From Evidence to Conclusion | How We Know What We Know

LEARNING OBJECTIVES

Learning objectives

By the end of this lesson, the reader should be able to:

Distinguish observation, evidence, inference, hypothesis, explanation, interpretation, argument, and conclusion.
Convert a broad historical assertion into one or more testable claims.
Explain why evidence supports claims through reasoning rather than by speaking for itself.
Compare competing hypotheses instead of evaluating only a favored explanation.
Distinguish deductive, inductive, abductive, analogical, and probabilistic reasoning.
Explain burden of proof and why it changes with the specificity and consequence of a claim.
Distinguish possibility, plausibility, probability, and established conclusion.
Explain correlation, causation, coincidence, common cause, and reverse causation.
Use Bayesian reasoning conceptually without inventing numerical precision.
Identify confirmation bias, motivated reasoning, cherry-picking, survivorship bias, availability bias, and hindsight bias.
Distinguish independent corroboration from repeated dependence on one source.
Explain how negative evidence becomes meaningful when a hypothesis predicts detectable traces.
Evaluate scholarly consensus and minority interpretations without treating either as infallible.
Build a transparent evidence matrix and claim card.
State confidence and uncertainty in proportion to the evidence.
Identify what evidence would strengthen, weaken, or overturn a conclusion.
Revise a conclusion without rewriting the original evidence record.
Apply an end-to-end workflow to historical and alternative-history claims.

INTRODUCTION: A CONCLUSION IS THE LAST STEP, NOT THE FIRST

Introduction: A conclusion is the last step, not the first

Historical arguments often begin where they should end.

A presenter opens with:

“This proves an advanced civilization existed before recorded history.”

The audience is then shown:

an unusual object;
a translated phrase;
a geological formation;
a map alignment;
a missing record;
or an expert disagreement.

The conclusion arrives first. Every later detail is recruited to support it.

Responsible inquiry reverses that order.

It begins with:

a clearly defined question;
documented observations;
identified sources;
alternative explanations;
explicit assumptions;
and a conclusion whose strength matches the evidence.

The Library of Congress primary-source framework asks researchers to observe, reflect, question, and investigate further. National Archives analysis tools likewise move from document characteristics and direct evidence toward historical use. The American Historical Association requires historians to be prepared to explain their methods, assumptions, the relationship between evidence and interpretation, and alternative interpretations.

These practices share one principle:

A defensible conclusion exposes the route by which the evidence leads to it.

01 · COURSE METHOD

The architecture of a conclusion

A historical conclusion can be analyzed as a sequence.

Question

What are we trying to determine?

Examples:

When was this wall constructed?
Who produced this inscription?
Did this text influence another composition?
Was this city destroyed in a named campaign?
Does this object depict a technological device?

Claim

A claim is a proposition that can be supported, weakened, contradicted, or left unresolved.

Evidence

Evidence is information relevant to evaluating the claim.

Inference

An inference connects evidence to a claim.

Hypothesis

A hypothesis is a proposed explanation of the evidence.

Argument

An argument organizes evidence and inference in support of a conclusion.

Conclusion

A conclusion states what the evidence justifies at a stated confidence level.

Limitation

A limitation identifies what the evidence cannot presently establish.

Revision condition

A revision condition states what new evidence would change the conclusion.

Archive standard

Every major conclusion should display its claim, evidence, reasoning, alternatives, confidence, limitations, and revision conditions.

02 · COURSE METHOD

Observations do not contain their own explanation

An observation can be accurate while the explanation is wrong.

Observation

A stone surface contains parallel grooves.

Possible explanations

tool marks from manufacture;
later sharpening;
weathering;
modern cleaning;
symbolic decoration;
or accidental abrasion.

The grooves do not announce which explanation is correct.

Observation

A flood narrative resembles another flood narrative.

Possible explanations

direct literary borrowing;
shared earlier tradition;
common regional experience;
independent development;
later harmonization;
or broad narrative similarity.

Observation

A monument aligns with sunrise on a date.

Possible explanations

intentional alignment;
topographic constraint;
cardinal orientation;
later calendar selection;
multiple-comparison coincidence;
or modern reconstruction error.

Archive standard

Describe the observation in language that remains true even if the preferred explanation is false.

03 · COURSE METHOD

Make the claim testable

Broad claims resist evaluation.

Vague claim

Ancient people possessed advanced knowledge.

Advanced relative to what? Which people? Which knowledge? At what date? Preserved in what source?

Testable claim

A first-millennium BCE astronomical tablet records a planet not visible to the unaided eye and identifies its orbital period accurately.

This claim predicts specific evidence:

an exact tablet;
secure date;
readable lines;
an unambiguous celestial identification;
an orbital period;
translation supported by specialists;
and exclusion of known visible bodies.

Claim decomposition

Large claims should be divided.

“The monument was built by a lost civilization” contains several claims:

The monument predates the known culture.
The known culture did not construct it.
Another population existed at the relevant time.
That population had the required technology and organization.
The monument belongs to that population.

Evidence for one proposition does not establish all five.

Archive standard

Break a dramatic conclusion into the smallest propositions that would need to be true.

04 · COURSE METHOD

Deductive reasoning

A deductive argument aims for a conclusion that follows necessarily from its premises.

Form

All objects of type X have feature Y.
This object is type X.
Therefore, this object has feature Y.

Historical limitation

Historical premises are rarely complete enough to produce strict certainty.

Example:

Every official year in this documented sequence has one eponym.
The inscription names a recognized eponym.
Therefore, it belongs to that year.

The reasoning is strong only if:

the sequence is complete;
the name is correctly read;
there are no duplicate names;
the inscription is contemporary;
and the calendar conversion is secure.

Archive standard

Deductive form does not make uncertain premises certain.

05 · COURSE METHOD

Inductive reasoning

Induction moves from observed cases toward a broader generalization.

Example

Hundreds of securely excavated vessels with a particular rim form cluster within one chronological phase.

A newly excavated vessel with the same regional fabric and rim form is probably associated with that phase.

Strength factors

number of cases;
representativeness;
independence;
consistency;
quality of classification;
and absence of relevant exceptions.

Weak induction

Three internet images show similar symbols, so every ancient culture shared one global technological system.

The sample may be:

selected;
unrepresentative;
misdated;
mislabeled;
or linked only by broad visual resemblance.

Archive standard

Induction is only as strong as the sample, classification, and comparison population.

06 · COURSE METHOD

Abductive reasoning: inference to the best explanation

Abduction asks which explanation best accounts for the evidence.

This is central to historical and archaeological reasoning.

Example

Evidence:

a burned destruction layer;
weapons;
abandoned storage;
a contemporary campaign inscription;
and a chronological match.

Competing explanations:

military destruction;
accidental urban fire;
earthquake followed by fire;
ritual abandonment;
or later disturbance.

The best explanation is the one that accounts for the greatest amount of evidence with the fewest unsupported assumptions while surviving tests against alternatives.

Criteria for explanatory strength

scope;
fit;
coherence;
specificity;
independent support;
predictive success;
simplicity;
and resilience under criticism.

Simplicity is not always “fewest words”

A simple explanation minimizes unnecessary assumptions.

It does not ignore known complexity merely to remain brief.

Archive standard

The best explanation is not the most exciting explanation; it is the explanation that performs best against the full evidence.

07 · COURSE METHOD

Analogical reasoning

Analogy compares a less-understood case with a better-understood one.

Legitimate use

An archaeologist compares wear on an ancient tool with experimentally used replicas.

Relevant similarities

material;
edge geometry;
motion;
worked substance;
microscopic wear;
and residue.

Weak visual analogy

An ancient carving resembles a modern astronaut helmet.

The comparison may identify a visual pattern but does not establish shared function.

Strong analogy requires

relevant similarities;
absence of decisive differences;
known causal connection;
and independent contextual support.

Archive standard

Resemblance generates a hypothesis; mechanism and context determine whether the analogy supports it.

08 · COURSE METHOD

Burden of proof

The burden of proof belongs to the person advancing a claim.

Positive claim

This inscription records an extraterrestrial visitation.

The claimant must provide relevant evidence.

Negative response

The proposed evidence does not establish that interpretation.

The critic does not need to prove that extraterrestrial visitation is metaphysically impossible. The critic must show why the evidence does not support the specific claim.

Competing positive claim

The inscription is definitely a metaphor for seasonal rain.

This alternative also bears a burden of proof.

Burden varies with claim strength

“This may be a ritual object” requires less evidence than “This was certainly a reactor component.”
“The traditions are related” requires less evidence than “Text A directly copied Text B in 720 BCE.”
“A regional flood may underlie the story” requires less evidence than “One flood covered the entire globe in one year.”

Archive standard

The more specific, consequential, and revisionary the claim, the stronger and more diverse the required evidence.

09 · COURSE METHOD

Possibility is not probability

A claim can move through several levels.

Logically possible

The statement contains no obvious contradiction.

Physically possible

It does not conflict with known physical constraints.

Historically possible

The required people, materials, chronology, and circumstances could have existed.

Plausible

The claim fits known context and has some supporting evidence.

Probable

The evidence favors it over major alternatives.

Highly probable

Strong independent evidence converges and alternatives explain less.

Established

The conclusion is supported by multiple robust lines of evidence and is unlikely to change in its core form, though details may be revised.

Common error

“You cannot prove it did not happen, therefore it probably happened.”

Inability to exclude a claim does not increase it to probability.

Archive standard

Possible means not excluded. It does not mean supported.

10 · COURSE METHOD

Correlation and causation

Two events occurring together do not automatically establish cause.

Correlation

A and B vary together.

Direct causation

A produces B.

Reverse causation

B produces A.

Common cause

C produces both A and B.

Coincidence

The relationship occurs by chance.

Selection effect

Only matching cases were selected.

Historical example

Several cities decline during a period of climatic stress.

Possible explanations:

climate directly caused collapse;
climate intensified war or migration;
political fragmentation reduced resilience;
trade failure caused decline while climate was coincidental;
different cities declined for different reasons.

Causal evidence improves with

correct temporal order;
mechanism;
dose or scale relationship;
comparative cases;
exclusion of common causes;
and independent evidence.

Archive standard

A chronological overlap is the beginning of a causal argument, not its completion.

11 · COURSE METHOD

Multiple comparisons and pattern hunting

When researchers test many possible alignments, names, dates, or numerical relationships, some matches will occur by chance.

Example

A monument has many edges, corners, sightlines, and possible observation points.

A researcher checks:

every solstice;
equinox;
lunar standstill;
bright star;
planet;
historical anniversary;
and calendar system.

One close alignment is likely.

Proper evaluation

Ask:

Was the target predicted before measurement?
How many alternatives were tested?
What tolerance counts as a match?
Is the observation point ancient and documented?
Does architecture show intentional viewing design?
Are similar alignments absent where the theory predicts absence?

Archive standard

A pattern found after unlimited searching requires stronger independent confirmation than a pattern predicted in advance.

12 · COURSE METHOD

Bayesian reasoning without false precision

Bayesian reasoning updates confidence when new evidence appears.

Conceptually:

Posterior confidence depends on prior confidence and how expected the evidence is under competing hypotheses.

Example

Hypothesis A: A tablet is an ordinary school exercise.

Hypothesis B: It is a technical record of an unknown aircraft.

Evidence: The text contains a word translated “wing.”

Ask:

How common is wing imagery in ordinary literary, divine, or animal contexts?
How likely would “wing” be under a real aircraft hypothesis?
Does the tablet contain measurements, materials, controls, procedures, or repeated technical vocabulary?

The word may be expected under both hypotheses. It therefore provides little discrimination.

Likelihood ratio concept

Evidence is powerful when it would be much more expected if one hypothesis were true than if alternatives were true.

Avoid invented numbers

Historical evidence rarely supports claims such as:

“There is an 87% chance this king existed.”
“The artifact is 93% alien.”

Numerical probabilities require a defensible model, defined evidence, and justified inputs.

Qualitative Bayesian reasoning

Use language such as:

substantially raises confidence;
modestly supports;
expected under both explanations;
strongly discriminates;
weakly relevant;
or lowers confidence.

Archive standard

Use Bayesian structure to discipline updating, not to manufacture numerical authority.

13 · COURSE METHOD

Evidence matrices

An evidence matrix compares evidence against competing hypotheses.

Example hypotheses

H1: The object is an ancient ritual vessel.
H2: The object is an ancient technological component.
H3: The object is a modern composite.

Matrix questions

For each item, ask:

Is it expected under H1?
Is it expected under H2?
Is it expected under H3?
Does it discriminate among them?
Is the evidence independent?
What is its quality?

Benefit

The matrix prevents one spectacular detail from overwhelming the complete record.

Limitation

A matrix is only as good as:

the hypotheses included;
the evidence quality;
and the honesty of the scoring.

Archive standard

Include the strongest plausible rival explanation, not only weak alternatives designed to lose.

14 · COURSE METHOD

Independent corroboration

Several sources can appear independent while sharing one origin.

Apparent multiplicity

five books;
ten videos;
twenty websites;
and several AI summaries

may all trace to one unverified translation.

Independent evidence

Information is more independent when it arises from different:

objects;
sites;
archives;
observers;
laboratories;
languages;
methods;
or causal traces.

Partial dependence

Two chronicles may use one lost source but also preserve independent additions.

Two laboratories may analyze separate samples but rely on the same calibration model.

Dependence is often a matter of degree.

Archive standard

Count evidentiary origins, not publications.

15 · COURSE METHOD

Convergence and consilience

A conclusion is especially strong when different kinds of evidence converge.

Example

A ruler’s campaign may be supported by:

the ruler’s inscription;
an enemy archive;
archaeological destruction;
administrative interruption;
coin or seal evidence;
and a compatible date.

Each source has different weaknesses.

Together they may create a stronger case than any one source.

False convergence

Several lines may share one hidden assumption.

Example:

translation;
chronology;
site identification;
and historical narrative

all depend on one disputed name restoration.

Archive standard

Convergence is strongest when methods fail in different ways.

16 · COURSE METHOD

Negative evidence and predicted traces

A hypothesis should predict what evidence would probably exist if it were true.

Weak negative evidence

No inscription naming a minor official survives.

Records may be incomplete.

Stronger negative evidence

A proposed large industrial civilization should leave some combination of:

settlements;
extraction sites;
waste;
manufacturing debris;
transport infrastructure;
altered landscapes;
tools;
chemical residues;
and widespread durable products.

If extensive investigation repeatedly fails to find predicted traces, confidence should decrease.

Preservation analysis

Before using absence, ask:

Would the evidence preserve?
Has the correct location been searched?
Is the expected signature distinctive?
Is excavation extensive enough?
Could later processes erase it?

Archive standard

Absence becomes evidence when presence was a clear and testable prediction.

17 · COURSE METHOD

Confirmation bias

Confirmation bias favors evidence that supports an existing belief.

Forms

searching only favorable terms;
accepting supportive sources with little scrutiny;
demanding impossible proof from critics;
remembering hits and forgetting misses;
interpreting ambiguity in one direction;
and stopping research after finding support.

Countermeasures

write rival hypotheses first;
define disconfirming evidence;
seek the strongest criticism;
separate source quality from agreement;
and record rejected evidence with reasons.

Archive standard

Apply the same evidentiary threshold to conclusions you prefer and conclusions you dislike.

18 · COURSE METHOD

Motivated reasoning

Motivated reasoning uses intelligence to defend a desired conclusion.

Motives may be:

religious;
anti-religious;
political;
national;
commercial;
professional;
personal;
or community-based.

Warning signs

conclusion cannot be revised;
critics are explained only through bad motives;
every contrary result becomes evidence of suppression;
standards change depending on direction;
and uncertainty feels morally unacceptable.

Countermeasure

Ask:

What would I conclude if the evidence pointed toward the opposite identity, religion, nation, or theory?

Archive standard

A meaningful revision condition is evidence that the conclusion is functioning as research rather than identity protection.

19 · COURSE METHOD

Cherry-picking

Cherry-picking selects favorable evidence while omitting relevant contrary evidence.

Examples

quoting one translation while ignoring stronger alternatives;
using the one radiocarbon date that fits;
displaying only matching symbols;
citing only supportive experts;
or ending a quotation before its qualification.

Complete reporting

A defensible argument reports:

all relevant results;
exclusion criteria;
major contrary evidence;
and unresolved conflicts.

The American Historical Association’s professional standards emphasize integrity in representing sources and readiness to explain evidence, interpretation, and alternatives.

Archive standard

A conclusion cannot be stronger than the evidence remaining after unfavorable evidence is restored to the record.

20 · COURSE METHOD

Survivorship, availability, and selection bias

Survivorship bias

Only surviving evidence is analyzed.

Durable elite monuments may dominate conclusions about societies whose ordinary records disappeared.

Availability bias

Memorable evidence feels more common or important.

A dramatic artifact may outweigh thousands of ordinary objects in public perception.

Selection bias

The sample was chosen in a way that favors a result.

Publication bias

Positive or dramatic results are more likely to be publicized.

Archive standard

Ask what was not preserved, not collected, not published, and not selected for presentation.

21 · COURSE METHOD

Hindsight bias and narrative compression

After an event, its outcome can feel inevitable.

Historical hindsight

The fall of an empire is presented as the unavoidable result of one visible weakness.

Contemporaries may have faced:

multiple possible outcomes;
incomplete information;
reversible crises;
and unexpected events.

Narrative compression

Decades of change become one sentence:

“The civilization collapsed because of drought.”

The compressed story may omit:

regional differences;
migration;
war;
political choices;
disease;
trade;
and recovery.

Archive standard

A clear narrative should not erase the uncertainty experienced by historical actors or the complexity of causal processes.

22 · COURSE METHOD

Consensus

Scholarly consensus is the current broad judgment of qualified specialists.

Why it matters

Consensus can summarize:

accumulated evidence;
criticism;
replication;
specialist methods;
and explanatory comparison.

What it does not do

Consensus does not make a proposition true by vote.

It can be:

incomplete;
biased;
slow to change;
or wrong.

Responsible use

Ask:

Which specialists?
How broad is the agreement?
What evidence supports it?
What parts are secure?
What parts remain debated?
What would change it?

Archive standard

Consensus is evidence about expert evaluation, not a substitute for the underlying record.

23 · COURSE METHOD

Minority and alternative positions

A minority position should be assessed by evidentiary performance.

Strong minority position

uses authentic sources;
engages current scholarship;
explains the consensus accurately;
identifies a specific weakness;
proposes testable alternatives;
and survives specialist criticism.

Weak minority position

relies on mistranslation;
misrepresents consensus;
treats criticism as suppression;
avoids peer engagement;
and changes claims when tested.

Unpopularity is not proof

A claim is not true because it is rejected.

A claim is not false because it is unconventional.

Archive standard

Give alternative positions a fair hearing and the same evidentiary examination—not automatic elevation or dismissal.

24 · COURSE METHOD

Disagreement does not imply equal support

Scholars may disagree about:

exact dates;
translation;
causation;
fragment placement;
or historical identity.

Bounded disagreement

Researchers may agree that:

the object is ancient;
the inscription is genuine;
the text concerns a campaign;

while disagreeing about:

which campaign;
one restored name;
or a twenty-year date range.

A documentary may present this as “experts cannot agree what the artifact is.”

That is misleading.

Archive standard

Map the location and scale of disagreement. Do not expand a narrow dispute into total uncertainty.

25 · COURSE METHOD

Confidence language

This framework is shared across the Archive methodology.

The Archive confidence scale should remain disciplined.

Established

Multiple strong independent lines support the core conclusion.

Highly probable

Evidence strongly favors the claim and serious alternatives explain less.

Probable

The claim is better supported than major alternatives.

Plausible

The claim fits the evidence but remains underdetermined.

Possible

The claim is not excluded, but positive evidence is weak.

Speculative

The proposal extends substantially beyond available evidence.

Unsupported

Relevant support has not been demonstrated.

Contradicted

Strong evidence conflicts with the claim.

Unknown

The evidence cannot presently decide.

Confidence belongs to a specific proposition

An object can be:

established as ancient;
probably from one region;
plausibly used ritually;
and unsupported as advanced technology.

Archive standard

Do not assign one confidence label to an entire story containing claims of different strength.

26 · COURSE METHOD

Precision and uncertainty

NIST guidance on measurement uncertainty emphasizes that a reported interval and confidence interpretation depend on assumptions about probability distributions and measurement models.

Historical conclusions add further uncertainty from:

source preservation;
context;
identification;
translation;
chronology;
and causal interpretation.

False precision

“The city fell because climate contributed 62% and warfare 38%.”

Such numbers require a defined causal model and measurable variables.

Appropriate precision

“Climate stress probably increased vulnerability, but the evidence does not isolate its contribution from political fragmentation and conflict.”

Archive standard

Precision should reflect the resolution of the evidence, not the confidence of the presenter’s voice.

27 · COURSE METHOD

Claim cards

A claim card makes reasoning visible.

Claim

One testable proposition.

Claim type

identity;
date;
origin;
function;
relationship;
causation;
authenticity;
or interpretation.

Primary evidence

Exact objects, records, measurements, and passages.

Source condition

complete or fragmentary;
excavated or unprovenanced;
contemporary or later;
preserved or restored;
direct or indirect.

Reasoning

Explain the inferential steps.

Competing hypotheses

Include the strongest serious alternatives.

Discriminating evidence

Identify evidence more expected under one hypothesis than another.

Dependencies

Identify evidence derived from the same underlying source.

Contrary evidence

State evidence that weakens the claim.

Confidence

Use the Archive scale.

Limitations

State what is not established.

Revision conditions

State what would change the conclusion.

28 · COURSE METHOD

Evidence matrices

A claim card explains one conclusion. An evidence matrix compares hypotheses.

Suggested fields

evidence item;
source identifier;
observation;
quality;
independence;
expected under H1;
expected under H2;
expected under H3;
discriminating value;
uncertainty;
and notes.

Avoid fake arithmetic

Do not assign arbitrary scores such as:

+5 for supportive;
−3 for conflicting;

unless the scoring system has a defined and validated meaning.

Qualitative classifications may be more honest:

strongly favors;
moderately favors;
weakly favors;
neutral;
weakly conflicts;
strongly conflicts.

Archive standard

The matrix organizes judgment; it does not automate historical truth.

29 · COURSE METHOD

Worked example: a celestial image

Claim

An ancient relief depicts a spacecraft orbiting a planet.

Observations

circular object;
radiating marks;
winged figure;
surrounding symbols;
damaged upper edge.

Hypotheses

H1: Astral or divine iconography.
H2: Royal symbol and protective deity.
H3: Technological spacecraft depiction.
H4: Modern misidentification caused by cropping.

Evidence needed

exact object record;
full scene;
date and context;
contemporary iconographic parallels;
ancient terminology;
material technological evidence;
and history of the photograph.

Discriminating evidence

A repeated technical diagram with components, measurements, materials, and physical remains would strongly favor H3.

A well-attested symbol appearing in religious scenes without technical context favors H1 or H2.

Conclusion

Visual resemblance alone supports investigation but not technological identification.

Confidence

ancient image: established if authenticated;
astral or religious context: probable if parallels are strong;
spacecraft: unsupported without independent technological evidence.

30 · COURSE METHOD

Worked example: a flood tradition

Claim

Two ancient flood narratives describe the same historical event.

Observations

Both may contain:

divine warning;
vessel;
preservation of life;
catastrophic water;
landing;
and sacrifice.

Hypotheses

H1: Direct literary dependence.
H2: Common inherited tradition.
H3: Independent stories shaped by recurring flood experience.
H4: Later harmonization.
H5: One global event directly remembered by both.

Evidence

textual wording;
motif sequence;
language;
manuscript dates;
geographic transmission;
archaeological flood data;
and geological scale.

Discriminating evidence

Highly specific shared wording and sequence may favor literary relationship.

Regional flood deposits may support historical environmental background.

A global-event claim requires globally synchronous geological evidence, not only narrative resemblance.

Conclusion

Literary relationship and historical event identity are different propositions.

31 · COURSE METHOD

Worked example: alleged suppression

Claim

A museum is concealing an artifact because it disproves conventional history.

Evidence required

exact object;
documentation of its location;
access requests;
unusual denial;
communications showing motive;
and evidence that the object supports the claimed conclusion.

Alternative explanations

conservation restriction;
legal dispute;
catalog backlog;
security;
misidentification;
false internet claim;
or ordinary unpublished research.

Circular form

Lack of access proves suppression.
Suppression proves the object is revolutionary.
Its revolutionary nature explains lack of access.

No independent support enters the loop.

Conclusion

Institutional concealment is historically possible and has occurred in various contexts.

A specific concealment claim must be documented rather than assumed from missing access.

32 · COURSE METHOD

Worked example: one anomalous date

Claim

One radiocarbon result proves the accepted chronology is wrong.

Evaluation

What material was dated?
Is the context secure?
Was the sample calibrated?
Could old wood, reservoir effect, contamination, residuality, or intrusion apply?
How many other dates exist?
Does stratigraphy support the result?
Was the result replicated?
Which part of the chronology would change?

Possible outcomes

laboratory anomaly;
sample-context problem;
real outlier;
misidentified phase;
or genuine chronological revision.

Conclusion

An anomaly deserves investigation.

It does not automatically validate every alternative chronology.

33 · COURSE METHOD

End-to-end investigation workflow

Step 1: Define the research question

Make it narrow enough to answer.

Step 2: Decompose the claim

Separate identity, date, function, origin, transmission, and causation.

Step 3: Identify the best evidence

Locate exact objects, lines, contexts, datasets, and records.

Step 4: Document source condition

Record preservation, provenance, dating, copying, restoration, and authenticity.

Step 5: Separate observation from interpretation

Write each in different fields.

Step 6: Build at least two serious hypotheses

Include the strongest conventional and alternative explanations.

Step 7: State predictions

What should be observed if each hypothesis is true?

Step 8: Identify discriminating evidence

Prioritize evidence that differs across hypotheses.

Step 9: Check independence

Trace each source to its evidentiary origin.

Step 10: Test chronology

Ensure objects, texts, and proposed events can overlap.

Step 11: Test language and translation

Check textual condition, grammar, lexical range, and variants.

Step 12: Test context and authenticity

Evaluate excavation, provenance, manufacture, and alteration.

Step 13: Seek contrary evidence

Search intentionally for disconfirmation.

Step 14: Evaluate negative evidence

Ask what expected traces are absent and whether they should survive.

Step 15: Compare explanatory performance

Assess scope, fit, assumptions, and unresolved evidence.

Step 16: Assign evidence lanes

Classify primary evidence, scholarly interpretation, contested interpretation, alternative interpretation, speculation, and unsupported claims.

Step 17: Assign proposition-specific confidence

Do not label the whole narrative at once.

Step 18: State limitations

Separate known, inferred, and unknown.

Step 19: Define revision conditions

State what evidence would change the judgment.

Step 20: Publish the reasoning trail

Provide citations, object identifiers, alternatives, and uncertainty.

Step 21: Reassess after criticism

Respond to the strongest objection, not the easiest one.

Step 22: Version the conclusion

Preserve earlier judgments and explain why confidence changed.

34 · COURSE METHOD

Revising conclusions

Revision is not failure.

Responsible revision record

previous conclusion;
previous confidence;
new evidence;
changed inference;
new conclusion;
new confidence;
and remaining uncertainty.

Avoid retrospective rewriting

Do not pretend the new conclusion was always obvious.

Preserve the history of interpretation.

Confidence can move in either direction

New evidence may:

strengthen;
weaken;
narrow;
broaden;
or divide one claim into several claims.

Archive standard

A trustworthy research platform records how and why it changed its mind.

35 · COURSE METHOD

Transparency and reproducibility

Historical work cannot always reproduce the past event.

It can make its research process inspectable.

Reproducible elements

exact citations;
object identifiers;
images;
transcription;
code;
calibration settings;
comparison corpus;
inclusion criteria;
and reasoning.

Irreversible evidence

Excavation removes deposits as it records them.

Destructive sampling consumes material.

This increases the need for:

field records;
sample archives;
photographs;
and transparent methods.

Archive standard

Where the event cannot be repeated, the evidentiary pathway should be documented well enough to be challenged.

36 · COURSE METHOD

Ethics of conclusion-making

Conclusions can affect:

descendant communities;
ownership claims;
national identity;
religious belief;
reputations;
legal disputes;
and cultural heritage.

Ethical responsibilities

avoid fabrication;
distinguish allegation from established fact;
represent communities accurately;
avoid sensationalizing human remains;
disclose uncertainty;
correct errors;
and protect sensitive information when necessary.

Ethical restraint is not censorship

Not publishing a burial location to prevent looting is different from suppressing evidence to protect a theory.

Archive standard

Transparency must be balanced with documented conservation, privacy, safety, and heritage responsibilities.

37 · COURSE METHOD

Course synthesis: six disciplines of evidence

Course 2 established six connected disciplines.

Lesson 1: What Counts as Evidence?

Separate claims, sources, observations, inferences, and conclusions.

Lesson 2: Dating the Ancient World

Distinguish material, object, deposit, and event dates; integrate stratigraphy, measurement, and historical chronology.

Lesson 3: Reconstructing Lost and Fragmentary Texts

Keep preserved wording, damaged traces, restorations, manuscript placement, and composite editions visible.

Lesson 4: Translation, Scribal Transmission, and Bias

Trace modern wording through script, language, grammar, variants, translators, institutions, genre, and audience.

Lesson 5: Archaeology, Provenance, and Authenticity

Evaluate context, ownership history, manufacture, alteration, legal-ethical status, and historical reliability separately.

Lesson 6: From Evidence to Conclusion

Compare hypotheses, test predictions, control bias, assign proportional confidence, and publish revision conditions.

Combined Archive rule

No extraordinary conclusion should bypass the evidence chain established across these six lessons.

38 · COURSE METHOD

The Archive final evidence protocol

Before publishing a major historical claim, confirm:

Identity

Exact source identified?
Object or manuscript number supplied?
Image and record matched?

Context

Provenience documented?
Provenance documented?
Deposit and association understood?

Authenticity

Material, manufacture, inscription, and restoration evaluated separately?

Chronology

What is dated?
Which method and chronology system?
What uncertainty remains?

Text

Preserved, damaged, restored, and supplied wording distinguished?
Witness and composite edition distinguished?

Translation

Original language checked?
Grammar and lexical range evaluated?
Translation type labeled?

Reasoning

Observation separated from inference?
Competing hypotheses included?
Dependence and corroboration checked?
Negative evidence evaluated?

Conclusion

Confidence proportional?
Contrary evidence included?
Limitations stated?
Revision conditions stated?

Publication

Sources linked?
Claims versioned?
Corrections possible?

39 · COURSE METHOD

Common conclusion failures

Conclusion first

Evidence is selected after belief is fixed.

Claim bundling

Evidence for one proposition is transferred to an entire narrative.

Possibility inflation

Not impossible becomes probable.

False dichotomy

Only two explanations are offered when others exist.

Weak rival

An implausible alternative is chosen to make the favored view look strong.

Correlation causation error

Temporal overlap becomes cause.

Dependence inflation

Many repetitions become independent confirmation.

Cherry-picking

Contrary evidence is hidden.

Gap filling

Missing data becomes support for the preferred theory.

Argument from ignorance

Lack of explanation becomes proof of a specific explanation.

Moving the goalposts

Evidence requirements change after the claim is challenged.

Conspiracy closure

Every contradiction becomes further proof of concealment.

Numerical theater

Unsupported percentages create false precision.

Consensus worship

Expert agreement replaces examination.

Anti-expert reflex

Past expert error is used to dismiss all expertise.

Narrative certainty

A clear story conceals complex and unresolved evidence.

40 · COURSE METHOD

Established knowledge, debate, and speculation

Established

Evidence supports claims through explicit or implicit inference.
Observations can fit multiple explanations.
Broad claims should be decomposed into testable propositions.
Historical reasoning commonly uses induction, abduction, analogy, and probabilistic updating.
Possibility is weaker than probability.
Correlation does not by itself establish causation.
Repeated dependence on one source is not independent corroboration.
Negative evidence becomes stronger when a hypothesis predicts detectable traces.
Confirmation bias and cherry-picking can distort evaluation.
Scholarly consensus can be informative without being infallible.
Confidence should be assigned to specific propositions.
Conclusions should state limitations and revision conditions.
Transparent methods and source trails allow criticism and correction.

Strong methodological interpretation

Comparing serious competing hypotheses produces stronger conclusions than testing one favored theory alone.
Evidence that discriminates among hypotheses is more valuable than evidence expected under all of them.
Convergence among methods with different failure modes is especially strong.
Qualitative Bayesian reasoning is useful when numerical inputs are not defensible.
An evidence matrix can expose hidden assumptions and source dependence.
Versioned conclusions increase trust.

Contested

How much prior knowledge should influence extraordinary claims.
When a broad consensus becomes sufficiently stable to call a proposition established.
How to weigh simplicity against historically documented complexity.
Whether historical confidence should ever be expressed numerically.
How much contrary evidence must be included in public-facing lessons.
How to balance open evidence with heritage protection and privacy.

Speculative or unsupported

A possible explanation is therefore the best explanation.
A mystery is evidence for a preferred extraordinary cause.
Many online repetitions are independent corroboration.
One anomaly invalidates an entire chronology or discipline.
Expert disagreement means every interpretation is equally supported.
Consensus proves truth without source examination.
Past scientific mistakes make all current methods arbitrary.
A cover-up can be assumed whenever evidence is missing.
Numerical confidence is valid merely because software produced it.
A conclusion that cannot be revised is a historical finding rather than a protected belief.

KEY TERMS

Key terms

Question: The problem an investigation seeks to resolve.

Claim: A proposition open to evaluation.

Evidence: Information relevant to a claim.

Observation: Directly documented feature or result.

Inference: Reasoned connection between evidence and claim.

Hypothesis: Proposed explanation.

Argument: Organized evidence and reasoning supporting a conclusion.

Conclusion: Judgment justified by the evidence at a stated confidence.

Deduction: Reasoning in which the conclusion follows from premises if they are true.

Induction: Generalization from observed cases.

Abduction: Inference to the best explanation.

Analogy: Reasoning from relevant similarity.

Burden of proof: Responsibility to support an asserted claim.

Possibility: Status of not being excluded.

Plausibility: Fit with context and some supporting evidence.

Probability: Degree to which evidence favors a claim.

Correlation: Variables or events occurring together.

Causation: One factor producing or contributing to another.

Common cause: A third factor causing both correlated events.

Confounder: Factor that distorts an apparent relationship.

Bayesian reasoning: Updating confidence by considering prior expectations and evidentiary likelihood.

Prior: Confidence before considering new evidence.

Posterior: Updated confidence after considering evidence.

Likelihood: How expected evidence is under a hypothesis.

Likelihood ratio: Comparative evidentiary expectation under competing hypotheses.

Corroboration: Evidence strengthening another source or claim.

Independence: Degree to which evidence does not derive from the same informational origin.

Convergence: Different evidence pointing toward one conclusion.

Consilience: Agreement among independent methods or domains.

Negative evidence: Significance of an expected but absent trace.

Confirmation bias: Favoring information that supports prior belief.

Motivated reasoning: Reasoning shaped toward a desired outcome.

Cherry-picking: Selecting favorable evidence while omitting relevant contrary evidence.

Survivorship bias: Distortion caused by analyzing only what survived.

Availability bias: Overweighting memorable or easily recalled evidence.

Selection bias: Distortion caused by nonrepresentative selection.

Publication bias: Preference for publishing positive or dramatic results.

Hindsight bias: Treating an outcome as more predictable after it occurred.

Consensus: Broad specialist judgment on current evidence.

Minority position: Interpretation accepted by fewer specialists.

Evidence matrix: Structured comparison of evidence across hypotheses.

Claim card: Structured record of claim, evidence, reasoning, alternatives, and confidence.

Discriminating evidence: Evidence expected more strongly under one hypothesis than another.

Revision condition: Evidence that would change a conclusion.

False dichotomy: Presenting two options when additional alternatives exist.

Argument from ignorance: Treating lack of disproof or explanation as proof.

Multiple-comparison problem: Increased chance of apparent patterns when many possibilities are tested.

Reproducibility: Ability to inspect or repeat an analysis from documented data and methods.

Versioning: Preserving successive forms of a conclusion and reasons for change.

KEY TAKEAWAYS

Key takeaways

A conclusion must expose the inferential route from evidence.
Observations and explanations are different categories.
Large historical narratives should be divided into testable propositions.
Historical conclusions usually rely on induction, abduction, analogy, and probabilistic reasoning.
The burden of proof belongs to the claimant.
Possible does not mean probable.
Correlation requires mechanism and comparative evidence before becoming causation.
Pattern hunting must account for the number of possibilities tested.
Bayesian reasoning is useful without fabricated percentages.
Competing-hypothesis matrices reveal which evidence actually discriminates.
Repetition is not independence.
Negative evidence matters when a hypothesis predicts preserved traces.
Confirmation bias, motivated reasoning, and cherry-picking require procedural controls.
Consensus is relevant but not infallible.
Alternative interpretations deserve accurate representation and equal evidentiary standards.
Confidence must be proposition-specific.
Limitations and revision conditions are part of the conclusion.
A transparent change of mind strengthens rather than weakens research credibility.
Course 2 provides the mandatory evidence protocol for all later Archive courses.

KNOWLEDGE CHECK

Knowledge check

Distinguish observation, inference, hypothesis, and conclusion.
Why should a broad claim be decomposed?
What makes a claim testable?
Why does deductive form not guarantee a historical conclusion?
What factors strengthen induction?
Define abductive reasoning.
What makes one explanation better than another?
When is analogy evidentially useful?
Who bears the burden of proof?
Why does inability to disprove a claim not make it probable?
Distinguish logical, physical, historical, and evidential possibility.
Distinguish correlation, reverse causation, common cause, and coincidence.
Why does chronological overlap not establish causation?
What is the multiple-comparison problem?
How can astronomical alignment claims be tested responsibly?
Explain Bayesian updating in plain language.
What makes evidence discriminating?
Why should historical studies avoid unsupported numerical confidence?
What is an evidence matrix?
Why must a matrix include strong rival hypotheses?
Why are many books not necessarily many independent sources?
What is convergence?
Why is convergence stronger when methods have different failure modes?
When does absence become evidence?
What is confirmation bias?
Give four safeguards against confirmation bias.
What is motivated reasoning?
What is cherry-picking?
Distinguish survivorship, availability, and selection bias.
What is hindsight bias?
Why does scholarly consensus matter?
Why is consensus not proof by vote?
What distinguishes a strong minority interpretation from a weak one?
Why does disagreement not imply equal evidentiary support?
Why must confidence be assigned to propositions rather than stories?
What is false precision?
What fields belong in an Archive claim card?
What is a revision condition?
Why should earlier conclusions be preserved when revised?
Apply the end-to-end workflow to the claim: “A carved image proves ancient aircraft existed.”

Primary and institutional sources

Library of Congress. Primary Source Analysis Tool and Teacher’s Guides. Observe, reflect, question, and investigate further as a disciplined source-analysis sequence.

National Archives and Records Administration. Document Analysis Worksheets. Structured examination of written documents, photographs, artifacts, maps, audiovisual records, and perspective before historical use.

American Historical Association. Statement on Standards of Professional Conduct. Professional obligations concerning source integrity, methods, assumptions, evidence, interpretation, and alternative interpretations.

National Institute of Standards and Technology. Guide to the Expression of Uncertainty in Measurement and uncertainty-reporting resources. Measurement models, distributions, uncertainty intervals, coverage, and explicit assumptions.

NIST. Concepts, Principles, and Methods for the Assessment of Measurement Uncertainty. Probability, independence, exchangeability, and interpretation of uncertainty.

Course 2 institutional corpora and standards. Oracc, CDLI, Papyri.info, Historic England, Oxford Radiocarbon Accelerator Unit, IntCal, the British Museum, the Israel Antiquities Authority, UNESCO, ICOM, AIA, and relevant museum or laboratory records. The final conclusion protocol integrates the object, chronology, text, translation, provenance, and authenticity standards established in Lessons 1–5.

RECOMMENDED SCHOLARLY FOUNDATION

Recommended scholarly foundation

Marc Bloch, The Historian’s Craft
Martha Howell and Walter Prevenier, From Reliable Sources
David Hackett Fischer, Historians’ Fallacies
Richard J. Evans, In Defence of History
John Lewis Gaddis, The Landscape of History
Carlo Ginzburg, essays on clues and historical inference
C. Behan McCullagh, Justifying Historical Descriptions
Aviezer Tucker, Our Knowledge of the Past
Alison Wylie, work on archaeological reasoning and evidential security
Merrilee Salmon, Philosophy and Archaeology
Colin Renfrew and Paul Bahn, Archaeology: Theories, Methods and Practice
Peter Kosso, Knowing the Past
Ian Hacking, work on probability, evidence, and inference
Deborah Mayo, work on severe testing and evidence
Elliott Sober, Evidence and Evolution and work on likelihood reasoning
Judea Pearl, Causality and The Book of Why
Richard Lempert, work on relevance and inferential evidence
Daniel Kahneman, Paul Slovic, and Amos Tversky, work on judgment and bias
Raymond Nickerson, work on confirmation bias
Jonathan Baron, Thinking and Deciding
Philip Tetlock and Dan Gardner, Superforecasting
Naomi Oreskes, work on expert consensus and historical analysis of science
National Academies reports on reproducibility, replicability, and open science

The Archive should use these works as methodological foundations, not as substitutes for the object-level evidence relevant to each historical claim.

YOUR COURSE PROGRESS

Ready to preserve your progress?

Completion is saved to your member account and follows you across browsers and devices.