Understanding evidence and uncertainty
What this chapter does
Not all gaps mean the same thing. A missing document, a question not yet checked, a disputed interpretation and an operation that failed should not all become “false” or “unknown” in the same way. This chapter teaches how to preserve the reason for uncertainty.
Reader outcome
Interpret a statement in context, identify what evidence it has, distinguish assessment states, and describe precisely what a receipt or document does and does not establish.
Our example: MH370 and the limits of an investigation
The 2018 Malaysian safety-investigation report and the Australian Transport Safety Bureau’s search investigation address different questions about Malaysia Airlines flight MH370. The Malaysian team could not determine the real cause from the available evidence; a search analysis can identify a high-probability region or evaluate searched areas without identifying an observed crash coordinate or determining a perpetrator.
Both are genuine official work. Neither should be turned into certainty the documents did not claim, nor treated as useless because it leaves questions unanswered.
1. Start with the exact assertion
Choose a sentence such as The 2018 Malaysian report did not determine the cause of MH370’s disappearance. Identify that sentence as a proposition about the report’s conclusion. Contrast it with the much stronger unsupported claim Investigators proved exactly why MH370 disappeared. The same document cannot be cited indiscriminately for both.
2. Read the source’s scope
Open the Malaysian report’s stated findings and limitations. Record its publication date, investigative mandate, relevant passage and evidence that was unavailable. The ATSB search material is a separately authored analysis of a search effort and its assumptions. A search-area model is not itself a recovered flight recorder.
3. Describe an observation, a citation and an assessment separately
An observation or recorded graph relationship tells you what a provider represented. A cited report offers text to inspect. A registered Source Record identifies a document; retained bytes can preserve a particular copy. A Check, if executed, evaluates specified questions against specified evidence. Do not transfer an outcome from one object to another.
4. Compare the kinds of absence
| State | Accurate interpretation | Not a synonym for |
|---|---|---|
| Not evaluated | No Check question was run or recorded for this proposition | Evaluated and disproved |
| Unresolved | A bounded assessment did not establish a conclusion | Execution failed |
| Refused | An operation or decision was denied under an identified rule | Evidence contradicts the proposition |
| Execution failure | The operation did not complete as designed | A substantive assessment |
| Missing or inaccessible source | Material wasn’t available in that context | No evidence exists anywhere |
| Competing interpretations | Sources or methods yield differing readings | All views are equally supported |
Expected result: you can describe the absence and choose a suitable next step rather than forcing a yes/no verdict.
5. Understand support without flattening it
Ask which exact passage bears on the claim, what methods or assumptions it uses, and whether it is temporally relevant. A reliable report can contain a firm finding about one part of a case and explicit uncertainty about another. Do not inflate a citation, graph edge or document count into multiple independent confirmations.
6. Read a receipt’s verified property
When a receipt is available, read what the verification operation actually compared: a serialized content identity, a signature or issuer binding when supported, a particular time attestation, or a separate substantive reassessment. Verifying one property is not proof of the others. An issued-at timestamp alone is not independent evidence that those bytes existed by that time.
7. Record the uncertainty that remains
Finish with a source-specific account: The named Malaysian report did not determine the cause; the named ATSB report describes the search and its constraints; these documents do not themselves settle every causal hypothesis. Note the date and version you read, the missing material, and the new question a relevant source could address. You need not run Check to describe the limits of these reports accurately.
Continue to Using Counterpedia Check for evaluated claims, or Source tracing for the document itself.
Optional tutorials
The main procedure does not require these exercises. Choose a tutorial for a field-level comparison, or skip to the closing explanation below.
| Tutorial | Output |
|---|---|
| Read an assessment without extending its scope | Input binding, per-question table, and scope-preserving result note |
| Compare attributed positions | Attributed comparison table; source-set omissions; separate disagreement and evidence notes |
Skip the optional exercise and read the closing explanation.
Tutorial: Read an assessment without extending its scope
Inspect the ECF-514 example one test at a time. Separate the recorded assessment from quotation inspection, receipt identity, and questions the result does not answer.
Starting material
Open the ECF-514 result associated with the Check chapter. Record the result identity and the source context shown by that result. This tutorial reads an existing result; the main Check procedure is where you prepare inputs and run a new assessment.
1. Preserve the assessed input
Copy the assessed statement, its source reference, and the source version or capture identity supplied by the result. Keep the result contract or evaluator version when recorded. Comparing results for different statements or source versions is not a repeat of the same test.
2. Record each question separately
For every question listed by the selected result, record its wording, outcome, reason, and associated material. Do not replace several outcomes with one overall verdict.
| Field | Record from the result |
|---|---|
| Question | Exact test question or identifier |
| Outcome | Displayed result, retaining its supplied meaning |
| Reason | The explanation or reason code carried by that result |
| Material | The passage, source version, or artifact the test references |
| Method | Recorded evaluator or verification procedure |
| Unanswered part | The specific question left unassessed or unresolved |
Use the result you are reading, not a fixed expected answer copied from another specimen. A question marked not evaluated differs from an evaluated question that did not establish the tested property. An operation that failed before returning a result is different again.
3. Inspect the quotation and the proposition
Open the cited passage and read its context. Determine which recorded question concerns matching words and which concerns support for the assessed statement. An exact quotation can still be used to make a broader assertion than the passage addresses. Report a support outcome only when the result supplies one, with its stated scope.
Do not turn this exercise into an independent legal ruling. Describe the assessment and the attributed passage that you inspected.
4. Inspect receipt verification separately
Identify what the receipt-verification operation compares: the representation, expected identity, verification method, and comparison outcome. Recomputing a receipt identity is not the same operation as rerunning the assessment. A signature check answers a different question from a digest comparison; neither alone establishes when the source event occurred.
5. Write a result note
Write one sentence for each relevant recorded question: what was evaluated, what outcome was returned, and the reason. Then identify a question that the result leaves open. Keep source existence, quotation accuracy, proposition support, currentness, and the receipt’s identity distinct wherever those questions appear in the selected contract.
Output and completion check
Keep the input reference, per-question table, and result note. Another reader should be able to trace each sentence to a question or passage in the identified result. Reading the result must not be reported as a new Check execution.
Apply the procedure elsewhere
Repeat on a second existing result with a different outcome or on a Theranos assessment bound to a selected statement. Retain the new input and method identities. A difference in outcome is interpretable only in the context of what each assessment examined.
Return to Evidence and uncertainty, or continue with comparing artifact bytes.