Gold Standard · Intelligence specialist guide
Find the nearest question, then follow the evidence
Change analysis asks what is different, compared with which baseline, for whom and over what time—then tests whether the apparent shift reflects the target, the observer or the measurement system.
A new data point is not automatically a change, and a change is not automatically important. Weekly activity can look unusual against yesterday but ordinary against the same season last year. A sudden increase in reports can reflect new collection access rather than new behaviour. A stable total can conceal a large shift between categories.
Intelligence change analysis creates an explicit baseline, defines comparable observations, describes deviations, tests possible drivers and retains alternative explanations. It complements indicator validation and timeline analysis: indicators watch for specified signs, timelines reconstruct sequence, while change analysis explains how a state, pattern or relationship differs from a defensible comparison.
Choose a reading route
Build the baseline
Question, comparison and time discipline.ROUTE 2 · CHAPTERS 4–5
Measure deviation
Comparable variables and change registers.ROUTE 3 · CHAPTERS 6–7
Work a case
Analyse a fictional shift and alternatives.ROUTE 4 · CHAPTERS 8–10
Test explanation
Drivers, indicators and judgement updates.ROUTE 5 · CHAPTERS 11–12
Transfer and audit
New contexts, provenance and review.
Open the full chapter index · Use the field check · Return to the Intelligence root
Chapter contents
Chapters 1–3 · Question, baseline and temporal control
Chapters 4–6 · Comparable measures and deviations
Chapters 7–9 · Drivers, alternatives and updating
Chapters 10–12 · Transfer, governance and audit
CHAPTER 1 OF 12 · Build the baseline
1. State the change question before selecting a baseline
A baseline is a comparison frame chosen for a stated question. The useful baseline for detecting a short operational disruption may differ from the one needed to assess a five-year structural shift. Selecting the comparison after seeing the result invites hindsight and can make ordinary movement look exceptional.
Write the target, variable, population, geography, interval and decision need first. List candidate baselines such as previous period, seasonal peer period, pre-event average, matched unit or modelled expectation. Select one by a rule tied to the question and preserve reasonable alternatives for sensitivity.
A fictional distribution centre ships 18 percent fewer parcels than last week. Against the same holiday week last year, volume is unchanged. The analyst states both comparisons and avoids describing a collapse until the decision context is clear.
Changed-context transfer check
Transfer the method to school attendance after a calendar change. Decide whether the previous week, previous term or matched weekdays provide a fair baseline.
Limit and misconception
No baseline is neutral. Structural breaks, policy changes and composition shifts can make historical averages misleading even when calculations are correct.
Evidence practice
Keep the written question, candidate baselines, selection rule, decision need and sensitivity set. Label every field as observed, computed, assumed or chosen. Give the record to a second reader and ask them to reproduce the result without the original author explaining what was meant. Differences reveal hidden definitions, missing versions or decision rules that still need to be made explicit.
Deliberately test for this failure: choosing the comparison that produces the most dramatic deviation after inspecting outcomes. Record the earliest point at which the process can detect it, the correction applied and whether later conclusions change. A method is transferable only when its controls work on an altered case, not merely on the example used to teach it.
CHAPTER 2 OF 12 · Build the baseline
2. Distinguish level, trend, seasonality and composition
Change can occur in the absolute level, growth rate, seasonal pattern, volatility or composition of a total. A flat aggregate may hide offsetting movements, while a rising series may merely continue an established trend. The analytic claim should name which property changed.
Plot raw levels and rates, compare matched seasonal periods, decompose important categories and inspect volatility and missingness. Use rolling summaries only with declared windows. Mark policy, sensor and classification changes on the same time axis before interpreting a breakpoint.
Total website visits remain stable, but mobile access rises while desktop access falls. The level is unchanged; composition has shifted. A service response based only on the total would miss the operational implication.
| Change form | Question | Common trap |
|---|---|---|
| Level | Is the amount different? | Ignoring trend |
| Rate | Did growth accelerate or slow? | Small denominator |
| Composition | Did the mix change? | Stable total |
| Volatility | Did uncertainty widen? | Focusing only on average |
Changed-context transfer check
Apply the distinction to energy use in hotter weather. Separate demand growth from temperature, customer mix and tariff effects.
Limit and misconception
Statistical decomposition is a model, not a discovery of true causes. Short histories and changing behaviour can make seasonal estimates fragile.
Evidence practice
Keep raw series, definitions, category shares, windows, event markers and decomposition assumptions. Label every field as observed, computed, assumed or chosen. Give the record to a second reader and ask them to reproduce the result without the original author explaining what was meant. Differences reveal hidden definitions, missing versions or decision rules that still need to be made explicit.
Deliberately test for this failure: calling a continuing trend a new change or letting offsetting categories disappear inside one total. Record the earliest point at which the process can detect it, the correction applied and whether later conclusions change. A method is transferable only when its controls work on an altered case, not merely on the example used to teach it.
Contents · Previous chapter · Next chapter
CHAPTER 3 OF 12 · Build the baseline
3. Separate target change from collection and reporting change
Observed records are produced by a collection system. New sensors, changed access, analyst attention, reporting incentives, definitions or source duplication can alter the data even when the underlying target is stable. Change analysis must model this observation process.
Maintain a collection diary with sources, coverage, outages, tasking, definitions and processing revisions. Compare stable-source subsets with the full series and examine observation opportunity by group and time. Label retroactive updates separately from events that actually occurred later.
Reports of equipment faults double after a mobile reporting app launches. A stable sample of inspected equipment shows no comparable deterioration. The analyst treats the increase as partly a reporting-system change rather than immediate proof of worsening reliability.
Changed-context transfer check
Change the case to social-media data after an API restriction. Explain why fewer observed posts do not establish lower public attention.
Limit and misconception
Collection-effect correction is often incomplete because unobserved coverage is unknown. Bound the conclusion instead of manufacturing a precise adjusted series.
Evidence practice
Keep source coverage, collection changes, stable subsets, processing versions and observation opportunities. Label every field as observed, computed, assumed or chosen. Give the record to a second reader and ask them to reproduce the result without the original author explaining what was meant. Differences reveal hidden definitions, missing versions or decision rules that still need to be made explicit.
Deliberately test for this failure: attributing an observed jump to the target while the measurement system changed at the same time. Record the earliest point at which the process can detect it, the correction applied and whether later conclusions change. A method is transferable only when its controls work on an altered case, not merely on the example used to teach it.
Contents · Previous chapter · Next chapter
CHAPTER 4 OF 12 · Measure deviation
4. Make variables comparable before calculating deviation
A deviation is meaningful only when numerator, denominator, unit, population and time window are comparable. Counts can rise because the population at risk grew; rates can swing because the denominator is small; averages can shift because the group composition changed.
Create a comparison contract for each variable: definition, unit, denominator, inclusion rule, time basis, source and revision status. Recompute earlier periods under the current rule when defensible, or keep a visible discontinuity. Report absolute and relative change together when each answers a useful question.
Incidents rise from 20 to 25 while operating hours rise from 1,000 to 1,500. The count increases 25 percent, but the rate falls from 20 to 16.7 per thousand hours. Both observations are true and imply different follow-up questions.
Changed-context transfer check
Transfer the calculation to a survey whose respondent mix changes. Design a stratified comparison rather than treating two raw percentages as equivalent.
Limit and misconception
Normalisation can hide real burden when the denominator itself is part of the problem. Keep counts and affected people visible alongside rates.
Evidence practice
Keep variable definitions, units, denominators, inclusion rules, source versions and calculations. Label every field as observed, computed, assumed or chosen. Give the record to a second reader and ask them to reproduce the result without the original author explaining what was meant. Differences reveal hidden definitions, missing versions or decision rules that still need to be made explicit.
Deliberately test for this failure: comparing unlike populations or reporting a percentage change without its starting value. Record the earliest point at which the process can detect it, the correction applied and whether later conclusions change. A method is transferable only when its controls work on an altered case, not merely on the example used to teach it.
Contents · Previous chapter · Next chapter
CHAPTER 5 OF 12 · Measure deviation
5. Build a deviation register instead of one dramatic chart
A deviation register records what moved, how much, compared with what, when it emerged, which sources support it and how sensitive it is to alternative baselines. It prevents attention from selecting only the most visually striking pattern.
For each candidate change, record direction, magnitude, duration, affected subgroup, baseline, uncertainty, source quality, collection caveat and current disposition. Rank follow-up by relevance and information value, not by colour or z-score alone. Keep reversed and unresolved deviations.
A dashboard shows a sharp one-day spike and a modest six-week drift. The register reveals that the spike came from duplicated records, while the drift persists across sources and merits investigation.
Changed-context transfer check
Apply the register to a network where density rises after entity-resolution rules change. Separate graph-processing effects from relational change.
Limit and misconception
A register organises evidence; it does not make incomparable deviations commensurate or replace domain expertise about materiality.
Evidence practice
Keep the deviation rows, baselines, uncertainty, source notes, dispositions and correction history. Label every field as observed, computed, assumed or chosen. Give the record to a second reader and ask them to reproduce the result without the original author explaining what was meant. Differences reveal hidden definitions, missing versions or decision rules that still need to be made explicit.
Deliberately test for this failure: following the largest visual spike while durable smaller changes and data revisions go unexamined. Record the earliest point at which the process can detect it, the correction applied and whether later conclusions change. A method is transferable only when its controls work on an altered case, not merely on the example used to teach it.
Contents · Previous chapter · Next chapter
CHAPTER 6 OF 12 · Work a case
6. Worked example: explain a fictional logistics shift
A fictional relief warehouse records slower dispatch, more split shipments and longer vehicle dwell time over six weeks. The question is whether the operation has changed enough to threaten delivery reliability and which explanation should be tested first.
The analyst builds matched seasonal and pre-policy baselines, normalises by order size, checks data-system changes and maps the deviations. Large orders increasingly arrive late from one supplier, packing teams split them across waves and vehicles wait for final pallets. Weather and staffing alternatives remain plausible but fit fewer observations.
A policy change that grouped urgent and routine items created a queue interaction. The team pilots separate lanes, measures dwell time and watches substitution effects. The assessment states moderate confidence because two weeks of source data remain incomplete.
| Observation | Baseline result | Interpretation status |
|---|---|---|
| Dispatch time | Higher across two baselines | Sustained deviation |
| Split shipments | Higher for large orders | Possible mechanism |
| Vehicle dwell | Higher after policy change | Outcome to explain |
| Staffing | Within normal range | Alternative weakened, not disproved |
Changed-context transfer check
Change the case so supplier lateness is corrected but dwell time remains high. Identify which alternative explanation gains weight and what evidence should be collected next.
Limit and misconception
The warehouse is fictional. Real operational and security decisions require authorised data, responsible owners and safeguards for workers and affected communities.
Evidence practice
Keep baseline choices, normalised deviations, collection diary, competing explanations, pilot and update record. Label every field as observed, computed, assumed or chosen. Give the record to a second reader and ask them to reproduce the result without the original author explaining what was meant. Differences reveal hidden definitions, missing versions or decision rules that still need to be made explicit.
Deliberately test for this failure: selecting one plausible driver before checking whether it explains every important deviation. Record the earliest point at which the process can detect it, the correction applied and whether later conclusions change. A method is transferable only when its controls work on an altered case, not merely on the example used to teach it.
Contents · Previous chapter · Next chapter
CHAPTER 7 OF 12 · Work a case
7. Use alternative explanations to protect against narrative lock-in
A coherent story can fit a change without being the best explanation. Alternatives should include target behaviour, collection effects, definitional change, random variation, external conditions, strategic adaptation and interacting causes. They become useful when each makes distinguishable predictions.
List alternatives before final wording, identify evidence expected if each were true and seek diagnostic rather than merely consistent evidence. Use a matrix to compare observations, note dependence between sources and preserve hypotheses that remain viable. Invite review from someone not invested in the leading account.
For rising returns, alternatives include lower product quality, changed return policy, different customer mix and duplicated transactions. A policy timestamp and stable product-test results shift weight toward the second explanation without proving it exclusively.
Changed-context transfer check
Transfer the method to a falling participation rate. Add one explanation based on numerator change and one based on denominator or eligibility change.
Limit and misconception
Alternative analysis reduces premature closure but cannot enumerate every cause. Interacting mechanisms may make several explanations partly true.
Evidence practice
Keep the hypothesis list, diagnostic predictions, evidence matrix, source dependencies and review comments. Label every field as observed, computed, assumed or chosen. Give the record to a second reader and ask them to reproduce the result without the original author explaining what was meant. Differences reveal hidden definitions, missing versions or decision rules that still need to be made explicit.
Deliberately test for this failure: adding token alternatives that cannot change the judgement or counting repeated reporting as independent support. Record the earliest point at which the process can detect it, the correction applied and whether later conclusions change. A method is transferable only when its controls work on an altered case, not merely on the example used to teach it.
Contents · Previous chapter · Next chapter
CHAPTER 8 OF 12 · Test explanation
8. Distinguish drivers, mechanisms, triggers and background conditions
A driver influences the direction or magnitude of change; a mechanism explains how that influence produces observed effects; a trigger marks onset; background conditions shape vulnerability. Blurring these roles turns correlation into causal language and weakens collection priorities.
Draw a causal sketch with temporal order, necessary links and competing pathways. Identify evidence for each edge and look for cases where the driver changes without the outcome. Test lag, dose, subgroup and intervention patterns while keeping confounding and feedback visible.
A price increase precedes lower demand, but a concurrent service outage affects the same region. The analyst separates trigger timing from the mechanism and compares unaffected regions before assigning causal weight.
Changed-context transfer check
Apply the framework to a rapid migration from one platform to another. Identify background conditions that made the trigger effective.
Limit and misconception
Observational evidence rarely isolates one mechanism. Causal diagrams clarify assumptions but do not create experimental identification.
Evidence practice
Keep the causal sketch, timestamps, pathway evidence, counterexamples, confounders and intervention results. Label every field as observed, computed, assumed or chosen. Give the record to a second reader and ask them to reproduce the result without the original author explaining what was meant. Differences reveal hidden definitions, missing versions or decision rules that still need to be made explicit.
Deliberately test for this failure: calling a preceding event the cause while mechanism, comparison and competing pathways remain untested. Record the earliest point at which the process can detect it, the correction applied and whether later conclusions change. A method is transferable only when its controls work on an altered case, not merely on the example used to teach it.
Contents · Previous chapter · Next chapter
CHAPTER 9 OF 12 · Test explanation
9. Connect deviations to indicators and signposts without circularity
A change analysis may reveal measurable signs worth monitoring, but an indicator should be defined prospectively with a rationale, threshold, source and response. If the same data both invent and validate the sign, apparent performance will be exaggerated.
Translate a tested mechanism into observable indicators, reserve separate periods or cases for validation and specify base rates, lead time and false-positive costs. Distinguish early signposts from confirmation and document what action each threshold triggers.
The warehouse case identifies persistent supplier lateness plus growing split shipments as a joint warning. The team validates the combination on earlier seasons before using it as an operational threshold.
Changed-context transfer check
Change the example so only one indicator fires. Decide whether to collect more evidence, take a reversible action or revise the rule.
Limit and misconception
Indicators monitor a specified model of change. Targets can adapt, relationships can drift and thresholds can become obsolete.
Evidence practice
Keep the mechanism-to-indicator map, development set, validation set, base rates, thresholds and actions. Label every field as observed, computed, assumed or chosen. Give the record to a second reader and ask them to reproduce the result without the original author explaining what was meant. Differences reveal hidden definitions, missing versions or decision rules that still need to be made explicit.
Deliberately test for this failure: defining an indicator from one anomaly and reporting success on the same episode. Record the earliest point at which the process can detect it, the correction applied and whether later conclusions change. A method is transferable only when its controls work on an altered case, not merely on the example used to teach it.
Contents · Previous chapter · Next chapter
CHAPTER 10 OF 12 · Test explanation
10. Update analytic judgements and explain what changed
A revised assessment should identify whether new evidence changed the estimated state, probability, timing, scope or confidence. Consistency is not a virtue when evidence changes, and revision is not failure when reasons and prior uncertainty remain visible.
Keep a judgement ledger with prior wording, date, confidence, key assumptions and expected indicators. On update, state the new evidence, source quality, affected assumption, unchanged elements and why confidence moved. Use calibrated estimative language and distinguish correction from ordinary learning.
An assessment moves from “likely temporary” to “likely persistent” after the deviation survives two seasonal comparisons and the policy mechanism is replicated at a second site. The update explains the change rather than rewriting history.
Changed-context transfer check
Transfer the ledger to a judgement whose confidence falls even though the central estimate stays the same. Explain how widening uncertainty can matter for action.
Limit and misconception
Versioned judgements can still become bureaucratic if analysts defend past positions. Review should reward transparent updating, not apparent infallibility.
Evidence practice
Keep prior and current judgement, assumptions, new evidence, confidence rationale and consumer-facing change note. Label every field as observed, computed, assumed or chosen. Give the record to a second reader and ask them to reproduce the result without the original author explaining what was meant. Differences reveal hidden definitions, missing versions or decision rules that still need to be made explicit.
Deliberately test for this failure: replacing an earlier judgement without showing why it changed or quietly shifting the claim being assessed. Record the earliest point at which the process can detect it, the correction applied and whether later conclusions change. A method is transferable only when its controls work on an altered case, not merely on the example used to teach it.
Contents · Previous chapter · Next chapter
CHAPTER 11 OF 12 · Transfer and audit
11. Transfer change analysis only after rebuilding comparability
A baseline, threshold or driver learned in one location may not transfer to another with different seasonality, population, incentives, collection access or institutional rules. The method transfers; the numerical comparison and explanation require rebuilding.
Restate the question, recreate variable definitions, test measurement equivalence and compare baseline distributions. Check whether the proposed mechanism and alternatives exist in the new setting. Use local validation, subgroup review and staged deployment before operational use.
A staffing indicator works in one warehouse with fixed shifts but fails at a site using contractors and dynamic rosters. The team retains the change register yet redesigns the variable and threshold.
Changed-context transfer check
Move the method from logistics to public-service demand. Identify which data represent people, eligibility, access barriers and institutional capacity.
Limit and misconception
Transfer failure may reveal a different system rather than poor data. Do not force comparability merely to preserve a universal dashboard.
Evidence practice
Keep the new comparison contract, distribution tests, mechanism review, local validation and subgroup results. Label every field as observed, computed, assumed or chosen. Give the record to a second reader and ask them to reproduce the result without the original author explaining what was meant. Differences reveal hidden definitions, missing versions or decision rules that still need to be made explicit.
Deliberately test for this failure: reusing historical thresholds across populations or systems whose observation processes differ. Record the earliest point at which the process can detect it, the correction applied and whether later conclusions change. A method is transferable only when its controls work on an altered case, not merely on the example used to teach it.
Contents · Previous chapter · Next chapter
CHAPTER 12 OF 12 · Transfer and audit
12. Audit change from raw record to revised judgement
A reproducible audit follows the evidence from source records through cleaning, baseline choice, deviation calculation, alternative analysis, driver testing and judgement update. It also shows where analyst choices and incomplete collection affect the conclusion.
Freeze source snapshots and code, reproduce calculations, rerun alternative baselines, trace every chart to rows and review excluded records. Check source dependence, missingness, version changes and time alignment. Require a second analyst to reconstruct the leading and strongest competing explanation.
A fictional audit finds that late-arriving records were included in the current period but not historical comparators. Reprocessing reduces the deviation but does not erase it; the judgement remains while confidence and magnitude are revised.
Changed-context transfer check
Design the audit for a source that cannot be retained verbatim. Specify hashes, queries, access logs and aggregated evidence that preserve review without violating controls.
Limit and misconception
Reproducibility cannot remove uncertainty or guarantee objectivity. It makes assumptions, transformations and disagreements inspectable.
Evidence practice
Keep source snapshots, transformation code, baseline variants, deviation register, hypothesis matrix and judgement ledger. Label every field as observed, computed, assumed or chosen. Give the record to a second reader and ask them to reproduce the result without the original author explaining what was meant. Differences reveal hidden definitions, missing versions or decision rules that still need to be made explicit.
Deliberately test for this failure: publishing a polished trend while data revisions, exclusions or alternative baselines cannot be recovered. Record the earliest point at which the process can detect it, the correction applied and whether later conclusions change. A method is transferable only when its controls work on an altered case, not merely on the example used to teach it.
Contents · Previous chapter · Connected articles
Advanced boundary cases
Structural breaks and regime change
When the generating process changes, one historical baseline can become irrelevant. Test breakpoints, connect them to evidence and consider separate regimes rather than averaging incompatible periods.
Multivariate and network change
Several variables or relationships may change together. Preserve interpretable component evidence and guard against multiple testing, boundary drift and entity-resolution changes before assigning a systemic explanation.
Adversarial adaptation
Targets may change behaviour when they learn indicators or collection methods. Monitor indicator decay, substitution and deception, and keep collection and analytic assumptions reviewable.
A practical field check
Use this check when an assessment claims that a state, pattern, behaviour or risk has changed.
- Write the change question and decision need.
- List and justify reasonable baselines.
- Separate level, rate, composition and volatility.
- Audit collection, reporting and definition changes.
- Create comparable variables and a deviation register.
- Test alternatives and causal pathways.
- Validate indicators outside the discovery episode.
- Version the judgement and explain each update.
Run the check as a small workbook
Use sheets for raw observations, collection history, comparison contracts, baseline variants, deviations, hypotheses, drivers, indicators and judgement updates. Every chart should trace to exact source rows and transformation versions.
| Gate | Pass condition | Stop condition |
|---|---|---|
| Baseline | Question and comparison rule are explicit | Post-hoc dramatic baseline |
| Measurement | Units, population and time are comparable | Definition or collection shift hidden |
| Explanation | Alternatives make testable predictions | One narrative fits everything |
| Update | Evidence and confidence changes are versioned | Earlier judgement silently replaced |
Chapter-by-chapter evidence audit
| Chapter | Evidence to keep | Failure to catch |
|---|---|---|
| 1. State the change question before selecting a baseline | The written question, candidate baselines, selection rule, decision need and sensitivity set | Choosing the comparison that produces the most dramatic deviation after inspecting outcomes |
| 2. Distinguish level, trend, seasonality and composition | Raw series, definitions, category shares, windows, event markers and decomposition assumptions | Calling a continuing trend a new change or letting offsetting categories disappear inside one total |
| 3. Separate target change from collection and reporting change | Source coverage, collection changes, stable subsets, processing versions and observation opportunities | Attributing an observed jump to the target while the measurement system changed at the same time |
| 4. Make variables comparable before calculating deviation | Variable definitions, units, denominators, inclusion rules, source versions and calculations | Comparing unlike populations or reporting a percentage change without its starting value |
| 5. Build a deviation register instead of one dramatic chart | The deviation rows, baselines, uncertainty, source notes, dispositions and correction history | Following the largest visual spike while durable smaller changes and data revisions go unexamined |
| 6. Worked example: explain a fictional logistics shift | Baseline choices, normalised deviations, collection diary, competing explanations, pilot and update record | Selecting one plausible driver before checking whether it explains every important deviation |
| 7. Use alternative explanations to protect against narrative lock-in | The hypothesis list, diagnostic predictions, evidence matrix, source dependencies and review comments | Adding token alternatives that cannot change the judgement or counting repeated reporting as independent support |
| 8. Distinguish drivers, mechanisms, triggers and background conditions | The causal sketch, timestamps, pathway evidence, counterexamples, confounders and intervention results | Calling a preceding event the cause while mechanism, comparison and competing pathways remain untested |
| 9. Connect deviations to indicators and signposts without circularity | The mechanism-to-indicator map, development set, validation set, base rates, thresholds and actions | Defining an indicator from one anomaly and reporting success on the same episode |
| 10. Update analytic judgements and explain what changed | Prior and current judgement, assumptions, new evidence, confidence rationale and consumer-facing change note | Replacing an earlier judgement without showing why it changed or quietly shifting the claim being assessed |
| 11. Transfer change analysis only after rebuilding comparability | The new comparison contract, distribution tests, mechanism review, local validation and subgroup results | Reusing historical thresholds across populations or systems whose observation processes differ |
| 12. Audit change from raw record to revised judgement | Source snapshots, transformation code, baseline variants, deviation register, hypothesis matrix and judgement ledger | Publishing a polished trend while data revisions, exclusions or alternative baselines cannot be recovered |
Connected articles across the eduKate ecosystem
Start with the governing route
- The Gold Standard of Intelligence
- The Gold Standard of Education | Learning That Transfers into Life
- How X Works hub
Continue into verified specialist studies
- Indicator Validation, Base Rates and Warning Thresholds — turn tested change mechanisms into monitored signs without circular validation.
- Timeline Analysis, Event Sequencing, Evidence Gaps and Competing Explanations — reconstruct event and report order before attributing change.
- Source Triangulation, Circular Reporting and Corroboration Networks — separate independent confirmation from repeated reporting.
- Competing Hypotheses, Diagnosticity and Evidence Matrices — test explanations using evidence that differentiates them.
- Link Analysis, Entities, Relationships, Brokerage and Missing Links — analyse structural change without confusing graph construction with target behaviour.
Questions readers often ask
Is every unusual value a meaningful change?
No. Compare it with a defensible baseline and test noise, seasonality, composition, reporting and collection effects.
Which baseline is best?
The one justified by the question and decision, tested against reasonable alternatives and not selected only because it makes the deviation large.
Can a stable total hide change?
Yes. Composition, distribution, volatility or relationships can move while the aggregate stays flat.
Does a change prove its proposed cause?
No. Drivers and mechanisms need temporal, comparative and diagnostic evidence against alternatives.
Should an assessment admit that it changed?
Yes. A versioned explanation of new evidence and confidence improves accountability and reader understanding.
Can thresholds transfer between settings?
Only after measurement equivalence, local baselines, mechanisms and performance are revalidated.
Sources and scope
Office of the Director of National Intelligence: ICD 203 Analytic Standards — official standards for objectivity, source quality, alternatives, uncertainty and explaining change.
CIA Center for the Study of Intelligence: A Tradecraft Primer — official structured-technique guidance, including indicators and signposts of change.
CIA Center for the Study of Intelligence: Tradecraft Primer overview — official publication context for structured analytic techniques.
CIA Studies in Intelligence: Intelligence Analysis, A Target-Centric Approach — official discussion of collaborative, evidence-centred analytic work.
CIA Center for the Study of Intelligence: Analytic Culture in the U.S. Intelligence Community — official research on analytic process, baselines and organisational assumptions.
The warehouse, service and operational examples are fictional. This guide teaches disciplined change analysis and does not authorise surveillance, attribution, personnel action or operational decisions outside applicable law and responsible review.
“Gold standard” describes the educational benchmark explained in this guide. It is not an official accreditation, legal opinion, policy determination or professional certification.
Back to contents · Return to the Intelligence root · Return to the Gold Standard of Education