 13 rounds before a confirmation audit found nothing left to fix.

The problem

What ordinary LLM research gets wrong

Jumps from a vague question straight to a polished answer, skipping evidence collection entirely.
Treats its own training data as a source, instead of something that needs a citation.
Confuses a popular opinion with a validated finding, or one good example with proof it generalizes.
Turns a recommendation into a hard requirement without flagging that a judgment call happened.
Delivers a well-written report with no way to trace a claim back to where it actually came from.
Mistakes a completed draft, or a single finished pass, for a completed research process.
Operating model

The process

REPEATED UNTIL CONVERGED Intake & alignment Methodology selection Evidence & source discipline Pattern extraction & disconfirmation Convergence Human review & handoff Δ=0 IF Δ>0, THEN repair, repeat
A tinted zone marks the repeated stages; the accent arc is the only thing that means "try again."
Method selection

Methodology matched to context

Artifact-producing problem

Design Science Research Methodology governs by default, moving through problem definition, design, demonstration, evaluation, and communication.

Evidence-mapping-only question

A scoping/evidence-synthesis discipline governs instead, without claiming formal systematic-review status unless every requirement is actually met.

Causal question

Neither method applies alone. A causal, experimental, or systematic-review approach is required, and the process says so instead of forcing a fit.

Method turns out to be wrong mid-stream

The method gets replaced. A mismatched method doesn't get carried to the finish line for the sake of momentum.

Evidence & claim discipline

Every source is graded, every claim is classified

A

Primary, authoritative, or peer-reviewed evidence.

B

Strong external/institutional evidence or employer-syndicated primary content.

C

Practitioner guidance or a structured public artifact.

D

Discovery only, raising questions without establishing conclusions.

Source-supported fact Observed finding Estimate Inference Design requirement Design hypothesis Unknown

Four collapses this process is built to prevent:

  • Observation collapsing into causation.
  • Popularity collapsing into validation.
  • Example quality collapsing into effectiveness.
  • A recommendation collapsing into a requirement.
AI & human boundaries

Where the model stops and I start

AI supports

  • Discovery and query generation
  • Extraction and coding suggestions
  • Synthesis across sources
  • Contradiction checks
  • Drafting
  • QA and visual planning

AI may not

  • Treat its own memory as evidence
  • Fabricate support for a claim
  • Silently change methodology mid-stream
  • Hide uncertainty to sound more confident
  • Infer an outcome the data doesn't support
  • Convert my preference into a research fact
  • Make the final approval decision

Human authority retains

  • Defining and approving the decision problem
  • Approving consequential assumptions or changes
  • Owning how the final artifact gets used
  • Auditing for correctness
  • Determining human readability and usability
  • Redirecting further passes based on misdirected outputs or new research questions
  • Final decision authority
Decision authority hierarchy
  1. Explicit instructions
  2. Controlling project authorities
  3. Locked objectives and constraints
  4. Strongest applicable evidence
  5. Reasoned inference
  6. Bounded, documented assumption
  7. Escalation, if a blocker exists
What counts as a blocker

A blocker exists only when a decision materially affects validity or usefulness, the available evidence can't resolve it, more than one defensible path remains, and proceeding on an assumption would compromise the research. Everything short of that is resolved through the hierarchy above.

Disconfirmation & convergence

Execute, find the fault, repair, repeat

A convergence loop runs the same review pass repeatedly, finding and repairing faults or gaps until the delta between passes reaches zero. This page is itself a product of that loop. The same research framework it describes governed how this page's own content and structure took shape.

Convergence record excerpt

Four of the thirteen convergence rounds behind this page.

Round 1
Fault: "Non-code" only described an absence. It named no actual capability. Repair: reframed around Workflow, Decision, and Research/Intelligence system jobs. Delta: high.
Round 4
Fault: a planning case study risked reading as career-advice documentation. Repair: centered the decision architecture, controls, authority, and execution transition instead. Delta: high.
Round 9
Fault: a minimalist homepage risked hiding credibility. Repair: adopted layered progressive disclosure with visible status/evidence cues. Delta: moderate.
Round 13
Confirmation audit. No new substantive structural fault found. Delta: zero. Convergence reached.
Representative studies

Three different problems explored

Outcome & limits

What this proves, and what it doesn't yet

This research framework compiles findings through a transparent method, built specifically to inform strategy development and planning. It runs through intake, methodology gate, evidence discipline, convergence, and human review, and it has now been refined across the three studies above.

The framework is inherently limited by the nature of open-source information, and it can be further refined through individual use once hard data becomes available and connected. I'm also refining small details in the final PDF format, slowly minimizing the routine edits that can be defined before the process even begins. Any research here should be read as signal. Actual business decisions or assumptions still need multiple sources behind them before they're considered confident.

The framework document and full source research for every study are on GitHub.