Map Legacy Banking Systems
The documentation a regulator asks for can be reconstructed by agents that sit on the machines, and kept current by agents that stay there.
Every bank runs systems nobody fully understands any more: a core that has outlived three vendors, batch chains that run at 02:00 because they always have, integration code whose author retired, a schema with four hundred tables and a spreadsheet that explains sixty of them. The documentation exists in fragments, in people, and in the systems themselves. Only the last of those scales.
Lightspeed maps an estate by putting agents on the machines and leaving them there. Not a crawler that reads repositories from the outside: agents with a shell on the hosts that run and reach the core, reading what is actually deployed.
On the machine, not beside it
A session in Lightspeed borrows real machines. For an estate that already exists, the machine is the host you already have: an existing VM is taken over explicitly, or an external machine is attached by describing how to reach its data-plane daemon. From there the agent has the file and process tools of a workstation: it reads the deployed configuration rather than the one in the wiki, lists the cron table and the batch scheduler’s definitions, inspects the processes that are actually listening, follows the JDBC URL in the properties file to the schema it really points at, and reads the COBOL, PL/I or Java in the repository the host checks out.
Credentials for those hosts and databases are injected into the environment and never shown to the model. The agent can run the query; it cannot read the password. Access is scoped by the universe the estate belongs to, so a mapping exercise for one business unit cannot reach the systems of another.
Split the estate, merge the map
An estate is mapped by a fleet, not an agent. A coordinator session splits the landscape, one sub-agent per host group, repository family or schema, and each sub-agent works from its own profile with its own environment. The coordinator merges what they find: which job writes the file the next job reads, which service calls which endpoint, which table is written by three programs and read by thirty. Limits on depth, descendants and concurrency keep the fleet the size the change board approved.
The map itself lives in a virtual file system: documentation, diagrams as text, notes and the skills the agents learned about this estate. It has no operating system attached and stays available whether or not a machine is currently borrowed. Fork a session to explore two hypotheses about a batch dependency from the same state without repeating the reading that got there.
Stays current because the agents stay
The reason documentation rots is that the writer leaves. A Lightspeed session lasts months and costs nothing while idle, so the agent that mapped a system can keep watching it: a resident session on a schedule re-reads the cron table and the deployed configuration, diffs them against the map, updates the map, and posts what changed to the team’s channel. When a machine is not needed it is paused by idle policy and wakes on the next run. The map is current because it is regenerated, not maintained.
Evidence, not just output
Everything a mapping agent read is in its event log: which host, which file, which query, which result, under which principal. For a regulator asking how the bank knows its own systems, the map comes with its own provenance. For the migration that follows, the map is the input, and the migration is the subject of Accelerate and Automate Migrations.
What a mapping run looks like
An estate of forty hosts, three hundred repositories and a dozen databases. A coordinator with a limit of twelve concurrent sub-agents and a two-week deadline. Each sub-agent borrows a machine from a versioned image or attaches to an existing host, reads for a few hours, writes its section of the map, and releases the machine. The coordinator merges, cross-checks the interfaces both sides claim, and flags the ones that do not match. Idle time between batches costs the price of a database row. At the end of two weeks the bank has a map of its estate that a human can review, and the agents that made it are still there to keep it true.
Rests on: dedicated environments, environment file and process tools, virtual file system, sub-agents, session fork and clone, credential injection, idle power policy, long-running sessions. Mechanism in depth: Map & Document Your Enterprise.