Skip to main content

Preview build: sign-in and grading run on the server. MFA is not enabled, and storage is in server memory so it does not survive a restart.

LearnDefend
Threat Hunting Methodology
TheoryMedium14 minThreat HuntingLog Analysis

Building a Falsifiable Hypothesis

What is it?

A hunting hypothesis is a specific, falsifiable statement of the form 'an adversary is doing X, which would leave evidence Y in source Z'. Falsifiable means there is a clear result that would prove it wrong — without that, the hunt can never end.

Why it matters

The quality of a hunt is set before any query runs, by the hypothesis. A vague hypothesis ('find evil') returns endless anomalies and no conclusion. A precise one bounds the data, defines success and failure in advance, and makes the result defensible.

Where you see it

Hypotheses are seeded from three places: threat intelligence about a specific actor's techniques, the MITRE ATT&CK matrix as a checklist of what to look for, and anomalies noticed during normal work. Each seed is turned into a concrete, source-anchored claim.

What normal looks like

A good hypothesis names a technique, a data source and an expected artefact, and can be answered within the data you actually hold. If you cannot even say what a 'no' looks like, the hypothesis is not yet ready to hunt.

What suspicious looks like

A hypothesis is worth escalating in priority when it targets a technique the environment is exposed to, is grounded in current intelligence, and would produce high-fidelity evidence — a rare event a benign process almost never creates.

How analysts investigate

Write the hypothesis, then write its expected evidence and its refutation before touching the data. Scope it to a time window and a source set you can actually query. Only then run it — and let the pre-written refutation, not your hope, decide the outcome.

Common beginner mistakes

  • Writing a hypothesis so broad ('detect lateral movement') that no single result could ever confirm or refute it.
  • Targeting a technique the available data cannot even show, then blaming the hunt when it finds nothing.
  • Adjusting the hypothesis mid-hunt to fit whatever the data happened to show.

What you will be able to do

  • Write a specific, falsifiable hypothesis naming a technique, source, artefact and time window.
  • Seed a hypothesis from threat intelligence, the ATT&CK matrix, or an observed anomaly.
  • Write the refutation before querying, so the hunt can end honestly.

Everything downstream — which data you pull, what counts as a hit, when you stop — is decided by the hypothesis. Spend the effort here. A sharp hypothesis makes the rest of the hunt almost mechanical; a blunt one guarantees an inconclusive week.

Weak hypothesisStrong, falsifiable hypothesis
"Find persistence.""A non-admin account created a scheduled task running from a user-writable path in the last 7 days (Event 4698)."
"Look for C2.""A host makes outbound connections at a fixed sub-5-minute interval to a rare external destination."

Worked example — sharpening a blunt hypothesis

Intelligence says an actor uses PowerShell to download payloads. A junior analyst writes the hypothesis: 'hunt for malicious PowerShell.' Before reading on, decide: why can that hunt never end?

The answer: it names no source, no artefact and no time window, so no result could confirm or refute it — it would return endless PowerShell and never conclude. Sharpen it: 'A non-admin ran powershell.exe with an encoded command that made an outbound download, in the last 7 days.' Now it has a source (endpoint), a concrete artefact, a window, and a refutation: no such executions, or all are known-good automation.

Recap

  • A strong hypothesis names a technique, a source, an artefact and a time window.
  • It is falsifiable — you can state, in advance, the result that proves it wrong.
  • Seed it from intelligence, ATT&CK, or an anomaly — but always make it concrete and bounded.

Sign in to save your progress on the server.