Files

T

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

2026-04-05 10:23:56 +02:00

Sprint 31: Bedrock — Retrospective

Period: 2026-04-05 (single day) Velocity: 9/9 tickets completed (100%) PRs: 4 merged (#110 copy, #111 server, #112 ci, #113 client)

What Shipped

#766 Cultural diversity sweep — 10 systems enriched across 3 corridors, Afrikaans/Dutch subversions, per-body GTTR depth
#782 Per-body GTTR entries for 25 moons and stations at hub systems
#774 Nova Estrada and Entremeio enriched to batch standard
#773 D-168 Iserlohn IP evaluation — retained with documented rationale and Altena alternative on record
#780 Star map click-through popup with GTTR excerpt, edge-only-on-selected visibility rule
#775 main.gd decomposed from 28KB/644 lines to 13.4KB/322 lines (SnapshotConsumers, DialogueCoordinator, SnapshotHandler extracted)
#783 All 354 gdlint warnings resolved — zero warnings policy achieved
#776 atlas.rs split from 2119-line monolith into 8 focused modules
#777 Shared Python DB module (common.py), ruff linting, pre-push integration

First sprint on ephemeral branches — start-sprint.sh created worktrees, opened tabs, auto-ran /sprint-start. No cross-boundary issues, no stuck agents with uncommitted work, no stale index.lock collisions. The worktree migration paid off immediately.
Zero warnings achieved — gdlint went from 354 warnings to 0, ruff is clean, clippy is clean. Pre-push now catches all three. Quality baseline established.
4 teams in parallel, same day — copy, server, ci, client all delivered PRs within hours. The ephemeral branch model handled parallel execution cleanly.
Review cycles were fast — server needed 2 rounds (changelog fix), ci and copy merged on first review, client merged on first review. No busywork rounds.
Process improvements shipped inline — sprint-start process fix (team branches must not run pr-review) came in with the copy PR and merged naturally. Auto-unassign on sprint close fixed a recurring bookkeeping issue.

SI agent died during sprint planning — had to redo planning manually. The agent's death left a stale index.lock on main, which was the exact problem the worktree migration was designed to prevent. Root cause: SI was spawned in the main worktree, not an isolated one. Planning agents should use isolation: "worktree" in future.
Ptyxis tab placement — first start-sprint.sh run opened 3 windows instead of tabs in the current window. Fixed quickly (switched --new-window to --tab), but the initial experience was messy.
Cargo.lock conflicts on merge — server branch had a stale Cargo.lock from pre-v0.1.30. Required manual git checkout before merge. This will recur until sprint branches are created after the version bump.

The cascade is load-bearing — redirecting Sprint 31 from tycoon/character work back to Phase 1 wiki completion was the right call. The star map popup only works because 301 systems have GTTR content. Skipping ahead would have produced empty popups.
God object decomposition is safe in GDScript — the autoload parse-order constraint was the main risk. The load() inline pattern works, and non-autoload classes can use class_name freely. Documented as convention in CLAUDE.md.
Copy PR review scope rule continues to work — Paula's Fragua warning and Miri's GJ-1111 sync issue were correctly classified as content tickets, not PR blockers. Zero scope creep on copy reviews.

(none — 100% completion)

Done: Auto-unassign on sprint close added to tooling/db/sprint stop (committed)
Done: Zero-warnings policy added to pr-push and pr-review skills
Done: start-sprint.sh fixed to open tabs in current window
Backlog: Panel height bug in star map popup for hub systems (cosmetic — Hoshe flagged)
Backlog: show.rs SQL duplication maintenance trap (Tyre flagged — ticket-worthy)
Backlog: qdrant/audio connectors still have local load_config() — incomplete shared module adoption
Future: Planning agents should spawn with isolation: "worktree" to avoid main index.lock