THE GOLD STANDARD OF INTELLIGENCE · SPECIALIST STUDY
Collect to resolve a decision, not to accumulate material
Intelligence begins with a decision that must be informed under uncertainty. A collection requirement translates that decision into observable questions, sources, timing and stopping rules. Without that discipline, teams gather what is easy, mistake repetition for corroboration and discover too late that the decisive gap was never addressed.
This guide uses benign organisational and public-information examples. It does not teach covert access, evasion or intrusive surveillance. The gold standard is lawful, necessary and proportionate collection, explicit source validation, protected privacy and an analytic record that shows what remains unknown.
Start with the closest route. This guide belongs to the Intelligence subject guide, the Gold Standard of Education, and the How X Works learning library.
Collection and source-validation audit
Can the requirement change a decision?
Write the available actions, deadline and evidence that would move the choice. If every plausible answer leads to the same action, stop. If one answer would reverse the choice, raise its priority. This keeps collection tied to consequence rather than curiosity or institutional habit.
Is the indicator genuinely discriminating?
Place it against at least two rival explanations. Record whether the observation is expected, possible or surprising under each. Evidence shared by every explanation may describe context but does not discriminate. Seek the cheapest lawful observation with the largest difference in expectation.
Was absence actually observable?
Name the source, channel, period and reporting reliability that should have produced the signal. A missing public update means little if publication is optional or delayed. Negative evidence becomes stronger only when detection would have been likely under the hypothesis.
Are sources independent enough to corroborate?
Draw lineage back to first-hand access, records or measurements. Syndicated reports, shared databases and briefings may all repeat one origin. Count origins rather than documents and note common incentives or infrastructure that could propagate the same error.
What could the source know?
Assess access and competence for this exact claim. A senior official may know policy but not implementation; an operator may see a local event but not the wider pattern. Reputation cannot replace a source–claim assessment. Record whether access is known, inferred or merely asserted.
Which privacy and civil-liberties limits apply?
State purpose, lawful basis, minimum fields, access controls, retention and deletion. Prefer direct consented or public evidence over inference from personal traces. If the needed evidence cannot be collected proportionately, report the gap rather than treating safeguards as obstacles.
What is the leading analytic gap?
Describe the unknown, why it matters, attempted collection, next lawful option, owner and decision effect if unresolved. Distinguish a missing fact from uncertainty about the model and from an unstated value trade-off. Collection cannot repair the latter two by itself.
What rival explanation remains live?
Give the strongest alternative a fair test. State what it predicts, what evidence has been sought and what observation would raise or lower it. An alternative should not be a ceremonial sentence; it should influence tasking and confidence.
When should collection stop?
Use a deadline, sufficiency criterion, low expected value, privacy burden or inability to observe. Record the reason. Stopping is not failure when the decision is robust or further collection would cost more than the uncertainty it could reduce.
How will later evidence update learning?
Preserve the forecast, confidence, sources and gaps as they stood at decision time. When outcomes arrive, compare them without hindsight. Update source assessments and model assumptions only to the extent the case supports; one success or failure should not become a permanent reputation.
Choose a reading route
Frame the requirement · chapters 1–3
Plan collection · chapters 4–5
Validate sources · chapters 6–8
Analyse gaps · chapters 9–10
Transfer and govern · chapters 11–12
The whole problem at a glance
| Object | Purpose | Failure signal |
|---|---|---|
| Decision question | Defines choice and horizon | Interesting facts cannot change action |
| Collection requirement | Names the observable needed | Task says find everything |
| Source assessment | Tests access, competence and incentives | Credibility becomes a permanent score |
| Gap register | Shows missing evidence and consequence | Unknowns vanish from the judgement |
Every stage must remain traceable. A polished assessment cannot repair a requirement that never targeted the information needed for a decision.
CHAPTER 1 OF 12 · Frame the requirement
1. Start with the decision and its deadline
State who must decide what, by when and which alternatives are live. Understand supplier risk is too broad. By Friday, decide whether to delay launch, add a supplier or proceed, and assess a four-week interruption, is usable.
List evidence that could change the choice. If every answer leads to the same action, collection has no decision value. If one missing fact reverses the plan, elevate it. Intelligence is consequential uncertainty reduced, not pages delivered.
Record boundaries. Do not collect personal or confidential material merely because it is accessible. Use lawful public, contractual or consented sources; minimise attributes; set retention; involve the proper safeguarding, privacy or legal authority.
Set a decision time and an evidence time. Material arriving after the choice may support learning but cannot justify the original action. This distinction prevents retrospective confidence from rewriting what was known.
CHAPTER 2 OF 12 · Frame the requirement
2. Translate the question into indicators and observables
Break a judgement into competing explanations and observable implications. A late shipment could reflect port congestion, production failure, documentation error or reallocation. Each predicts different evidence.
An indicator is not proof. Define direction, time window, threshold and confounders. Fewer updates is weak; two missed contractual updates in five days while other customers receive them is testable.
Build an indicator matrix with hypotheses as rows and observations as columns. Seek evidence expected under one explanation but not rivals. Ten signals shared by every hypothesis add detail without discrimination.
State observability. Absence matters only if the source, channel and period should have revealed the signal. A missing scan in an unintegrated system is not proof that no handover occurred.
CHAPTER 3 OF 12 · Frame the requirement
3. Write collection requirements that can be answered
Name the information need, acceptable sources, scope, time range, confidence, deadline and format. State why the answer matters so a collector can propose a safer substitute when the preferred observable is unavailable.
Prioritise by decision impact, urgency, likelihood of success and cost or harm. A high-impact gap with no lawful path remains a constraint, not an invitation to improvise. Low-impact curiosities should not consume decisive time.
Define completion: two independent sources, a primary record, deadline expiry or low expected value from more effort. Without stopping rules, collection expands while analysis waits.
Write rejection criteria too. Information outside scope, with unverifiable provenance or gathered through an unauthorised method should not silently enter the evidence base.
CHAPTER 4 OF 12 · Plan collection
4. Build a plan with alternatives and safeguards
For each requirement, list primary and fallback sources, owner, due time, access basis, limits and validation method. Sequence cheap public checks before costly requests. Prefer a direct consented source over speculative inference.
Plan for noise and non-response without assuming deception. A supplier may be unable to share data for contractual reasons; a dashboard may lag. Label collection failure separately from the hypothesis.
Include minimum necessary fields, secure handling, role-based access, deletion date and escalation criteria. Ethical limits are part of quality because evidence without permission or context can be unreliable and harmful.
Estimate expected value of information. A costly task is justified only if plausible results can change the action enough to outweigh time, burden and risk.
CHAPTER 5 OF 12 · Plan collection
5. Coordinate collectors and analysts without contaminating evidence
Collectors need context but should not be pushed toward a preferred answer. Share hypotheses and indicators, not a demand to confirm management. What explains the delay invites alternatives; find proof the factory failed creates pressure.
Analysts report the most discriminating observations and update priorities. Collectors report access limits, source conditions and ambiguities rather than supplying a cleaned conclusion. The handoff preserves content and provenance.
Use a shared log: requirement ID, task, time, source class, raw reference, handling limits, result and quality note. Deduplicate before counting corroboration. Five summaries from one release are one lineage.
Keep analysis and tasking in conversation. New evidence may close a question or reveal a more valuable one, but scope changes need a recorded reason and the same safeguards.
CHAPTER 6 OF 12 · Validate sources
6. Assess access, competence, incentives and track record
Source quality is multidimensional. Access asks whether the source could observe; competence whether it could interpret; incentives what it gains; track record how comparable reporting performed. Timeliness and specificity also matter.
Do not turn dimensions into an eternal reputation. A logistics coordinator may know dispatch records but not factory output. A source reliable last year may face changed incentives. Assess source–claim pairs.
Separate source credibility from information plausibility. A credible source can err; an unknown source can supply verifiable primary evidence. Validate through records, consistency, independent lineage and predicted consequences.
Record uncertainty in the assessment itself. If access is inferred rather than known, say so. Source grades should compress evidence only after the underlying reasons remain available.
CHAPTER 7 OF 12 · Validate sources
7. Distinguish corroboration from repetition
Corroboration requires sufficiently independent access or method. Two articles quoting one spokesperson share a lineage. Two databases fed by one report also share a lineage. Draw a source tree before counting agreement.
Agreement can arise from copying, common incentives or a shared blind spot. Disagreement can arise from different timestamps, definitions or vantage points. Reconcile scope before inferring dishonesty.
Prefer triangulation across modes: a primary document, operational record and independent observation. Still record dependencies. Independence is rarely absolute; understand how error can propagate.
Weight contradictions by relevance and quality, not seniority. Preserve the dissenting observation and identify what new evidence would resolve it.
CHAPTER 8 OF 12 · Validate sources
8. Work a delayed-shipment example
A school expects laboratory equipment. The distributor says goods left the factory, a carrier portal shows label created, and a port bulletin reports congestion. The choice is postpone practical lessons, source locally or wait. The decisive gap is physical custody.
Request a carrier acceptance scan or bill of lading, ask for handover time, check whether the bulletin covers the terminal, and obtain local lead times. The portal and distributor are not independent if the distributor created the label.
If no custody record arrives, report dispatch unconfirmed; congestion plausible but unlinked; one-week arrival estimate low confidence. Choose a reversible local substitute rather than treating absence of proof as proof of failure.
After delivery, compare outcome with the recorded forecast. Late evidence updates source and model learning; it must not be used to pretend the earlier judgement was more certain.
CHAPTER 9 OF 12 · Analyse gaps
9. Maintain a live analytic gap register
Name the unknown, why it matters, current evidence, attempted collection, next lawful option, owner, deadline and consequence if unresolved. Rank by decision sensitivity, not embarrassment. Some unknowns can remain because the choice is robust.
Distinguish factual, model and value gaps. Facts concern what happened; model gaps how an observation relates to outcome; value gaps the decision-maker’s priorities. More collection cannot answer an unstated trade-off.
Update as evidence arrives. Closing dispatch may open customs clearance. Do not erase closed gaps; preserve why collection stopped and how the judgement evolved.
Include a no-further-collection rationale. It may cite deadline, marginal value, privacy burden or inability to observe. Explicit stopping is more defensible than silent abandonment.
CHAPTER 10 OF 12 · Analyse gaps
10. Express confidence and alternatives without false precision
Confidence reflects quality, consistency, relevance and model stability—not personal certainty. State judgement, confidence, support, gaps and the condition most likely to change the view. Avoid percentages without a defensible basis.
Keep a live alternative. Ask what evidence would be expected and whether collection searched where it could appear. Missing signals inform only when observable.
Use calibrated language. More likely than not and plausible but unsupported are different. When evidence cannot distinguish explanations before deadline, design a reversible decision instead of manufacturing resolution.
Separate confidence in evidence from confidence in forecast. Facts may be well established while their future consequence is model-sensitive. Reporting both prevents one score from hiding two uncertainties.
CHAPTER 11 OF 12 · Transfer and govern
11. Transfer the method to a school attendance concern
Missed lessons could reflect illness, timetable error, disengagement or data entry. The decision is correct records, offer support or escalate safeguarding. Seek minimum necessary school records and respectful conversation, not broad monitoring.
Register edits suggest a data problem; cross-subject absences a wider issue; a timetable collision another. None is character evidence. Source validation includes who entered the record and when.
The context raises duties of care, confidentiality and power. A learner’s account deserves attention without forced disclosure. Immediate safety follows safeguarding procedures; analytic discipline does not replace professional duty.
Report uncertainty in language that avoids stigma. The goal is support and accurate action, not prediction of a child’s character from incomplete traces.
CHAPTER 12 OF 12 · Transfer and govern
12. Publish a collection-and-analysis audit trail
Link each judgement to requirement IDs, source descriptions, quality factors, gaps and alternatives. Protect identities and sensitive details; transparency does not require public exposure.
Review collection value after the decision. Which tasks changed the judgement, arrived too late or repeated existing information? Feed learning into future requirements without turning one outcome into a permanent rule.
Set review triggers: a missed milestone, new primary record, changed source access or model failure. Intelligence is a revisable state, not a sealed verdict.
The final standard is disciplined humility: explain what was asked, why collected, what sources could know, what remained missing, how uncertainty affected action and when to revisit.
Practice and evidence check
- Rewrite a broad request as decision, deadline and three alternatives.
- List one observation that discriminates a hypothesis from its rival.
- Draw the lineage of three agreeing reports.
- Assess one source–claim pair across access, competence, incentives and timeliness.
- Name one unknown that would not change the decision and stop collecting it.
Practice is complete when every task has a decision purpose, lawful access basis, validation plan and stopping rule.
Frequently asked questions
Is more information always better?
- No. Late, repetitive, intrusive or non-decisive information can reduce clarity and cause harm.
Does a reliable source make every claim reliable?
- No. Assess access and competence for the specific claim.
Are two agreeing reports corroboration?
- Only if their informational lineages are sufficiently independent.
What if a decisive gap cannot close?
- Report it, lower confidence and prefer a reversible or robust decision.
Can this work in schools?
- Yes, with minimum necessary collection, consent and safeguarding.
Continue through the ecosystem
The Gold Standard of Intelligence — the subject root
How Intelligence Works | Source Monitoring — track claim origins
How Intelligence Works | Model Discrimination — separate rival explanations
Post-Decision Evidence Integration — learn after action
How Intelligence Works | Meta-Uncertainty — reason about uncertainty assessments
Sources and further reading
- ODNI Intelligence Community Directive 203 — Analytic Standards
- ODNI Intelligence Community Directives
- ODNI IC OSINT Strategy 2024–2026
- ODNI framework for commercially available information
Use sources to inspect definitions and limits, not as decoration. Apply methods in ways that fit the actual decision, data and people affected.