Methodology
How we turn 65,000 incidents
into actionable impact intelligence.
The data pipeline, classification framework, and proprietary metrics behind every Hackmanac product. Documented in full, because intelligence you can't audit is just an assertion.
Data pipeline
From source to score, in five steps.
Every incident in the Hackmanac dataset goes through the same five-step pipeline. It's deliberately slower than automated scraping, because the quality of every downstream metric depends on the quality of the inputs.
Collect
Continuous monitoring of open sources, regulatory filings, dark web channels, and verified industry feeds.
Verify
Cross-reference against multiple independent sources. Unconfirmed reports are flagged and held, not published.
Classify
Manually tagged across 200+ variables: actor, motivation, technique, target industry, sector, geography, and impact.
Score
ESIX© severity calculated. IM, QSI, and IEM updated. Outliers re-reviewed by senior analysts.
Publish
Published on HackRisk.io with full details. Used in HackLoss to calculate typical loss estimates.
Classification framework
200+ variables per incident. Every one curated by hand.
The dataset's depth is what makes everything else possible. Each incident is tagged across nine major dimensions, with sub-categories down to the technique and impact level. Below, a representative selection: not exhaustive.
Threat actor
- Actor class (state, criminal, hacktivist, insider)
- Named group attribution
- Attribution confidence level
- Operational tempo and history
Motivation
- Financial, espionage, sabotage, hacktivism
- Sponsorship or alignment
- Stated vs. observed intent
Technique
- Initial access vector
- Exploitation chain
- Tools, malware families, infrastructure
- MITRE ATT&CK mapping
Target sector
- Primary industry (19 categories)
- Sub-sector (142 sub-categories)
- Government, critical infrastructure, sensitive industry
Geography
- Country and region
- Regulatory jurisdiction
- Cross-border impact
Target profile
- Organization size and revenue band
- Public vs. private
- Military target designation
Impact
- Operational, financial, technical, legal, reputational
- Disclosed loss figures, where available
- Recovery duration
Temporal
- Date of incident, discovery, disclosure
- Dwell time
- Pattern with concurrent incidents
Provenance
- Primary and secondary sources
- Verification status
- Analyst confidence rating
FOUNDATIONAL METRIC
ESIX© - Estimated Severity Index
ESIX© compresses the multi-dimensional impact of a cyber attack into a single 1–10 absolute score. The score weighs operational disruption, financial loss, technical compromise, legal and regulatory consequences, and reputational damage, independently from dollar losses. Each dimension is scored independently against documented criteria, then combined using a weighted function calibrated against years of historical incidents.
This enables unique severity comparisons to be made between different threats, sectors and geographical areas.
ESIX© is incident-level: every attack in the Hackmanac dataset carries its own score. Aggregations at the sector, country, or actor level are weighted means based on thousands of incidents, never modeled or extrapolated.
Acceleration +42% vs. prior 4-quarter baseline.
THREAT TEMPORAL DYNAMICS
IM - Impact Momentum
Impact Momentum tells you whether things are getting worse, in which context, and how fast.
It quantifies the rate of change in aggregate severity across a chosen dimension: a sector, a country, a technique, or a threat actor.
IM can distinguish between accelerating and stabilising threats, even when the number of attacks remains unchanged compared to a previous period. It indicates where threats are escalating, regardless of attack counts alone.
THREAT ANOMALY DETECTION
QSI - Quiet Shift Index
QSI is an early warning indicator. This anomaly-detection metric identifies subtle changes in threat levels that are not yet visible on traditional dashboards. It alerts you to potential new threats before they escalate.
The index works by comparing current severity patterns against expected baselines across multiple dimensions simultaneously. When several signals correlate, even if they are not highly significant individually, QSI identifies the anomalous risk concentration as an emerging threat.
48 individual incidents below visibility threshold; cluster signal above 87th percentile.
STRATEGIC PRIORITY MAPS
IEM - Impact–Exposure Matrix
IEM maps every observed incident by industry, geography, and technique against two axes: their relative frequency, and their average impact.
It shows which threats generate critical impact in your context, based on the losses experienced by your peers.
IEM provides answers to the questions that arise with every budget cycle, such as 'Where exactly are we exposed?', 'Where will spending actually make a difference?' and 'What are our tail risks'?
IEM enables you to move beyond abstract risk assessments and make effective strategic decisions based on thousands of verified data points.
Quality assurance
Human analysts in every loop.
Senior OSINT experts manage collection and surfacing, handling every judgment call. The classification framework, the severity scoring, the attribution confidence levels: all of them are decisions made by senior analysts, not models.
Multi-analyst review
High-severity incidents (ESIX© 5+) require sign-off from at least two senior analysts before publication. Disagreements are escalated and documented.
Retrospective re-scoring
As new information emerges after an incident, prior scores are revised and the change tracked. The dataset is versioned and auditable.
Confidence tagging
Every classification carries an analyst confidence rating. Low-confidence tags are filterable, so users can choose their own quality threshold.
Limitations & disclosures
What our data can do, and what it can't.
Honest intelligence reports its own limits. The Hackmanac dataset is the deepest of its kind that we know of, but it isn't omniscient. The areas below are where we ask users to apply judgment, and where we're actively working to extend coverage.
Disclosure-dependent visibility
Attacks that are never disclosed never enter the dataset. Regulatory environments with mandatory breach disclosure (EU NIS2, US SEC rules) produce richer data than environments without. Sector-level conclusions in lightly regulated regions should be read accordingly.
Attribution confidence
Threat actor attribution is inherently probabilistic. Where attribution is contested or uncertain, the dataset records the uncertainty rather than forcing a single answer. Filter by confidence level when actor identity matters to your analysis.
Loss figure availability
Disclosed financial loss figures exist for roughly 60% of incidents. The remaining 40% carry estimated ranges derived from comparable incidents. Loss estimations from our Threat Impact Simulator© account for this and report confidence intervals accordingly.
Geographic coverage
The dataset is strongest in Europe, North America, the Middle East, and East Asia. Coverage in sub-Saharan Africa, parts of South America, and parts of Central Asia is improving but remains thinner. Country-level statistics from those regions should be interpreted as lower bounds.
The methodology is the moat.
Six years of curated research, 65,000+ classified incidents,
and four unique metrics built on top.
If you'd like to see how it applies to your specific context, let's talk.
Want to use our data in your own products or services? See our partner options