Add crane CLI to the runtime Dockerfile so the index builder can pull the
latest SPA shell from the Forgejo container registry on each cycle. The
existing syncSiteBuild logic checks for a newer image digest, extracts
the dist/ assets via crane export, and overlays generated JSON data files
on top before deploying to Cloudflare Pages.
- Dockerfile: install go-containerregistry crane binary (v0.20.2)
- sitebuild.go: new file with syncSiteBuild, craneDigest, craneExport,
digest caching, fallback to baked-in /app/web/dist
- main.go: wire initCraneAuth at startup, replace hardcoded webDistDir
with syncSiteBuild call in runBuildCycle
- sitebuild_test.go: 18 tests for extractRegistry, digest caching,
fallback logic, crane auth config, and copyWebAssets overlay behavior
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The bestCandidate function allowed game count to override pairing recency
when both candidates had non-zero LastPairedAt values. Per §6.1, pairing
recency must be the primary criterion with game count only breaking ties.
Also adds 10 tests covering: never-paired preference, oldest-pairing
preference, game count tiebreaking, Pareto distribution verification,
multi-opponent selection, and regression test for the priority inversion.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Per plan §10.8 (deployment pipeline) and §9.8 (Argo Workflows):
- Add waitForWorkflowCompletion() that polls Argo Workflow API
- Add getWorkflowStatus() to fetch workflow phase/status
- Update Promote() to wait for workflow completion before inserting bot record
- Update Promote() to wait for K8s deployment readiness (waitForDeployment)
- Update triggerArgoWorkflow() to return workflow name for polling
- Add acb-evolved-bot-deploy-workflowtemplate.yml to manifests
The promotion flow now:
1. Writes bot source to bots/evolved/<bot_name>/
2. Commits and pushes source to git
3. Triggers Argo WorkflowTemplate
4. Waits for workflow completion (build + manifest commit)
5. Waits for K8s deployment to be ready
6. Inserts bot record into bots table
7. Updates programs table with bot_id/bot_name
This ensures evolved bots have running containers before being marked active.
Adds generateMapsIndex to acb-index-builder, writing:
- maps/index.json — active/probation/classic maps grouped by player count
- maps/{map_id}.json — full map definition with walls, cores, energy_nodes
Queries maps.map_json column for geometry; adds RawJSON field to MapData
and updates fetchMaps to select it.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Implementation complete:
1. engine/thumbnail.go - New thumbnail rendering package
- GenerateMatchThumbnail() creates 640x360 PNG thumbnails
- Renders grid, bots, cores, walls, energy with player colors
- SelectThumbnailTurn() chooses most interesting turn
- Pure Go stdlib image rendering (no canvas required)
2. cmd/acb-worker - Upload thumbnails to B2 alongside replays
- uploadThumbnail() generates PNG and uploads to B2
- Key: thumbnails/{match_id}.png, content-type: image/png
- Called after match completion, non-blocking on failure
3. cmd/acb-index-builder/deploy.go - Promote thumbnails to R2
- promoteRecentReplays() copies both replays AND thumbnails from B2 to R2
- Thumbnails promoted to warm cache alongside replay promotion
4. cmd/acb-index-builder/generator.go - Populate thumbnail URLs
- buildPlaylistMatch() now includes thumbnail_url field
- URL pattern: https://r2.aicodebattle.com/thumbnails/{match_id}.png
- Enables playlist cards and embed OG tags to show preview images
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Add dedicated 10/hour-per-IP rate limiter for POST /api/feedback/{id}/upvote,
separate from the 20/hour feedback submission limiter. Wired in main.go init,
server_test.go helper, and RegisterRoutes.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace random 2-player pairing with the full §6.1 algorithm:
- Seed selection: bot with oldest last-match timestamp (tiebreak: lowest bot ID)
- Format selection: seed's least-played player count among {2, 3, 4, 6}
- Opponent selection: Pareto 80%/16-rank skill proximity + oldest last-pairing
with seed + fewest 24h games for game-count balance
- Map selection: least-recently-used active map for the chosen player count,
with map_scores.last_used_at updated after each match
- Random player slot assignment for all participant counts
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds ratelimit package with per-IP and per-key HTTP middleware.
Applied to register (5/hr), feedback (20/hr), predict (60/hr),
and job submission (5/day) endpoints. Includes metrics counter
for rejected requests and periodic bucket cleanup goroutine.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
PacifistBot never attacks; it survives by maximizing distance from enemies
and retreating toward own core when cornered. Pure evasion strategy that
wins via opponent elimination by third parties.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Add metrics server startup and HTTP middleware to acb-api, generation
counter metric to evolver, and R2 cache size metric to index builder.
Also remove dead measureR2CacheSize reference from index builder main.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The RecencyBoost test now uses balanced 5-5 splits for both pairs so
recency is the sole differentiating factor (previously one pair was 10-0
which conflated balance and recency). Also wires Prometheus build
duration metric in main loop.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Add crosspoll_state table to persist per-island generation counters
across evolver restarts. Load state on startup and save after each
cross-pollination check. Add persistence pattern and translation
structure tests.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- Add Exploration and Formation axis definitions with feature extraction
from source code pattern matching (exploration/formation indicators)
- Extend Grid key from (x,y) to (x,y,z,w) with 3⁴=81-cell behavior grid
- Update bin assignment, promotion gate, and persistence (JSON snapshot)
- Add Slice() for 2-D dashboard visualization across any axis pair
- Migration: old 2-D archives project at z=middle, w=middle
- Update cross-pollination to pad 2-element behavior vectors to 4
- Add Prometheus metrics to matchmaker (bot crashes, stale job count)
- Add rivalry detection to index builder (data/meta/rivalries.json)
- Web: batched bot list loading, leaderboard keyboard accessibility,
improved ARIA attributes on match/playlist cards
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Adds mock store/LLM implementations and tests for CheckAndPollinate:
generation boundaries, fitness penalties, translation triggers,
multi-boundary catch-up, and empty island handling.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Add crash_strikes and cooldown_until columns to bots table. Worker
increments strikes on crash (cooldown at 3), resets on success.
Matchmaker excludes cooldown bots from pairing, series scheduling,
and championship brackets. Fix erroneous cooldown filter on series
table in finalizeCompletedSeries (column only exists on bots).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The crash cooldown system was already implemented across engine, worker,
and matchmaker. This adds comprehensive integration tests that verify:
- Single crash does not trigger cooldown
- Two crashes do not trigger cooldown
- Three consecutive crashes trigger 30-min cooldown
- Successful match resets strike counter
- Interleaved crash/success resets counter correctly
- Cooldown extends on repeated crashes while on cooldown
- Matchmaker eligibility query excludes bots on active cooldown
- Matchmaker eligibility query includes bots with expired cooldown
- Full end-to-end flow: 3 crashes → excluded → cooldown expires → re-pair
Tests use ACB_TEST_DATABASE_URL env var for PostgreSQL integration tests
and skip gracefully when not configured.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Adds cross-pollination logic that copies the top program from each island
to a random other island every 50 generations. When source and target
islands use different languages, the LLM translates the code. Generation
boundaries are tracked per-island to prevent duplicate events.
- New crosspoll package with boundary detection, migration, and LLM translation
- Added MaxGenerationByIsland DB query for generation counter tracking
- Integrated into RunEvolutionLoop with observability logging
- Tests for boundary logic, translation prompts, and target selection
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Worker now gzip-compresses replays before uploading to B2 with
key replays/{match_id}.json.gz and Content-Encoding: gzip.
Updated B2 client Upload to accept contentEncoding parameter.
Fixed downstream web consumers (matches, bot-profile, playlists)
to reference .json.gz URLs.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The getEnv() function in server.go always returned the default value,
preventing ACB_R2_ENDPOINT/ACB_B2_ENDPOINT from being read at runtime.
Also updated Dockerfile from golang:1.24 to golang:1.25 to match go.mod.
K8s manifests for acb-evolver and acb-api already exist in
declarative-config/k8s/iad-acb/ai-code-battle/ (added Apr 21).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The community feedback endpoint was registered as /api/ui-feedback in
the Go API but the plan and annotation.ts client both use /api/feedback.
Rename the route and update agentation-overlay.ts to match. Add a
route-level test asserting the canonical path and that the old path
returns 404.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Playlist curation per §10 is fully implemented in the index builder:
- generatePlaylists() writes /data/playlists/index.json and {slug}.json
- curateWeeklyHighlights() selects best-of-week by upsets, elite
clashes, marathon turns, and closest finishes (last 7 days)
- persistGeneratedPlaylists() upserts to playlists/playlist_matches DB tables
- /data/playlists/ stub files seeded for all 12 curated collections
Replay viewer improvements shipped alongside:
- WinProbPoint refactored from {p0,p1} to {probs: number[]} for N players
- renderWinProbSparkline draws one line per player with matching colors
- replay.ts updated to build probs[] from replay.win_prob arrays
- Dynamic legend generated from replay.players instead of hardcoded P0/P1
New annotation overlay component (§16.8):
- AnnotationOverlay: timeline track, per-turn list, canvas markers
- createAnnotationForm: type selector, author, body, localStorage + API
- ANNOTATION_OVERLAY_STYLES: self-contained CSS for the overlay
Evolver: add mutations_per_hour metric to Totals (live.json §14)
Types: consolidate evolution types into types.ts, re-export from api-types.ts
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add debug telemetry UI to replay viewer with player toggles,
priority-based target markers, and stacked reasoning boxes.
Fix undefined generateTestImage in main_test.go.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Covers buildFirstMatchPerBot, isNewBotDebutFast, buildPairFrequency,
isRivalryMatchFast, and integration test for playlist generation with
the optimized lookups.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- Fix deploy.go to query actual table names (series_games not series_matches,
join through series_games for seasons instead of non-existent season_matches)
- Add playlist_matches table to exempt match IDs from R2 pruning
- Pre-build lookup maps for O(1) playlist match filtering instead of O(n²)
- Enhance home page featured replay to prefer AI-enriched matches
- Add enrichment test coverage (shouldEnrich criteria validation)
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Implement auto-curated playlists in the index builder: 12 playlist types
(closest finishes, upsets, comebacks, marathons, rivalry classics, etc.)
with weekly highlight curation. Add DB persistence, R2 pruning exemptions,
frontend pages, and AI commentary enrichment pipeline.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The buildSpotlightPrompt function accepted a rivalries parameter but never
used it. This adds top rivalry data to the LLM prompt so the generated
Counter-Strategy Spotlight can reference active rivalries. Test updated to
verify rivalry data appears in prompt output.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The series scheduler was scheduling games and creating series_games rows,
but never updated winner_id or incremented a_wins/b_wins when individual
matches completed. This left series in perpetual "active" state since
finalizeCompletedSeries checks win counts that were never incremented.
Add updateSeriesGameResults step that:
- Finds series_games with completed matches but NULL winner_id
- Updates winner_id from match_participants
- Increments a_wins or b_wins on the series table
Called as step 0 in tickSeriesScheduler, before finalization checks.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Enhance the WASM game sandbox with production-accurate Go engine:
- Add multi-player support (2-4 players) to Go WASM engine via JS callbacks
- New acbEngine.addPlayer/clearPlayers/runMatchMulti API for N-player matches
- Sandbox auto-loads Go WASM engine in background, falls back to TS engine
- Engine selector: Auto (Go WASM → TS fallback), Go WASM only, or TS only
- Engine status indicator shows which engine is active
- Performance panel reports which engine was used
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- Worker resolves open predictions after writing match results (resolvePredictions + upsertPredictorStats)
- API endpoints: POST /api/predict, GET /api/predictions/open, GET /api/predictions/history
- Frontend /watch/predictions page with polling, prediction submission, and history display
- predictor_stats table tracks streaks and accuracy per predictor
- Series format selection: fix threshold from >200 to >=200 for bo3 eligibility
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- Evolution page: live polling (10s), activity feed, candidate tracking,
statistics section, island overview with live.json schema
- Series page: detailed series view with game-by-game results
- Seasons page: season list with status and champion display
- Predictions page: enhanced prediction UI with open matches
- API types: add CycleInfo, Candidate, ActivityEntry, Totals for live.json
- Embed: improved embeddable replay widget
- Mobile CSS: responsive breakpoints and bottom tab bar
- Exporter: enhanced live.json generation with full cycle/candidate data
- Matchmaker: series scheduling support with config
- Worker: additional database queries for series/season data
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- POST /api/register: bot registration with URL + shared secret validation
- GET /api/job: worker polls for next pending match job (authenticated)
- POST /api/job/:id/result: worker submits match result (winner, replay JSON)
- GET /api/replay/🆔 serve replay JSON from R2 warm cache (falls back to B2)
- GET /api/bot/🆔 bot profile JSON (rating, elo, record, metadata)
- GET /api/bots: leaderboard snapshot with pagination
- POST /api/ui-feedback: accept Agentation UI feedback
Authentication via Bearer token (worker API key). Shared secrets encrypted
with AES-256-GCM using ACB_ENCRYPTION_KEY.
Installs Python 3, Node.js/TypeScript for bot validation sandbox.
Base image includes Go; Java/Rust/PHP validation is deferred to follow-up bead.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace os.ReadFile+os.WriteFile with io.Copy so large files (e.g. the 21MB
demo-replay-v2-6p.json) are never fully loaded into RAM during copyWebAssets.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add a web-builder stage to the Dockerfile (Vite/TS build) and copy the
resulting dist/ into the runtime image at /app/web/dist. Call copyWebAssets
each build cycle so HTML/JS/CSS is merged into the output dir before wrangler
deploys — previously only JSON data files were uploaded, causing CF Pages to
serve 404 at the root.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
wrangler creates .wrangler/tmp relative to its working directory.
The container runs as non-root user acb with WORKDIR=/app (root-owned),
so mkdir /app/.wrangler/tmp fails with EACCES. Setting cmd.Dir=/tmp
gives wrangler a writable CWD while keeping the /data output path
(absolute) unchanged.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
matches.winner is an INTEGER (player slot), not a bot_id VARCHAR.
Fix two queries that compared mp.bot_id = m.winner (type mismatch)
to use mp.player_slot = m.winner.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
COALESCE(parent_ids, '[]'::json) fails because the column is JSONB
and PostgreSQL won't coerce json to jsonb. Change to '[]'::jsonb.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Fixed TestSelectBestCandidate_GoHttpBonus: HTTP bonus (1.5x) on 150-char code
(225 score) doesn't beat 500-char plain text (500 score). Test now expects
the longer code to win.
- Fixed TestScoreCandidate_Bonuses: adjusted minScore expectations to match
actual code lengths with 1.5x bonus applied.
- Fixed TestBehaviorDistance: use epsilon comparison for floating-point
precision instead of exact equality. sqrt(2) ≈ 1.414214 is not exactly
representable in floating-point.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add the 'evolve' subcommand that ties together the LLM prompt builder
and ensemble components:
- Load programs from target island
- Select parents via tournament selection
- Analyze optional replay files for strategic context
- Build meta description from current ladder state
- Assemble evolution prompt with all context
- Run LLM ensemble (fast tier + strong tier refinement)
- Output generated bot code
Usage: acb-evolver evolve -island alpha -lang go [-replay file.json] [-out file.go]
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add parent sampling via tournament selection (selector/tournament.go)
- Add replay analyzer to extract key moments, strategies, weaknesses
- Add meta builder for leaderboard summary and dominant strategies
- Add prompt assembler combining parent code + replay + meta context
- Add LLM ensemble with fast tier (GLM-5-Turbo) for bulk generation
and strong tier (GLM-5) for refinement passes
- Add code extraction from LLM responses with language validation
- Add convert utilities for type conversion between packages
- Comprehensive test coverage for all components
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add Go implementations of 5 strategy bots directly into the engine:
- GathererBot: prioritizes energy collection, avoids combat
- RusherBot: aggressively rushes enemy cores
- GuardianBot: defends cores with cautious expansion
- SwarmBot: formation-based coordinated movement
- HunterBot: targets isolated enemy units
Update acb-local with bot selection flags:
- -bot0/-bot1: select bot strategies
- -list-bots: list available strategies
- Default to gatherer vs rusher for interesting gameplay
Enables demo replays with real strategic behavior without K8s infrastructure.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add Delete, List, ListTopByIsland, and GetLineage methods to the programs
Store. These complete the CRUD operations needed for the evolution pipeline:
- Delete: Remove programs by ID
- List: Paginated listing of all programs
- ListTopByIsland: Get top N programs by fitness for a specific island
- GetLineage: Recursively traverse parent chain for lineage tracking
Also adds comprehensive tests for all new operations including lineage
tracking through grandparent-parent-child chains.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Upload replays to B2 (Backblaze) instead of R2 for cold archive storage
- Write match results directly to PostgreSQL instead of HTTP API
- Perform Glicko-2 rating updates in worker after match completion
- Update config: ACB_R2_* env vars → ACB_B2_*
- Remove obsolete api_test.go (tested removed HTTP client)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
