Why

This is Phase 1 of turning reposystem into the estate cockpit — the unifying representation → visualisation layer. It fixes the root problem found during exploration:

• The representation was fragmented across five disagreeing sources (Rust types, spec/DATA-MODEL.adoc, config/*.ncl, legacy graph.toml/GRAPH-STRUCTURE.md, and repos.toml).
• The real ~297-repo estate (repos.toml) was completely disconnected from the visualisation engine (GraphStore), so web/export.json was an empty stub and the UI rendered nothing real.

Per the agreed plan: strict repos-only boundary (aerie/ambientops are loose external seams, not absorbed), curated-hybrid cockpit, estate-parameterised now / multi-tenant later, and first step produces the representation and begins converging the front-ends.

What changed

Representation — Rust types is now the single schema of record

• Estate (multi-estate parameterisation; hyperpolymath = estate chore(deps): bump the cargo group across 1 directory with 2 updates #1), ExternalSeam + SeamDomain (loose cross-links to aerie=network / ambientops=machine; seams are edge sinks), RelationType::RefersTo (the only relation allowed to target a seam), and estate/metadata fields on Repo. All new fields #[serde(default)] → old data still loads.
• Seam-sink invariant enforced in EcosystemGraph::add_edge; seams rendered in to_dot; persisted for free via the existing split-file JSON.

The missing bridge + unified interchange

• reposystem import manifest (src/importers/manifest.rs) maps repos.toml → GraphStore (289 unique repos + aerie/ambientops seams), reusing Forge::from_url and scanner::parse_owner_name. Sibling systems matched by basename (catches systems-ecosystem/ambientops).
• export --format estate-json → the envelope (schema: reposystem/estate-export@1) serialising all stores + estates + seams. One serializer, many transports.

Visualisation (begin convergence)

• web/app.js consumes the envelope: renders external seams (distinct dashed style) and adds an estate selector + estate filtering. index.html/styles.css add the control + seam node style.

Schema alignment + the "super power collection" registry

• config/{contracts,reposystem}.ncl aligned to the Rust taxonomy (dropped the competing EdgeKind; added Estate/Seam/RefersTo).
• config/tools.ncl — new typed registry of the tool collection (embedded core vs extracted standalone, role in the stack).
• spec/DATA-MODEL.adoc made normative; GRAPH-STRUCTURE.md marked superseded.

Pre-existing build/test fixes (surfaced once the build was unblocked)

• Cargo.toml: enable gix sha1 feature — gix-hash's Kind::Sha1 (the #[default]) was cfg'd out, failing to compile on rustc 1.94 (this blocked all builds, independent of these changes).
• scan now honours --json; logs routed to stderr so piped stdout stays clean; fixed an unannotated verisimdb doctest fence.

Verification

• cargo test → 113/113, 0 failures (fresh run; supersedes the stale 2026-04-04 baseline).
• just web-refresh → non-empty web/export.json (289 repos, aerie/ambientops seams, estate scoping); just web-serve renders the real estate with external seam nodes.
• reposystem export --format dot renders seams (shape=note, dashed).

Notes / follow-ups

• web/export.json is regenerated with real estate data (it was an empty stub) — reproducible via just web-refresh.
• Deferred to follow-up PRs (per plan): consolidation gitlink surgery (orphan avatar-fabrication-facility/claim-forge, rpa-elysium path conflict) and Phases 2–6 (GUI/TUI convergence, tools-as-interface, cross-forge control surface, interop).
• nickel isn't installed in CI here, so the .ncl files weren't machine-validated (they're not consumed by the build); syntax follows standard Nickel.

🤖 Generated with Claude Code

https://claude.ai/code/session_01KhFfkfjKaB7Rg957u3uCT6

Generated by Claude Code