threat intelligence protective intelligence

Protective-intelligence case products need the same source and uncertainty controls as threat assessments.

7 min read

The seam

Threat-intelligence quality standards touch protective-intelligence case management when sourced observations become a bounded assessment, management action, review or closure decision.

What you leave with

Give every intake, assessment, plan and review product a decision audience, information cut-off and next review point.
Separate source report, observed fact, inference and judgement so a reviewer can reproduce the assessment path.
Adopt a calibrated probability lexicon and state whether each underlying standard is binding, internally adopted or informative.
Convert every management action into an owner, intended effect and observation measure.
Require transfer acceptance and record reopening criteria before case ownership ends.
Assure samples, corrections, decision use and equity without claiming a count of attacks prevented.

A referral arrives with a screenshot and a conclusion: the sender is dangerous. The intake lead copies that conclusion into the case system. A later assessment repeats it. Human resources then restricts access, while the next review records that the case remains high concern. Nobody can recover who saw what, when the screenshot was captured, or which step turned a report into a judgment.

This failure travels. A loose sentence in an intelligence note may misdirect a decision. The same sentence in a protective case can change a person’s employment, education, liberty or access to support. Every later product inherits the first product’s uncertainty unless you preserve the analytic path.

Control the whole case record as six linked products. Quality collapses when tradecraft applies only to the assessment memorandum. Intake, assessment, management plan, review note, transfer or closure record, and programme assurance report form one six-product chain. Each product can change a decision.

Put the decision audience, purpose, information cut-off and next review point on every product. For intake, the cut-off is the receipt time and the audience includes the triage decision-maker. For an assessment, it is the last information considered and the authority choosing management. The plan goes to each action owner. The review note goes to whoever can continue, change or end those actions.

This stops a bounded judgment becoming a timeless label. The October 2024 NTAC framework uses a six-step unit model and treats assessment as management guidance rather than prediction. Your record should therefore say which current decision it supports and which change forces another look.

Preserve four evidential layers. Case systems encourage analysts to compress a source report, observed fact, inference and judgment into one narrative. Compression conceals the point where uncertainty entered. It also makes correction needlessly broad.

Keep the four layers visible. “A colleague reported that the employee asked about the director’s schedule” is a source report. A preserved message showing the question is an observed fact about that message. “The employee may be seeking access information” is an inference. “Further inquiry is likely warranted before changing access” is a judgment for a named decision.

Attach origin, date, handling and retrieval details to each decisive source. The FBI advises recipients to preserve electronic threats with subject, sender, date and time. ICD 206 recognises four sourcing mechanisms for covered US Intelligence Community products. ICS 206-01, dated 2 December 2024, extends citation rules to public and commercially available information, including AI-related material.

Those instruments bind their stated US Intelligence Community actors. A company can adopt their source-traceability controls internally. Otherwise they remain informative. Put that status in your standards register beside the rule, owner, effective date and exception route.

Use one probability lexicon and a separate confidence statement. Words such as possible, probable and unlikely drift between readers. A concern label adds a second problem because urgency, likelihood and potential harm become one colour.

Adopt one controlled mapping. ICD 203 contains five analytic standards and nine tradecraft standards. Its seven probability bands span 01–05% through 95–99%. The directive binds the US Intelligence Community. Your corporate policy becomes binding internally only when the authorised owner adopts the mapping and defines where it applies.

State confidence separately. The UK assessment approach treats probability and Analytic Confidence Ratings as two uncertainty dimensions. A judgment can carry higher likelihood and low confidence when sources are sparse. Record the gaps, assumptions, plausible alternatives and observations that would change it. Editors must preserve the terms through summaries and dashboards.

Show how instruments influenced the judgment. A score creates false precision when reviewers cannot see its evidential role. WAVR-21 has 21 items. Its initial reliability work used 11 raters and 12 workplace case summaries, with a psychologist-subgroup coefficient of 0.76. Those figures measure inter-rater reliability. An actuarial attack probability requires separate validation.

Name the instrument version, intended population, unknown items and method limits. Then explain which findings changed the formulation and record the remaining findings as neutral or inconclusive. A 2025 systematic review found no prospective designs among the included violent-extremism tool studies and could not recommend one tool over another for particular populations and contexts. Method selection therefore needs a recorded rationale tied to population, context and evidence.

Turn each action into a testable intervention. “Monitor the person” and “refer to support” fail because completion can be claimed without any defined effect. The next review then counts activity while learning nothing about whether management changed the scenario.

For every action, name the owner, authority or consent, intended effect, due condition, observation measure, review trigger and fallback. Link it to the person, potential target, situation or setting, the four management domains identified by the FBI. A support referral might aim to restore regular contact. The measure is whether contact occurs and continues under the agreed arrangement. An access change might aim to remove an identified route to a target. The measure is whether credentials and physical access actually change.

Completion and effect need separate fields. A sent referral records completion. Accepted contact records an early effect. A missed contact, renewed communication, target change, boundary testing or sudden silence after scrutiny can force earlier review. The owner reports the observation; the case authority decides whether the formulation changes.

Make every review a new bounded decision. Review notes often repeat the old concern label and list completed tasks. That practice hides contrary evidence, missed actions and drift in the original assumptions.

Record what changed since the previous cut-off. Include new sources, action effects, changed circumstances, rival explanations, dissent, current likelihood, confidence and the next trigger. NTAC’s October 2024 guide directs units to monitor progress, intervention effects and life changes before closure. The review should show which observation altered which decision.

Treat a material error as a correction. Preserve the superseded product, identify affected judgments and actions, and notify the original decision audience. ICD 203 requires annual standards training and product-evaluation programmes within the US Intelligence Community. An internally adopted corporate control needs its own cadence, sample rule and correction owner.

Require acceptance before transfer and criteria before closure. Ownership disappears when one organisation sends a file and assumes another has taken it. Jurisdiction, employment or enrolment may already have changed, leaving approaching triggers with nobody watching.

The transfer record names both owners, the disclosure basis, current assessment, active actions, approaching triggers, restrictions and records location. Keep the sending owner visible until the receiver accepts. In Virginia public higher education, § 23.1-805(F) requires specified written notices within 24 hours after its preliminary articulable-and-significant-threat determination. That jurisdiction-bound deadline does not govern the transfer rule in subsection J.

Closure needs the same precision. State which concern factors changed, what uncertainty remains, which other processes continue and what observations reopen the case. Renewed contact, changed access or failure of a central assumption are usable criteria. “Safe” is a permanent label that the evidence cannot support.

Assure decisions, corrections and equity. A clean template proves little when analysts cannot retrieve sources or managers ignore actions. Sample products by case risk and product type. Test the source trail, uncertainty wording, overdue review, correction handling, access history and evidence that the stated audience used the product.

Examine equity across the full decision path. A review of 24 school threat-assessment articles published from 2007 through 2017 found poor coverage of cultural competence. A 2025 Florida study then examined 18,411 assessed students across 1,646 schools. Approximately 1% received law-enforcement action, with no reported disparity on the characteristics studied. That result belongs to that study and outcome. Your assurance report needs its own defined denominators from referral through reopening.

Do not invent a count of attacks prevented. The missing counterfactual cannot show which event would have happened. Report whether the six products were timely, sourced, reviewed, corrected and used. Those controls can be reproduced, challenged and improved. They also show whether intervention burden falls unevenly while the case is still open to correction.

Questions answered

What controls should appear on every protective-intelligence case product?

Every case product should state its decision audience, purpose, information cut-off, sources, uncertainty, owner, dissemination limits and next review point. The judgment should be bounded to the recorded information and circumstances.

How should evidence and analysis be separated in a case record?

A case record should distinguish the source report, the observed fact, the inference drawn from it and the professional judgment that informs a decision. Each layer should retain its source reference, date and relevant limitation.

What probability and confidence controls belong in a protective-intelligence assessment?

A protective-intelligence assessment should use one controlled probability lexicon and express analytic confidence separately. The lexicon should map each probability term to a range, while confidence should reflect the quality, consistency and gaps in the supporting evidence.

Does ICD 203 bind a corporate protective-intelligence team?

ICD 203 binds the United States Intelligence Community. A corporate team may adopt named controls from it through internal policy or use it as an informative model. Its standards register must state the actual status.

What does ICD 203 require for uncertainty language?

ICD 203 defines seven probability bands spanning 01–05% through 95–99% and requires likelihood to be distinguished from analytic confidence. Those requirements bind the United States Intelligence Community and may be adopted explicitly elsewhere.

How should a structured professional judgment instrument appear in an assessment?

An assessment should name the instrument, version, intended population, unknown items and role in the conclusion. A score can represent attack probability only when valid evidence establishes that use. WAVR-21 has 21 items, and its development study reported initial inter-rater reliability.

What fields make a threat-management action testable?

A testable management action names the owner, lawful authority or consent, intended effect, due condition, completion evidence, observation measure, review trigger and fallback. It also identifies the person, target, situation or setting addressed.

What should a protective-intelligence review note record?

A review note should record participants, date, method, new sourced information, completed or missed actions, observed intervention effects, changed circumstances, rival explanations, revised judgments, decisions, owners and the next review trigger.

When is a protective-intelligence assessment corrected rather than merely updated?

A material error in a source, fact, inference or judgment requires a correction. The correction should identify the affected product and decisions, preserve the superseded version, notify the original decision audience and state whether management actions must change.

What prevents ownership loss during a protective-intelligence case transfer?

A transfer record should name the sending and receiving owners, disclosure basis, current assessment, active actions, approaching triggers, information restrictions and records location. The sending owner remains responsible until the receiving owner explicitly accepts the case.

What belongs in protective-intelligence case reopening criteria?

Reopening criteria should be observable and tied to the closed formulation. Examples include renewed communication, changed target access, boundary testing, a failed support contact, a new loss or evidence that a closure assumption was wrong.

How should a protective-intelligence programme measure equity?

Equity assurance should compare referral, triage, assessment, intervention, law-enforcement action, closure and reopening patterns across relevant groups, with defined denominators and lawful data handling. A 2025 Florida study examined 18,411 assessed students at 1,646 schools and found approximately 1% received law-enforcement action without reported disparity on the studied characteristics.

Can a protective-intelligence programme report attacks prevented?

A programme should exclude counts of attacks prevented because the counterfactual is ordinarily unknowable. Assurance can measure source traceability, timely review, action completion, intervention effects, corrections, decision use, access events and equity outcomes.