release: cut v0.3.4 — buffer ring for scope def (#208) #209

Owner direction, 2026-10-19: run #208 through dev, then tag for prod.

Filed before the tag this time. v0.3.2 (#202) and v0.3.3 (#205) were both backfilled after the fact, which left prod's running version with no readable record twice in a row.

Contents

One commit [verified: git log v0.3.3..origin/master]:

829 insertions / 18 deletions across 9 files. No migrations, no schema change [verified: git diff --name-only v0.3.3..origin/master | grep -iE 'migration|schema|\.sql$' → none], so no database backup is required.

What #208 adds

• The ring. The region-def page draws a dashed ring (default: largest nearMiles of any layer, else 10 miles; settable in the UI or via ?ring=) and lists every region it touches as NEAR rows with distances. Display only.
• Tree-shaped commands. New layer keys root, parentProperty and nearMiles turn the generated region def command from a flat chain into a tree, so a border repeater carries the regions on both sides. With none of those keys set the command is the plain chain it always was.
• scripts/dev-server.js — dependency-free front-end dev server, and devcontainer wiring to start it on port 8080.
• tests/unit/test-region-def-near.js (349 lines) asserting exact command strings for the Ohio / Kentucky / Indiana / cvg / day cases, plus E2E additions.

Version: v0.3.4, and the reasoning is worth stating

Going v0.3.4, a patch bump, to match the immediately preceding precedent: v0.3.3 shipped #187 — the entire region-def page, a far larger feature — as a patch. Being inconsistent release-to-release is worse than being technically wrong about semver.

Strict semver would call a user-visible feature a minor bump (v0.4.0). If the owner prefers that, this is the moment to say so, and it is worth settling the convention once rather than re-deciding every release.

Pre-flight

Dev is the gate here. Verified before tagging:

• the master push run for ec540b91 completes with Deploy Staging success
• dev reports commit: ec540b91 from /api/stats
• dev container healthy, no exited: corescope- lines
• the region-def page's data path still resolves on dev (/region-layers/layers.json returns JSON, not index.html) — this is the chore(prod): region-layers mount reuses ${DATA_DIR}, so the region defs page serves the data dir #204 defect class and dev has its own compose with its own mount

Deploy

Tag v0.3.4 at ec540b91 → release-fast-path.yml re-tags :edge as :v0.3.4 → deploy-prod.

Expected cost: prod cold start measured at ~260s previously, against a 300s healthcheck window with auto-rollback. Roughly four minutes unavailable.

Verification on prod

• /api/stats reports version: v0.3.4, commit: ec540b91
• container healthy, RestartCount unchanged, no in-container crashes
• /region-layers/layers.json still returns JSON (regression check on chore(prod): region-layers mount reuses ${DATA_DIR}, so the region defs page serves the data dir #204)
• node and packet counts still climbing

Note on process

#208's body carried Closes #188, which auto-closes the issue on merge and bypasses the Testing gate (CLAUDE.local.md; DifferentWire/standards#201). Raised in review and not changed before merge; the owner's call was that this work is not going through Testing. Recording it because #188's state will reflect the keyword rather than a QA decision.

Done when

Prod serves v0.3.4 and the region-def ring works against prod's region data.

Refs #208, #188, #205, #202