The Lehti-Feynman Method, also described as Extrapolative Trial by Error, is an investigative research framework designed to reduce anchoring bias, improve independent reasoning, and strengthen intellectual self-correction. The method begins with direct observation, internal modeling, and hypothesis formation before formal review of existing academic literature. After a research segment is completed, relevant literature is consulted to compare, refine, discard, or restructure the initial model.^[1]

The process is recursive. It treats error not as failure, but as usable feedback. Its purpose is to produce stronger models through independent exploration, delayed external validation, cognitive-bias reassessment, and repeated refinement.

The method does not reject academic literature. It changes the order of engagement. Instead of beginning with the dominant explanation and building from it, the researcher first builds a provisional model from observation, pattern recognition, inference, and source inspection. The model is then tested against literature, primary evidence, technical standards, counterarguments, and competing explanations.

Purpose

Lehti-Feynman Method
Other name	Extrapolative Trial by Error
Type	Investigative research method; model-building framework
Primary use	Independent reasoning, source analysis, evidence reconstruction, conceptual modeling, and interdisciplinary inquiry
Core sequence	Define → Gather → Preserve → Separate → Model → Review → Test → Correct → Repeat
Central principle	Build the model first, then test it against external knowledge
Key controls	Delayed literature review, post-segment review, discard rules, perspective shifting, source comparison, reassessment
Related Metopedia pages	Metopedia:Research method, Metopedia:Article standards, Metopedia:Source standards, Cognitive Impasse, Selective-Mindedness

The Lehti-Feynman Method is intended for independent research, conceptual development, behavioral analysis, historical reconstruction, systems analysis, source criticism, mathematical exploration, linguistic analysis, forensic reasoning, and other fields where premature exposure to established interpretations may influence the investigator’s conclusions.

Its central purpose is to create a structured delay between initial reasoning and external validation. By postponing academic comparison until after the investigator has formed an initial model, the method attempts to reduce anchoring, imitation, inherited framing, and unexamined dependence on authority.

This delay is not a rejection of scholarship. It is a sequencing device. Academic literature, institutional records, peer-reviewed studies, technical manuals, archives, and expert interpretations remain necessary, but they enter the process after the investigator has produced a visible model that can be tested.

The method is especially useful when a topic contains:

strong institutional framing;
repeated public slogans;
disputed evidence;
missing records;
conflicting expert claims;
ideological pressure;
poorly defined terms;
unverifiable assumptions;
inherited explanations that may have become difficult to question;
a risk that the first explanation encountered will become the interpretive frame for all later evidence.

Core principle

The central principle of the method is:

Build the model first, then test it against external knowledge.

This reverses the normal research sequence. Instead of beginning with established theories and building from them, the researcher begins with observation, inference, pattern recognition, and internal modeling. Once the model reaches a defined stopping point, it is tested against relevant academic work, primary sources, technical evidence, and established standards.

This creates a form of self-directed blind review. The investigator first works without knowing the dominant interpretation in detail, then submits the resulting model to external correction.

The value of the method lies in the tension between independence and correction. Independent modeling protects original observation from premature framing. External review prevents private speculation from becoming insulated belief.

Research sequence

The method follows a repeated five-stage cycle.

Stage	Name	Function	Output
1	Learn	Establish direct contact with the subject.	Raw observations, questions, anomalies, starting definitions.
2	Model	Build a provisional explanation from observation and inference.	A visible model that can be tested.
3	Explore	Expand the model through comparison, note-taking, and structured speculation.	Hypotheses, patterns, subquestions, weak points.
4	Assess	Compare the model with external evidence, literature, standards, and counterarguments.	Retained, refined, discarded, or suspended findings.
5	Repeat	Begin the next segment from the corrected model.	A stronger model and a new research boundary.

Learn

The researcher begins with a fundamental concept, observation, problem, behavior, structure, document, image, event, claim, or anomaly.

This stage involves direct exposure to the subject itself rather than immediate exposure to commentary about the subject. The goal is to establish a raw starting point. A researcher may examine a document, inspect a photograph, observe a behavior, test a calculation, read a source passage, compare two claims, or define a problem before searching for the accepted explanation.

The Learn stage should record:

what was observed;
what question emerged;
what assumptions were already present;
what terms need definition;
what evidence is available directly;
what evidence is missing;
what the researcher noticed before importing outside interpretation.

Model

The researcher creates an initial model by extrapolating from existing knowledge, observed patterns, prior experience, and logical inference.

This model is provisional. It is expected to contain errors, gaps, overextensions, and incomplete assumptions. Its value comes from making the researcher’s reasoning visible enough to test.

A model may be expressed as:

a written explanation;
a diagram;
a timeline;
a causal chain;
a mathematical relation;
a linguistic reconstruction;
a source map;
a forensic comparison table;
a list of competing possibilities;
a hypothesis with stated failure conditions.

A model should not be protected merely because it is original. Originality only creates a testable object. It does not create truth.

Explore

The researcher expands the model through additional observation, comparison, note-taking, and structured speculation.

During this stage, the researcher avoids broad academic review of the subject. The purpose is not to avoid evidence, but to avoid importing the existing interpretive frame too early. Primary sources, direct measurements, source texts, images, logs, artifacts, and raw data may still be examined if they are part of the subject itself.

Exploration may include:

comparing repeated patterns;
checking whether a claim holds across multiple cases;
looking for exceptions;
writing alternative explanations;
testing a simple version of the model;
identifying what evidence would falsify the model;
marking weak assumptions;
separating observation from inference.

The Explore stage is the most vulnerable to overextension. For that reason, speculative notes should be labeled as speculative until the Assess stage.

Assess

After the model has reached a segment boundary, the researcher compares it with academic literature, primary sources, technical evidence, and established findings.

This stage determines which parts of the model align, which parts require revision, and which parts should be discarded.

Assessment should include:

primary-source comparison;
literature review;
search logs or search notes;
counterargument review;
competing-model comparison;
source hierarchy assessment;
internal-consistency testing;
falsification attempts;
uncertainty statement;
limits of the finding.

The Assess stage is where the method becomes disciplined. A researcher who refuses to revise after assessment is no longer using the method.

Repeat

The researcher begins a new segment using the refined model as the next foundation.

The cycle continues until the model becomes stable, coherent, evidence-supported, and resistant to obvious counterexamples.

The Repeat stage prevents a finding from becoming final too early. It also allows one subject to branch into multiple research paths without collapsing them into a single premature conclusion.

Segment-based structure

The method operates through research segments. A segment is a bounded period of investigation focused on a specific question, model, domain, or subproblem.

A segment may last a few hours, several days, months, or longer depending on the complexity of the subject. Multiple segments may run in parallel when they involve related but distinct areas of inquiry.

A segment boundary should be declared when:

the model has reached a coherent draft;
the available direct evidence has been reviewed;
new observations are becoming repetitive;
the researcher feels increasing attachment to the model;
the question has expanded beyond its original scope;
a literature comparison is necessary to prevent private drift;
the model has produced testable claims.

At the end of each segment, the researcher performs a structured review.

Post-segment review

After completing a segment, the researcher reviews relevant external material and applies five operations: align, refine, discard, shift perspective, and reassess.

Operation	Question	Result
Align	Which parts match strong external evidence?	Retain with caution.
Refine	Which parts are too broad, vague, or overstated?	Narrow and clarify.
Discard	Which parts fail against stronger evidence?	Remove or suspend.
Shift perspective	What happens after cognitive distance?	Reduced attachment.
Reassess	What remains after returning with less attachment?	Cleaner model.

Align

The researcher compares independent findings with established research, source material, data, or technical standards.

Aligned findings are retained, but not automatically treated as final. Agreement may indicate validity, but it may also reflect shared assumptions. A researcher must avoid using agreement as a substitute for evidence.

Alignment should record:

what matched;
which source supported it;
whether the supporting source is primary, secondary, technical, journalistic, testimonial, or interpretive;
whether other sources conflict;
whether the agreement is exact or partial.

Refine

Broad, vague, overstated, or partially correct hypotheses are narrowed.

The purpose of refinement is to preserve useful insight while removing exaggeration, ambiguity, or unsupported reach. A refined claim is usually shorter, more specific, and more defensible than the original model.

For example:

Weak form	Refined form
This source proves the claim.	This source supports one part of the claim, but does not establish motive.
The pattern is always present.	The pattern appears in the inspected cases and requires broader sampling.
The institution suppressed the evidence.	The institution removed or failed to preserve the record; motive remains unresolved.

Discard

The researcher removes observations, notes, hypotheses, or interpretations that conflict with stronger evidence or fail to cohere with the broader model.

Discarding is not treated as failure. It is a necessary quality-control step. A discarded claim may still be recorded in notes if it explains the research path, but it should not remain in the conclusion.

A claim should be discarded if it:

depends on unsupported assumptions;
requires selective evidence;
fails against established facts;
cannot be distinguished from speculation;
exists mainly because the researcher is attached to it;
introduces unnecessary complexity;
treats absence of evidence as direct proof;
confuses possibility with probability.

Shift perspective

The researcher temporarily disengages from the subject and works on an unrelated discipline or problem.

This creates cognitive distance, reduces attachment to the model, and allows weak assumptions to become easier to identify after returning. The shift may be short or extended depending on the intensity of the research and the degree of attachment.

Perspective shifting is useful when:

the researcher begins defending the model instead of testing it;
the same claim is being repeated without progress;
anger, pride, or certainty increases;
criticism feels threatening;
the researcher has stopped looking for failure points;
the model has become narratively satisfying but evidentially weak.

Reassess

The researcher returns to the subject with reduced attachment and reviews the model again.

The goal is to identify hidden motives, emotional investment, confirmation bias, narrative bias, unnecessary complexity, and untested assumptions.

Reassessment should ask:

What part of the model would I resist losing?
What evidence would make me change my conclusion?
What did I ignore because it weakened the model?
Which claim is strongest?
Which claim is weakest?
Which claim should be suspended rather than asserted?
What would a critic correctly attack?

Cognitive functions involved

The method depends on several cognitive operations.

Function	Role in the method
Data collection	Gathers raw information from observation, experience, documents, events, behavior, patterns, or source material.
Pattern recognition	Identifies recurring structures and organizes them into meaningful relationships.
Working memory	Holds variables, examples, contradictions, and possible explanations while testing relationships.
Synthesis	Combines separate observations into a coherent model.
Inference	Draws provisional conclusions from incomplete data while marking them as testable.
Source analysis	Evaluates reliability, context, internal consistency, incentive structure, provenance, and evidentiary value.
Critical thinking	Challenges assumptions, tests alternatives, and looks for failure points.
Temporal reasoning	Considers sequence, historical development, causality, delayed effects, and change over time.
Systems thinking	Evaluates interacting causes, incentives, feedback loops, constraints, and institutional effects.
Metacognition	Monitors bias, emotional attachment, premature certainty, and resistance to correction.
Iterative refinement	Improves the model by removing errors, strengthening valid parts, and clarifying unresolved areas.

Bias controls

The method is designed to reduce several common research distortions.

Anchoring bias

Delayed literature review reduces the chance that the first external explanation becomes the frame through which all later evidence is interpreted.

This does not remove anchoring entirely. A researcher still brings prior assumptions, language habits, personal interests, and existing knowledge. The method controls anchoring by making the initial model visible and then forcing comparison with outside evidence.

Confirmation bias

Post-segment comparison forces the researcher to test independent conclusions against external evidence.

Confirmation bias is controlled by requiring the researcher to state what would weaken the model, not only what supports it. A model that only accumulates confirming examples is unfinished.

Belief perseverance

The discard step requires removal of unsupported claims even when they were personally generated.

This is the method’s most important honesty test. If the researcher cannot discard their own claim after it fails, the method becomes self-protection rather than self-correction.

Narrative coherence bias

The researcher must separate a satisfying explanation from a supported explanation.

A model may feel true because it forms a clean story. The method requires the researcher to test whether that story survives evidence, counterexamples, and alternative explanations.

Emotional attachment

Perspective shifting reduces attachment to early conclusions and makes revision easier.

Attachment is not treated as a moral failure. It is treated as a normal research hazard. The solution is not to pretend detachment, but to use a process that exposes attachment before it controls the conclusion.

Methodological advantages

The Lehti-Feynman Method offers several practical advantages.

First, it encourages independent reasoning before inherited frameworks are introduced. This can reveal assumptions that would otherwise remain invisible.

Second, it makes mistakes productive. Incorrect models become diagnostic tools that expose weak reasoning, missing evidence, invalid assumptions, or hidden dependence on prior belief.

Third, it creates a balance between originality and discipline. The researcher is allowed to speculate, but only temporarily. Speculation must later survive comparison, refinement, and correction.

Fourth, it improves long-term learning. Because the researcher builds and repairs models directly, the resulting knowledge is more deeply integrated than memorized information.

Fifth, it helps distinguish discovery from repetition. If the independent model matches established findings after review, the researcher gains both a stronger understanding of the subject and a clearer view of how the established finding was reached.

Sixth, it can reveal gaps in dominant interpretations. A model built before literature review may notice a relationship that established literature leaves implicit, treats as background, or misses because of disciplinary boundaries.

Methodological risks

The method also has risks.

A researcher may overvalue their initial model. They may delay literature review too long. They may mistake pattern recognition for proof. They may discard external evidence too quickly if it conflicts with the model. They may also create complex explanations where simpler ones are sufficient.

The method can fail when the researcher:

treats speculation as evidence;
delays external review beyond the segment boundary;
searches only for supportive sources during assessment;
uses institutional distrust as a substitute for analysis;
treats novelty as correctness;
refuses to discard personally generated claims;
confuses private certainty with public proof;
uses the method to avoid peer criticism.

These risks are controlled through strict post-segment review, source comparison, discard rules, and repeated reassessment.

The method requires humility. Without willingness to be wrong, the process fails.

Quality-control rules

A finding should be retained only if it satisfies at least one of the following conditions:

It aligns with strong external evidence.
It explains observed data better than competing interpretations.
It identifies a gap in existing explanations without contradicting stronger evidence.
It remains coherent after perspective shifting and reassessment.
It survives repeated attempts at falsification or reduction.
It can be stated without hiding uncertainty.
It can be separated from the researcher’s motive for wanting it to be true.

A finding should be discarded or suspended if it:

Depends on unsupported assumptions.
Requires selective evidence.
Fails against established facts.
Cannot be distinguished from speculation.
Exists mainly because the researcher is attached to it.
Requires the rejection of stronger evidence without a clear reason.
Uses a complex explanation where a simpler supported explanation is available.
Treats institutional error as proof of a specific hidden motive.

Relationship to the Feynman Technique

The Lehti-Feynman Method resembles the Feynman Technique in its emphasis on simplification, self-testing, and exposure of misunderstanding.

The distinction is that the Feynman Technique is often framed around explaining a concept clearly to another person, while the Lehti-Feynman Method is aimed at internal model construction and self-correction.

The researcher is not primarily teaching someone else. The researcher is testing whether their own model can survive error, comparison, and revision. In this sense, the researcher teaches the model back to themself through construction, failure, repair, and simplification inspired by and evolved from Richard Feynman's method.

The original standalone method name is Extrapolative Trial by Error. The expanded name, Lehti-Feynman Method, identifies the method as a model-building and self-correction framework while acknowledging its relationship to explanation-based self-testing.

Application areas

The method may be applied to:

research design;
behavioral analysis;
historical analysis;
source criticism;
forensic reasoning;
conceptual modeling;
systems analysis;
education theory;
language reconstruction;
mathematical exploration;
image analysis;
institutional analysis;
evidence review;
fact-checking;
interdisciplinary hypothesis development.

It is best suited for subjects where the investigator must distinguish between observation, inherited explanation, institutional framing, and independent inference.

Use in Metopedia research

Metopedia can apply the Lehti-Feynman Method as an article-development process. In that context, the method does not replace citation standards, source hierarchy, or editorial review. It supplies the pre-literature modeling and self-correction sequence used before a page reaches a public conclusion.

A Metopedia page using this method should identify:

the research question;
the direct observations or sources inspected first;
the provisional model;
the segment boundary;
the literature and source review performed afterward;
what was retained;
what was refined;
what was discarded;
what remains unresolved;
how readers can reproduce the method.

Evidence separation

The method requires separation between evidence and interpretation.

Layer	Meaning	Example wording
Observation	What was directly seen, measured, read, or recorded.	“The page displays a removal notice.”
Inference	What the observation may suggest.	“The removal appears to affect DOI accessibility.”
Hypothesis	A possible explanation requiring further testing.	“The removal may reflect repository-scope enforcement or viewpoint-sensitive moderation.”
Conclusion	A claim supported after comparison, testing, and review.	“The removal record is insufficiently itemized to permit public audit.”

This separation prevents the model from becoming stronger than the evidence.

Automated tools

The method may use automated search tools, language models, OCR systems, image-processing tools, scripts, databases, and other software. These tools are treated as aids, not authorities.

Automated output should be used to generate leads, locate sources, compare text, extract data, or test repeatable procedures. It should not be treated as proof unless independently verified.

When automated tools are used, the researcher should record:

tool name or type;
prompt or query when relevant;
date of use;
output retained;
human verification performed;
errors found;
whether the tool affected the conclusion.

Conclusion

The Lehti-Feynman Method is a recursive research framework based on independent modeling, delayed external validation, error correction, and repeated refinement. It treats wrong conclusions as useful material rather than intellectual failure.

Its strength lies in its sequence: observe, model, explore, test, discard, refine, and repeat. By separating initial reasoning from immediate academic framing, the method attempts to preserve original pattern recognition while still requiring later accountability to evidence.

The method does not replace academic standards. It adds a preliminary stage of independent reasoning before formal comparison. Used carefully, it can strengthen objectivity, reduce anchoring bias, and produce research models that are both original and disciplined.

References

↑ Lehti, Andrew. Extrapolative Trial by Error. figshare, 2024. https://doi.org/10.6084/m9.figshare.27643080

[lehti-method-1] Lehti, Andrew. Extrapolative Trial by Error. figshare, 2024. https://doi.org/10.6084/m9.figshare.27643080

[1]