Dependency Analysis Before a Legacy Migration: Unsupported Components and Undocumented Interfaces in Government Systems
How a government program can inventory what a legacy system depends on before migrating it: declared, vendored, platform, and interface dependencies, support status for each, and a migration order derived from the result.
A legacy migration plan usually lists the applications to be moved. It less often lists what those applications depend on: the libraries compiled into them, the runtime and operating system underneath, the scheduled jobs that feed them, and the other systems that read their output. Those dependencies decide what can be moved first, what has to move together, and what will break if it is left behind.
What the guidance now expects
Three recent documents point the same way, each for a different layer of the system.
In July 2025, the Government Accountability Office (GAO) reported in GAO-25-107795 on the 11 federal legacy systems most in need of modernization. Four of the 11 run on unsupported hardware or software, and seven have known cybersecurity vulnerabilities. Only three agencies had modernization plans with all three elements GAO looks for: milestones, a description of the work, and details on the disposition of the legacy system.
In January 2026, the Office of Management and Budget issued memorandum M-26-05, which rescinded the earlier government-wide software attestation requirements. The inventory obligation stayed. The memo states that agencies “shall continue to maintain a complete inventory of software and hardware” and develop assurance policies that match their own risk determinations. It also says agencies may adopt contract terms requiring a producer to provide a current software bill of materials (SBOM) on request. An SBOM is a machine-readable list of the components inside a piece of software.
In February 2026, the Cybersecurity and Infrastructure Security Agency (CISA) issued Binding Operational Directive 26-02, which requires federal civilian agencies to inventory and decommission internet-facing network devices that have reached end of support, and within 24 months to establish continuous discovery so that devices are replaced on or before the date their support ends. The directive covers network edge devices only, but its method (find everything, record the support end date, act before that date) applies equally to the software inside a legacy application.
The underlying control is older than all three. Control SA-22 in NIST Special Publication 800-53, Revision 5 calls for replacing system components when the developer, vendor, or manufacturer no longer supports them, or arranging an alternative source of support when a component must stay.
Four kinds of dependency, and where each one hides
A package manifest is the usual starting point for dependency analysis. Legacy systems often predate package managers, so the manifest covers only part of the picture.
| Dependency class | Where it hides | How to find it | Question to answer |
|---|---|---|---|
| Declared libraries | Package manifests, build files, installer scripts | Manifest parsing, including transitive dependencies | Is each version still supported, and by whom? |
| Vendored and copied code | Source files copied into the repository, statically linked libraries, modified forks | Source and binary analysis, hash comparison against known releases | Which upstream version is this, and what was changed locally? |
| Platform | Compiler, language runtime, operating system, database engine, middleware, hardware | Build logs, deployment records, inspection of running hosts | What does the application assume about the platform that the target will not provide? |
| Interfaces | Batch file exchanges, shared database tables, message queues, scheduled jobs, reports consumed by other offices | Code analysis for input and output, job schedules, network and file transfer logs | Who sends data in, who reads data out, and what format do they expect? |
Vendored code does not appear in any manifest. A parsing or compression library copied into a source tree in 2008 will not be flagged by a scanner that reads only package files.
Interfaces are dependencies in the other direction: other systems depend on the one being migrated. A nightly fixed-width file sent to a partner office is a contract, even if the only record of it is a line in a job scheduler. If the new system produces the same data in a different layout, the migration has broken something outside its own boundary.
Record support status as a date
For each component found, the inventory should record four facts: the version in use, who supports it, the date support ends or ended, and what replaces it. “Unsupported” as a yes or no flag is less useful than a date, because a component with 14 months of support left needs a different plan than one that lost support in 2017.
Sorting the inventory by support end date shows which components have already passed end of support, which will pass it during the migration, and which will still be supported when the migration is done. The middle group is the easiest to miss: a three-year migration can begin with every component supported and end with a third of them expired.
Components that must stay past their support date need an entry of their own, consistent with SA-22: who provides alternative support, what compensating measures are in place (isolation, input filtering, reduced privileges), and when the decision will be reviewed.
Turn the inventory into a migration order
A dependency inventory is a graph: components are nodes, and “depends on” relationships are edges. Three properties of that graph are directly useful for sequencing.
- Clusters that must move together. If three applications share a database schema and write to the same tables, moving one of them alone means building and testing a temporary bridge.
- Components that many things depend on. A shared library or service used by 20 applications is a high-value early target if it can be replaced behind the same interface, and a high-risk one if it cannot.
- Edges that cross the system boundary. Each interface to an outside system needs an owner on the other side, an agreed format, and a test. These take calendar time that engineering effort cannot compress, so they should start first.
The same graph supports the disposition element of a modernization plan. A legacy system can be retired only when nothing still depends on it, and the inbound edges are the list of things to check.
An illustrative example
Consider a benefits eligibility system built in the late 1990s, with illustrative numbers throughout. The program’s migration plan lists 6 applications. The package manifests, where they exist, list 85 third-party libraries.
Dependency analysis of the source and the production hosts finds more. There are 23 additional libraries copied directly into the source tree, 9 of them modified locally. The applications run on a language runtime whose vendor support ended four years ago. The job scheduler runs 41 jobs, of which 12 send files to or receive files from other organizations. Interface documentation exists for 5 of those 12.
Sorted by support date, 31 of the 108 libraries are already past end of support, and 17 more will pass it within the planned 30-month migration. One of the expired, locally modified libraries handles encryption of the outbound files.
The plan changes in three ways. Work on the 7 undocumented external interfaces starts immediately, since each needs a counterpart in another organization. The encryption library is replaced first and on its own schedule, because it protects data leaving the boundary. Two applications that the original plan migrated a year apart are moved together, since they write to the same 14 tables.
Keep the inventory current after migration
The new system has its own dependencies with their own support dates, and the continuous discovery approach in BOD 26-02 is the right model: regenerate the inventory on each release, compare it with what is deployed, and review any component within a set window of its support end date. Our post on memory safety roadmaps shows how the same inventory drives per-component treatment decisions, and the post on evidence for the authorization to operate covers how it feeds change review.
How QuantumWorks approaches this
PRISM analyzes a legacy codebase and maps its entry points, behavior, dependencies, and outputs, each anchored to file and line, so that interfaces and copied code are found from the source itself. For cryptographic dependencies, Sentinel helps organizations understand and change cryptography across many systems at once, and CAM provides asset-level cryptographic evidence for assets in production after migration.
If your program is planning a migration and wants a dependency inventory it can sequence work from, we would be glad to talk. Contact QuantumWorks.