# 2026-06-13 — floor live + trail root-listing fix ## Floor in production R1+R2+floor rolled to the cluster (server 0.17.18, broker 0.1.36, library 0.1.5). The worlds bump needed a one-time orphan-delete recreate (root + world-a) — caused by the chart stamping churning version labels on the immutable volumeClaimTemplates; fixed at the chart layer (#184: VCT uses selectorLabels), so future server bumps are clean. Full rollout detail in journal 2026-06-12. ## Trail root-listing bug (library PR #16 → v0.1.6) Fritz hit `{"message":"malformed trail"}` on `/t/u/~/world-a/d/` — the floor's world-node link (opens a world's stacks = root listing, path "/"). Root cause: the trail parser rejected any pane chunk with an empty value (`if !ok || value == ""`), but the world-root listing legitimately encodes as `/d/` (doc-kind, empty value → "/"). Predated the floor — every root-listing link was broken since R2; the round-trip test only covered a non-empty subdir (/plans/) and missed root. Three CodeRabbit findings, all real, fixed in sequence: 1. Root-listing rejected — drop blanket `value==""`, handle per kind (doc empty → "/", tag still non-empty). 2. **Two parsers diverged** — `parsePaneChunk` (/t/ chunks) and `paneAddrFromRoute` (/w/ route hrefs in the trailize pass) both had the bug; fixed one, not the other. Extracted the shared semantics into `paneAddrFromParts(world, kind, value)` — single source of truth; the two wrappers keep only their input-specific framing (chunk-split+floor vs URL-parse+fragment). Divergence now structurally impossible. 3. Tag slash check ran on the RAW value, so `tags/foo%2Fbar` slipped through then unescaped to `foo/bar`. Moved empty/slash checks AFTER PathUnescape (in the shared helper → fixed for both decoders at once). Pattern worth keeping: two parsers for the same logical address = a divergence bug waiting to happen. Share the semantics, not the syntax framing. Deploy bump library 0.1.5 → 0.1.6 staged (deploy `fix/library-0.1.6`; clean Deployment roll, no recreate).