.svg){kind=icon}
Learn
Cross-INT correlation is the process of determining when records from different intelligence disciplines - SIGINT, GEOINT, HUMINT, OSINT - refer to the same real-world entity or track, then linking them into one assessed, provenance-bearing object an analyst can trust.
Last updated: October 18, 2026
Why it matters to your mission. The same person, track, or unit appears across SIGINT, GEOINT, HUMINT, and OSINT under different identifiers, and the connection that matters is the one most easily missed. Correlating those records into one assessed, provenance-bearing object closes that gap. The outcome is a faster, more complete picture an analyst can trust, with the links across disciplines made explicit instead of assumed.
Cross-INT correlation is the process of determining when records from different intelligence disciplines - SIGINT, GEOINT, HUMINT, OSINT - refer to the same real-world entity or track, then linking them into one assessed, provenance-bearing object an analyst can trust.
This is the technical companion to the pillar What Is Multi-INT Fusion?. The pillar explains what fusion is and why it matters; this page explains how correlation actually works under the hood, where it breaks, and what a mission should require of it.
Key takeaways
Different disciplines describe the same object in incompatible ways. A SIGINT emitter, a GEOINT location, a HUMINT name, and an OSINT handle may all point to one person or platform, but nothing in the raw records says so. Each arrives with its own identifiers, its own spelling and transliteration conventions, its own timing, and its own classification, and none was built to line up cleanly with the others.
Independent oversight has documented how hard this is in practice. The U.S. Government Accountability Office found that analysts routinely perform the integration by hand, describing a process in which "analysts then collect, assess and package the data by manually entering and transferring them into yet another system," and that "data may be classified at different levels or have different structures and formats" (U.S. Government Accountability Office, Defense Command and Control, GAO-25-106454, 2025). The same report notes that military departments "continue to pursue their command and control projects largely in isolation," so the data an analyst needs to correlate is spread across federated systems rather than a single repository, with "overly restrictive data classification" adding further friction.
Correlation is the step that turns many partial descriptions into one defensible assessment. That assessment is the basis of the all-source intelligence product doctrine expects: an integrated judgment drawn from every available discipline rather than a single source read in isolation (U.S. Joint Chiefs of Staff, Joint Publication 2-0, Joint Intelligence, 2013). Doing it well is what separates a trustworthy track from a pile of unmatched reports.
Different identifiers, one resolved entity
A SIGINT emitter, a GEOINT location, a HUMINT name, and an OSINT handle may all point to one platform, but nothing in the raw records says so. Cross-INT correlation aligns them into one entity.
GAO found analysts routinely do this integration by hand, re-entering data into yet another system. Automated alignment keeps provenance on every contributing source.
Cross-source entity alignment is the decision that two records from different sources describe the same entity. It compares identifiers, attributes, time, and location across disciplines and resolves three outcomes: a match, a near-match that needs review, and a conflict that has to be surfaced rather than hidden.
The comparison only works if the records can be compared on equal terms in the first place. A SIGINT selector, a HUMINT name, and an OSINT registration are written in different vocabularies, so alignment depends on first giving each source a common semantic representation. With that in place, an alignment becomes a visible, inspectable link between records rather than a hidden inference buried in a score. This is the same matching discipline described for a single record type in entity resolution for intelligence analysis; cross-INT correlation applies it across disciplines, where the heterogeneity and the adversary's attempts to defeat it are at their sharpest.
Co-display puts multiple sources on one screen and leaves the correlation to the analyst's eye. An assessed, or correlated, track is a single object the system has judged to be one real-world entity across sources, carrying the evidence and the confidence behind that judgment. Co-display shows data together; an assessed track asserts, with traceability, that the data is about the same thing.
The difference is not cosmetic. Co-display scales badly: as the number of sources grows, the analyst is asked to hold more simultaneous comparisons in their head, and the connection most easily missed is the one that matters. An assessed track moves that comparison into the system, where it can be made consistently, scored, recorded, and challenged. The table below separates the two so each can be evaluated on its own.
| Dimension | Co-display | Assessed (correlated) track |
|---|---|---|
| What it is | Sources shown together on one view | One object judged to be a single real-world entity across sources |
| Who correlates | The analyst, by eye | The platform, with analyst review |
| Confidence | Implicit | Explicit and recorded |
| Provenance | Per source, not linked | Linked across sources, auditable |
| Reasoning | Manual | Graph traversal across find-fix-track-target-engage-assess |
| Scaling | Degrades as sources multiply | Holds as sources multiply |
| Failure mode | Missed or double-counted entities | Surfaced conflicts and near-matches for review |
When entities, observations, and relationships live in a graph, reasoning becomes a traversal. An analyst can connect an emitter to a location, a location to a platform, and a platform to an intent, following the find-fix-track-target-engage-assess sequence as a path through linked evidence rather than as a set of disconnected lookups across separate systems.
Graph-based reasoning changes what the analyst is asked to trust. Instead of accepting an opaque correlation score, they can follow and test each link in turn: this emitter was observed at this location, this location matches this track, this track belongs to this platform. Every step stays tied to the source observation underneath it, so a weak link in the chain is visible as a weak link rather than averaged away. The same traversal that builds the assessment is the one a reviewer walks to challenge it.
Every correlation records which source records were linked, by which logic, and when. An analyst can open a correlated track and see why two observations were joined, and a reviewer can audit the link or reverse it. Provenance is not a report generated afterward; it is the structure the correlation is built on, so it survives as the track is enriched with more sources over time.
That persistence depends on the underlying repositories staying interoperable rather than each hand-off flattening the sources into a single merged record. This is the government-owned, interoperable-repository model the Chief Digital and Artificial Intelligence Office describes under Open DAGIR (Open Data and Applications Government-owned Interoperable Repositories): combining data from many sources into a decision-ready picture while the government retains ownership of its data (Chief Digital and Artificial Intelligence Office, Open DAGIR Fact Sheet, 2024). When the repositories stay interoperable, provenance survives across sources instead of being lost at each hand-off, and the correlated track remains something a commander can defend.
Consider a single platform of interest observed by four disciplines that never reference each other. A SIGINT collection reports an emitter active on a particular night. A GEOINT track places a vehicle at a known compound on the same night. A HUMINT report names an individual associated with that compound under a local spelling. An OSINT pull surfaces a business registration tying a slightly different spelling of that name to the same address.
Handled as separate records, this is four weak, disconnected leads, each easy to dismiss on its own. Cross-INT correlation does the connective work. Semantic representation puts the four sources into comparable terms; cross-source entity alignment proposes that the HUMINT name and the OSINT registration describe one person and that the emitter and the vehicle track are co-located in time; graph traversal links the person to the compound, the compound to the vehicle, and the vehicle to the emitter; and the whole assessment is recorded as one track with a confidence score and the conflicting spellings flagged for analyst review rather than silently merged.
The failure modes are just as instructive. Correlate too aggressively and the system fuses two different platforms into one track, and the assessment becomes fiction. Correlate too timidly and the four leads stay disconnected, and the pattern is never seen. That tension, between missing a correlation and manufacturing one, is why confidence scoring, provenance, and analyst adjudication are the core of doing this responsibly, not optional extras.
Torch.AI builds cross-INT correlation as government-owned reasoning infrastructure that complements existing systems of record rather than replacing them. NEXUS gives each source a common semantic representation, so a SIGINT selector, a HUMINT name, and an OSINT registration can be compared on equal terms instead of in four incompatible vocabularies. HALO expresses the resolved entities and their relationships as a graph, so a correlation is a visible, traceable link an analyst can follow and a reviewer can reverse, and so kill-chain reasoning runs as a traversal rather than an opaque score. ORCUS keeps the underlying repositories available and interoperable, consistent with the Open DAGIR model, so provenance survives across sources and the authoritative data stays intact.
Because the reasoning layer sits on top of the customer's existing systems rather than replacing them, the correlated track is portable and inspectable rather than trapped in a proprietary model. In a government-owned deployment, the customer keeps control of the layer where data becomes an assessed track, and can inspect it, govern it, and evolve it. You can see how this is packaged as a capability on the Torch.AI software page.
The distinctions above explain why correlation has to be more than co-display; the checklist below is what to actually require in an evaluation.
Evaluating a capability? The seven requirements above are the backbone of a cross-INT correlation evaluation you can score against your own environment: request a technical walkthrough.
Each discipline describes the same object in its own vocabulary, with partial identifiers, conflicting details, different timing, and different classification. Nothing in the raw records says a SIGINT emitter, a GEOINT track, a HUMINT name, and an OSINT handle are the same entity, and the records usually live in federated systems that were never built to line up. The GAO has documented analysts doing this integration by hand across systems classified at different levels.
Cross-source entity alignment is the decision that two records from different sources describe the same entity. It compares identifiers, attributes, time, and location across disciplines and resolves matches, near-matches, and conflicts, after first giving each source a common semantic representation so the attributes can be compared on equal terms.
Co-display puts multiple sources on one screen and leaves correlation to the analyst's eye. An assessed track is a single object the system has judged to be one real-world entity across sources, carrying the evidence and confidence behind that judgment, so it can be inspected, scored, and reversed rather than re-derived by hand each time.
With entities, observations, and relationships in a graph, reasoning becomes a traversal: connect an emitter to a location to a platform to an intent across the find-fix-track-target-engage-assess sequence, with every inference tied to its source observation so the analyst can follow and test each link instead of accepting an opaque score.
Every correlation records which source records were linked, by which logic, and when, so an analyst can see why two observations were joined and a reviewer can audit or reverse it. Keeping the underlying repositories interoperable, consistent with the Open DAGIR government-owned model, lets provenance survive across sources instead of being flattened at each hand-off.
Describe the problem.
Define the constraints. We will structure the session.
No demo.
No slide deck.
Real work.
Cross-INT correlation turns partial reports into one defensible assessment. We will build it on your data. No manual re-entry. Provenance on every link. Real work.
