A persuasive story can be coherent without being well supported
Narrative analysis makes a story auditable by decomposing it into claims, causal links, evidence sources, assumptions, omissions, alternatives and explicit revision triggers.
People understand complex events through stories: actors have motives, events have causes and one development leads to another. Narratives help organise information, but they can compress uncertainty, hide source dependence and make a conclusion feel inevitable after the fact.
The analytic task is not to reject narrative. It is to expose its structure. A claim network shows which observations support which inferences, which assumptions connect them, and where one source quietly supports several apparently independent conclusions.
Chapter contents
- 1. Frame the narrative as an answer to a question
- 2. Separate observations, claims, inferences and judgments
- 3. Build a claim network rather than a flat evidence list
- 4. Expose evidence dependencies and circular support
- 5. Test causal links, timing and mechanism
- 6. Worked example: explain a fictional supply recovery
- 7. Compare narrative structure across alternatives
- 8. Analyse omissions, silences and boundary choices cautiously
- 9. Identify framing, metaphor and actor roles
- 10. Use counter-narratives without turning analysis into advocacy
- 11. Calibrate confidence at claim and network level
- 12. Revise the narrative and preserve the change record
Choose a reading route
- Route 1: Define the narrative — question, audience and boundary
- Route 2: Build the network — claims, evidence and dependencies
- Route 3: Work the case — causal sequence and alternatives
- Route 4: Challenge coherence — omissions, framing and counter-narratives
- Route 5: Revise responsibly — confidence, indicators and updates
How to use this guide
Start with the chapters that match your immediate decision, then return to the worked example and audit chapters. Each chapter separates a model from the evidence needed to use it responsibly. A claim is not strengthened merely by being technical; it is strengthened when assumptions, observations, alternatives and limits can be inspected.
| Layer | Question | Minimum record |
|---|---|---|
| Concept | What exactly is being represented? | Definitions and boundary |
| Method | How will the reasoning be carried out? | Steps, inputs and assumptions |
| Transfer | Would it work in a changed setting? | A deliberately altered case |
| Audit | What could prove it wrong? | Evidence, limits and revision trigger |
Chapter 1. Frame the narrative as an answer to a question
The idea
A narrative selects events and links them into an account for an audience. Analysis begins by naming the decision question, time boundary and narrator rather than treating the story as neutral fact.
A disciplined method
Write the core question, identify narrator and audience, state the proposed beginning and ending, and list what the account excludes.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which decision the story is being used to support. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
A report says a policy “caused recovery”. The analytic question is which mechanisms and counterfactuals support that claim.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Change the audience from internal planners to the public and compare emphasis, certainty and omitted detail.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Purpose and audience shape a narrative but do not alone prove manipulation or error.
Evidence to retain: versions, authorship, intended audience, publication time and decision context
Common misconception: A chronological account is automatically objective.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 2. Separate observations, claims, inferences and judgments
The idea
Observations are reported or measured items; claims assert propositions; inferences connect them; judgments express an assessed conclusion. Mixing levels hides where uncertainty enters.
A disciplined method
Tag every sentence by analytic role, preserve quotations and raw measures, and rewrite compound claims into testable propositions.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which step adds interpretation beyond the source. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
A sales figure is an observation; “demand strengthened” is an inference; “the strategy succeeded” is a broader judgment.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Transfer to eyewitness testimony and distinguish what was perceived, remembered, interpreted and reported.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Observations still depend on measurement and sourcing; the categories clarify, not purify, evidence.
Evidence to retain: tagged text, original sources, definitions and analyst notes
Common misconception: Facts need no provenance while only opinions require scrutiny.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 3. Build a claim network rather than a flat evidence list
The idea
A claim network maps support, contradiction, qualification and dependence among propositions. It reveals load-bearing claims and unsupported jumps.
A disciplined method
Create nodes for atomic claims and evidence, draw typed edges, identify conclusions with many downstream dependants, and mark uncertainty on both nodes and edges.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which claim would collapse the largest part of the account. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
One source supports both “capacity rose” and “deliveries accelerated”; those claims jointly support “production expanded”. The shared source limits independence.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Remove the shared source and recalculate which conclusions remain supported.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Graph complexity can create false precision if edge meanings and confidence are vague.
Evidence to retain: node definitions, edge types, provenance, revisions and disagreement log
Common misconception: More incoming arrows always mean stronger support.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 4. Expose evidence dependencies and circular support
The idea
Reports may copy one origin, rely on the same dataset or cite each other. Circular support makes a narrative appear corroborated without adding independent observation.
A disciplined method
Trace each item to its earliest recoverable source, map shared methods and datasets, and distinguish direct observation from repetition and commentary.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is how many independent observation paths reach the claim. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
Three briefings cite two news stories that quote one official release. The network contains one origin, not six confirmations.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Transfer to multiple models trained on the same dataset and identify methodological dependence.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Independence is rarely absolute; shared context and partial dependence require qualitative judgment.
Evidence to retain: source graph, timestamps, dataset lineage, transformations and access records
Common misconception: Different authors guarantee independent evidence.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 5. Test causal links, timing and mechanism
The idea
Narratives often move from sequence to causation. A defensible causal link needs temporal order, a plausible mechanism, alternatives and evidence that differs across explanations.
A disciplined method
For each causal edge, state the mechanism, expected indicators, lag, confounders and counterfactual; compare with at least one rival mechanism.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which observation would distinguish mechanism from coincidence. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
Hiring rises after a subsidy. The narrative tests financing, demand and seasonal mechanisms rather than assuming the subsidy caused the change.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Move the policy date earlier while outcomes remain fixed and assess which causal account weakens.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Observational evidence may not identify one cause, and complex outcomes can have several interacting causes.
Evidence to retain: timeline, mechanism evidence, comparison group, confounders and sensitivity analysis
Common misconception: Because A preceded B, A caused B.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 6. Worked example: explain a fictional supply recovery
The idea
A fictional firm reports that a new coordination policy restored deliveries. Operations data, supplier messages and public claims partly overlap.
A disciplined method
Construct claim nodes, trace sources, test timing, compare policy, seasonal demand, inventory release and reporting-change explanations, then state what remains unresolved.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which evidence is diagnostic across the four explanations. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
Delivery time falls after the policy, but inventory was released the same week. Independent supplier lead times support a partial operational improvement.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Change the metric definition mid-series and revise the evidential weight of the apparent improvement.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: The example teaches method and does not establish facts about a real company.
Evidence to retain: raw series, definitions, messages, source graph, alternative matrix and revision history
Common misconception: A coherent account with a plausible mechanism needs no competing explanation.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
| Claim | Direct support | Dependency |
|---|---|---|
| Delivery time fell | Operations series | Metric definition |
| Suppliers responded faster | Supplier logs | Shared platform clock |
| Policy caused recovery | Timing and mechanism | Inventory release alternative |
| Recovery is durable | Later observations | Future demand |
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 7. Compare narrative structure across alternatives
The idea
Alternative narratives should explain the same core evidence, not cherry-pick different subsets. Comparing plot, actors, mechanism and predicted next events exposes diagnostic differences.
A disciplined method
Hold the evidence set constant, write two or more concise accounts, identify shared claims, then list observations that each expects and cannot easily explain.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which future or hidden observation separates the accounts. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
One account predicts supplier lead times continue improving; another predicts gains vanish when released inventory is exhausted.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Transfer to a case with two interacting causes and construct a combined account without making it unfalsifiable.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: An elegant alternative may still lack evidence; comparison organises judgment but does not automate it.
Evidence to retain: parallel narratives, common evidence set, unexplained items and indicators
Common misconception: Alternatives must be mutually exclusive to be useful.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 8. Analyse omissions, silences and boundary choices cautiously
The idea
What a narrative leaves out can matter, but absence becomes evidence only when the narrator had reason, ability and opportunity to include it.
A disciplined method
List excluded periods, actors and outcomes; ask whether exclusion follows scope, missing access, privacy, irrelevance or strategic framing.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is whether an omission changes the conclusion or merely expands context. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
A success story reports averages but omits distribution. The missing subgroup pattern materially changes the justice claim.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Change the report mandate so subgroup data are legally protected and reassess the meaning of omission.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Suspicion based on silence can become unfalsifiable when inclusion expectations are not stated.
Evidence to retain: scope documents, data availability, disclosure rules, drafts and author explanation
Common misconception: Anything omitted was deliberately concealed.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 9. Identify framing, metaphor and actor roles
The idea
Labels and metaphors assign agency, responsibility and inevitability. “Wave”, “failure”, “rescue” and “natural” can shape causal interpretation before evidence is evaluated.
A disciplined method
Extract key labels, passive constructions and actor roles; rewrite the account with neutral alternatives and observe which inferences change.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which wording carries an unstated causal or moral premise. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
Calling a price rise a “shock” foregrounds external cause; calling it “mark-up” foregrounds firm choice. Evidence must decide between them.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Translate the narrative and examine where metaphors or agency do not survive.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Language analysis cannot determine factual truth on its own and should not replace source evaluation.
Evidence to retain: original language, translation, alternative wording, actor map and reviewer notes
Common misconception: Removing emotive words removes all bias from a narrative.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 10. Use counter-narratives without turning analysis into advocacy
The idea
A counter-narrative surfaces neglected evidence and alternative agency, but it should face the same evidential standards as the dominant account.
A disciplined method
Invite a separate analyst to build the strongest rival account, score evidence symmetrically, and document which claims survive both.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is whether the challenge introduces new evidence or merely reverses emphasis. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
A community account adds unpaid care effects missing from the official efficiency story; both accounts use the same budget data.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Change the challenger’s source access and note which disagreements are evidential versus value-based.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: A counter-narrative can contain its own omissions, dependencies and rhetorical shortcuts.
Evidence to retain: commissioning question, evidence parity, source access, disagreements and synthesis
Common misconception: The marginal or contrarian account is automatically more truthful.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 11. Calibrate confidence at claim and network level
The idea
A narrative may contain well-supported observations but a weak final synthesis. Confidence should follow source quality, independence, coherence, alternatives and missing evidence at each layer.
A disciplined method
Rate load-bearing claims, identify the weakest necessary edge, state overall confidence and explain why high-confidence details may not justify the broad conclusion.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which uncertain edge limits the whole judgment. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
Delivery records are high confidence, the mechanism is moderate and durability is low; the overall success narrative is therefore moderate, not high.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Strengthen one weak edge with independent evidence and show exactly which downstream judgments change.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Ordinal confidence labels are communication aids, not mathematical probabilities unless calibrated data support them.
Evidence to retain: claim ratings, edge rationale, source independence, alternatives and review sign-off
Common misconception: Overall confidence equals the average confidence of all facts.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 12. Revise the narrative and preserve the change record
The idea
Responsible revision states what changed, why, which claims moved and what remains stable. Silent rewriting erases learning and can exaggerate foresight.
A disciplined method
Set indicators and review dates, version the claim network, compare old and new judgments, and retain superseded accounts with reasons.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which new observation would reverse or narrow the conclusion. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
When supplier gains fade after inventory exhaustion, the policy narrative narrows from “restored deliveries” to “improved coordination temporarily”.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Change the update to evidence that contradicts only one subgroup and avoid overcorrecting the entire account.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Frequent revision can confuse readers without clear versioning and confidence language.
Evidence to retain: dated networks, indicator log, new sources, change summary and distribution record
Common misconception: A revised assessment should hide earlier errors to avoid undermining trust.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Field workbook
Take a short public report and convert one paragraph into a claim network. Mark observations, inferences, causal edges and shared sources. Write a rival account using the same evidence and specify one indicator that would favour each.
- Write the decision question in one sentence and name the decision owner.
- List observations separately from interpretations, with dates and provenance.
- Record at least two plausible alternatives before choosing a preferred account.
- Run the changed-context test and note which assumptions fail.
- State the limit, confidence and next review trigger in plain language.
| Layer | Question | Evidence |
|---|---|---|
| Claim | Is it atomic and testable? | Definition and source |
| Edge | What connects the claims? | Mechanism and alternatives |
| Network | Where is dependence? | Source lineage |
| Revision | What would change? | Indicator and version history |
Continue through the verified ecosystem
- Competing hypotheses — compare explanations with diagnostic evidence
- Source triangulation — map independence and circular reporting
- Timeline analysis — test sequence and missing evidence
- Link analysis — entities, relations and missing links
- Deception analysis — test deliberate shaping without paranoia
Frequently asked questions
Is narrative analysis literary criticism?
It can borrow attention to framing and structure, but intelligence use also maps claims, sources, mechanisms, alternatives and revision triggers.
What is a claim network?
A graph whose nodes are claims or evidence and whose typed edges record support, contradiction, qualification or dependence.
Why analyse source dependence?
Several reports may repeat one origin or dataset, creating an illusion of corroboration.
Does coherence prove truth?
No. A story can be internally coherent while relying on false premises, omitted evidence or weak causal links.
What makes an alternative narrative fair?
It should explain the same core evidence and face the same source and confidence standards.
How should a narrative be revised?
Version the network, state what new evidence changed and preserve the earlier account and reasoning.
Authoritative sources and further reading
- CIA Tradecraft Primer — structured techniques and alternative analysis
- ODNI Intelligence Community Directive 203 — objectivity, sourcing, alternatives and uncertainty
- Psychology of Intelligence Analysis — mental models, evidence and bias
- CIA Center for the Study of Intelligence monographs — official analytic tradecraft library
- Sherman Kent and the Profession of Intelligence Analysis — validity, diagnosticity and open-minded assumptions
This article teaches defensive analysis of public or authorised information. It does not validate allegations, profile individuals or replace legal, investigative or domain-specific review.