DETECT hunt Reference

Templates

Ready-made pieces for a detection program: the log source inventory, a blank detection record and variables file, the intake ticket, the review checklist, the naming rules, the quarterly report, and the emulation plan and run report.

Copy and pasteFramework v0.136.0. Everything here is CC BY-NC 4.0 and free to use in your own program, so adapt it for your own wiki and keep the credit line.

The meetings and roles that use these templates are in Operating rhythm. For a small team, the three standing meetings add up to about three hours and 40 minutes a month, plus 45 minutes after each incident.

On this page

The templates

Log source inventory

Paste into a spreadsheet as CSV and replace the example rows. One row per log source. Fewer than 30 days_kept caps that source's telemetry readiness at 2, fields_parsed no caps it at 3, and enriched no or silence_alert no caps it at 4. The usual_gap_minutes column is the normal time between two events from the source, read from a week of data. DET-0000 alerts after three times that gap, and never sooner than 1 hour. Every no in silence_alert is DET-0000's work queue.
source,owner,days_kept,fields_parsed,enriched,usual_gap_minutes,silence_alert
Windows Security events from domain controllers,Identity team,90,yes,yes,1,yes
EDR process events,Endpoint team,30,yes,no,1,no
Entra ID sign-in logs,Identity team,30,no,no,5,no

Detection record

Save as detections/DET-NNNN/ads.md, or let python tools/new.py write it. Each comment says what lint checks under that heading. A finished record shows every section filled in.
# DET-NNNN Name of the detection

## Goal

<!-- The behavior, the asset it puts at risk, and who asked for the detection. -->

## Categorization

## Strategy Abstract

<!-- One sentence each. Hypothesis: when an attacker does X, log source Y records Z. -->
Hypothesis:

Log source:

## Technical Context

<!-- Filled at Research. Lint counts these four lines from the research stage on. Put the baseline count in
     score.yml as baseline_7d too: the promotion check fails a first week live with no matches when it is above 0. -->
Plan approved by:
Run tag:
Baseline:
Teardown verified:

## Blind Spots and Assumptions

<!-- Name one thing the detection cannot see. -->

## False Positives

## Validation

## Priority

## Response

<!-- Numbered steps an analyst can follow. Lint needs them from the validate stage on. -->

## Additional Resources

## Retirement record

Detection variables

Declare each threshold, window, allowlist, and scope once, with its evidence. Detections in SQL explains each kind.
# detections/DET-NNNN/variables.yml. lint.py checks each declaration: a kind the framework defines, a default,
# an evidence line, and bounds for a threshold, a window, or a severity cutoff.
variables:
  name_of_the_value:
    kind: threshold          # threshold, window, allowlist, scope, severity_cutoff
    type: integer
    default: 10
    bounds: [5, 50]
    evidence: the count or the query behind the default

Playbook record

Save beside the playbook's code. Response playbooks gives the ladder, the stages, and the triggers. The kit has no playbook tools yet.
# playbooks/PB-NNNN/playbook.md. One section per field; a playbook that closes alerts needs the last one filled.

# PB-NNNN: Name of the playbook

## Serves
<!-- The detection ids or the alert class the playbook runs on, and the alert volume per day it will see. -->

## Trigger
<!-- The alert field values that start it. Nothing else does, so an alert the trigger does not name goes to a person. -->

## Enrichment
<!-- Each lookup, its source, and the field it fills. A lookup that fails leaves its field empty and says so, and the playbook goes on. -->

## Decision
<!-- The rule that turns the enriched alert into one outcome: close as benign, escalate, or act. Every branch is written down, and the default branch is escalate. -->

## Action
<!-- What it does on each outcome, at which tier of the ladder, and the exact call it makes. -->

## Approval point
<!-- Which actions wait for a person, who that person is, and how long the action waits before it expires unacted. -->

## Rollback
<!-- The call that undoes each action, tested once against a real object, and who may run it. -->

## Manual baseline
<!-- The steps an analyst took by hand before the playbook, timed over at least 20 alerts: median minutes, and the outcome each one got. -->

## Fixtures
<!-- One past alert per outcome, with the enrichment results as they were, and the outcome the analyst recorded. -->

## What it gets wrong
<!-- The alert the decision rule will misread, with its name. A playbook that closes alerts has at least one. -->

Intake ticket

Paste into your detection ticket type as required fields. Nothing enters the backlog without them.
- Candidate title, one line naming the behavior
- ATT&CK technique id, and the tactic it sits under
- Where the idea came from: incident, threat report, red team finding, audit, request
- The six scores with one evidence line each
- Log sources the detection would read, and their readiness
- Who asked, and what they expect to happen when it fires

Review checklist

Paste into the pull request template. The reviewer works down it in order.
- [ ] The record has all 11 sections and the blind spots section says something specific
- [ ] The true-positive fixture produces an alert and the false-positive fixture does not
- [ ] The converted query matches the approved copy for every target
- [ ] The response steps could be followed at 3 a.m. by somebody who has never seen this detection
- [ ] Every tunable number is a declared variable with a default, bounds, and evidence
- [ ] The score's six evidence lines are claims somebody could check
- [ ] The alert volume threshold is set, and the detection deploys with alerting off

Naming and layout

Paste into the team's readme. Decide once, argue never again.
- Detection id DET-NNNN, allocated once and never reused, including after retirement
- Detection title states the behavior and the condition: what happened, under what circumstance
- One technique per detection where possible. A detection covering three techniques is three detections
- File layout detections/DET-NNNN/ carrying detection, record, score, fixtures, and golden queries
- Tags carry the ATT&CK technique and the detection id, so alerts join back to the record

Quarterly report

Paste into the quarterly report. Each line names the metric it reports, as Appendix C defines it, and the page ends with one ask.
- Coverage: robustness_weighted_coverage by tactic, beside the raw count of live detections
- Outcome: incident_detection_rate, the share of this quarter's incidents a live detection caught first
- Health: the median precision_30d across live detections, minutes_per_true_positive, and rule_debt
- Flow: deploy_lead_time_days, first_pass_validation_rate, and rules_retired_per_quarter
- Capacity: the three ceilings against the live count, and which one binds
- One ask, costed, tied to the ceiling that binds
- Misses: misses_by_cause, most frequent code first, with the fix each code got
- Loop: incidents_without_followup, each one named, with the responder who ran it
- Risk: each KRI against its tolerance, red ones first
- Objectives: last quarter's OKRs scored, and next quarter's

Threat model handoff

Add to the threat model template. Each unmitigated threat becomes one intake candidate carrying these fields.
- Service or system modeled, and the owner who signed the model
- The threat, in one line: what the attacker does and what they reach
- ATT&CK technique id, or ATLAS for an AI system
- Why it is unmitigated: no control, control planned, or control accepted as a gap
- Impact score for the service, 1 to 5, and the data class or blast radius behind it
- The log source that would show the behavior, and whether it is collected today
- Candidate id once filed, so the next model review sees what the backlog did with it

Emulation plan

Fill this in before anything is provisioned, and have a second person approve it.
- [ ] State the objective, and what the source says and leaves out.
- [ ] Describe the threat model: what the adversary is trying to achieve.
- [ ] Describe the victim model, with its resources, controls, identities, and network paths.
- [ ] Name the starting identity, and what the attacker has to earn later.
- [ ] Write the operational flow: each call, its actor, its target, the expected result, and the log it should land in.
- [ ] Estimate the cost if metered services are involved.
- [ ] Name the second person who will approve the plan before anything is provisioned.

Emulation run report

Write this after the run. It is the artifact that outlives the commands.
- Objective, scope, and source material
- Threat model and ATT&CK mapping
- Victim model and assumptions, each with the date it was true
- Operational flow: action, API call, identity, result
- Observations, including denials and identity progression
- Telemetry that appeared, and telemetry that did not
- Coverage outcome per detection that fired
- Infrastructure summary and teardown status
- Recommendation: new detection, tune, logging change, hunt, or more research

The review checklist and the emulation plan are written as checkboxes, so they can be ticked off in a pull request. Review the record before the query, because a reviewer who starts with the query tends to review syntax and miss the logic.