Consumer story
discovery-lab computes a provenance graph (tools/provenance.mjs) over 236 on-disk supervise run directories using the kernel's own readers — FileSpawnJournal + materializeTreeView for spawn trees and spend, FileCoordinationLog for steering and analyst findings. Both adoptions worked, but each needed a read-the-first-line adapter because the reader demands an identity the file's own records already contain.
What was read (0.132.13 published package, verified live against lab run dirs)
FileSpawnJournal.loadTree(root) requires the root NodeId, and silently returns undefined when the requested root does not match the journal's begin record. In 1 of 204 lab journals the root id differs from the directory name (a GEPA-search id), so loadTree(dirname) reads as "no journal" — indistinguishable from an absent file. The begin record ({"kind":"begin","root":...}) already names the root.FileCoordinationLog.load(runId, ownerId) requires the ownerId (a profile digest), but every row of the log carries runId and ownerId. A reader that only has the file path must parse the first line manually before it may call the reader.
Required change
FileSpawnJournal: a way to read a journal file without pre-knowing the root — either loadTree() defaulting to the begin record's root, or an exported readJournalRoots(path). A root mismatch should be distinguishable from a missing tree.FileCoordinationLog: load() with both arguments optional, defaulting to the identities in the file, or an exported discovery helper.
What this deletes downstream
Two first-line-parse adapters in discovery-lab's tools/provenance.mjs (loadJournal, loadCoordination). Recorded in discovery docs/15-adoption-decisions.md.
Consumer story
discovery-lab computes a provenance graph (tools/provenance.mjs) over 236 on-disk
superviserun directories using the kernel's own readers —FileSpawnJournal+materializeTreeViewfor spawn trees and spend,FileCoordinationLogfor steering and analyst findings. Both adoptions worked, but each needed a read-the-first-line adapter because the reader demands an identity the file's own records already contain.What was read (0.132.13 published package, verified live against lab run dirs)
FileSpawnJournal.loadTree(root)requires the root NodeId, and silently returnsundefinedwhen the requested root does not match the journal'sbeginrecord. In 1 of 204 lab journals the root id differs from the directory name (a GEPA-search id), soloadTree(dirname)reads as "no journal" — indistinguishable from an absent file. Thebeginrecord ({"kind":"begin","root":...}) already names the root.FileCoordinationLog.load(runId, ownerId)requires the ownerId (a profile digest), but every row of the log carriesrunIdandownerId. A reader that only has the file path must parse the first line manually before it may call the reader.Required change
FileSpawnJournal: a way to read a journal file without pre-knowing the root — eitherloadTree()defaulting to the begin record's root, or an exportedreadJournalRoots(path). A root mismatch should be distinguishable from a missing tree.FileCoordinationLog:load()with both arguments optional, defaulting to the identities in the file, or an exported discovery helper.What this deletes downstream
Two first-line-parse adapters in discovery-lab's
tools/provenance.mjs(loadJournal,loadCoordination). Recorded in discoverydocs/15-adoption-decisions.md.