Find peers

A peer is a public repo that shares our stack lines or a concrete stack seam. Default discovery asks for two lines; a relevant XState runtime or integration can qualify without Alchemy. Peers hit breaking changes first and write idioms before the docs do. Effect 4 remains a shared line after its stable release; drift between beta, RC, and stable is recorded, not hidden.

Run

pnpm find-peers

The script reads our prerelease pins (and the Effect 4 major line when stable) from the workspace package.json files, searches Sourcegraph once per line, and prints a table of repos ranked by how many lines they share. Scoped packages that move with their parent, such as @effect/vitest with effect, count once. A repo missing from .brain/data/peers.json is marked new. It takes about twenty seconds. Pass --min 1 to see repos that share only one line.

Refresh peers, versions, and tiers

Refresh the structured rows in .brain/data/peers.json every week or two and after every bump of a shared line (Effect, Alchemy, XState, or @xstate/effect). The gardener pass owns this step; no external scheduler is required. Run discovery, add new peers, and revisit existing peers even when discovery finds nothing new.

Rat-stack's own pins are read from its root, infra, and core manifests at content-build time, so a bump changes the comparison without a hand-edited baseline. For every peer, resolve its default branch to a commit and read its actual manifests or lockfiles at that commit. Record all four lines, the checked date, and source links. Preserve exact pins; label ranges, workspace/catalog references, and missing dependencies honestly. Resolve references from the workspace catalog or lockfile where possible; never turn a range into an invented pin. Show differences beside rat-stack's pins; version drift is expected, not a reason by itself to demote a peer.

Keep every peer in .brain/data/peers.json, including its first-seen date, tier, one-line evidence-based reason, versions, checked date, default branch and commit, manifest/lockfile source paths, and study links. The content build reads that one dataset, sorts by tier then repository name, and renders .brain/resources/peers.svx: S, A, and B in the main table; C through F in a collapsed Also seen list. Edit structured rows, never hand-edit rendered table rows. Never delete a peer that fell behind.