yoki-monitor

Operational health dashboard for Yoki Arcade on Soneium. Live at astarnetwork.github.io/yoki-monitor.

Phase 1 (live)

A single-page SPA showing:

• MINTER hot-wallet ETH balance (0x3dAd…4DDD) with warn/critical thresholds (< 0.01 / < 0.005 ETH).
• Yoki Treasury ASTR balance (0xfA2B…907f) + last-24h inflow line.
• Matchmaking bot balances (ETH + ASTR + PnL since launch for funding + bots 1–4).
• Engagement row — Cores holders, JKP 24h active players, JKP 7d active players.

Balances are read live via viem on page load and refreshed every 60 seconds while the tab is visible. JKP active-player counts and Cores holders are pulled from Soneium Blockscout REST and cached 5 min.

Phase 2

• Treasury growth chart (2.6): LIVE. Public Recharts line chart of the treasury balance JSONL, with KPI tiles for total inflow, mint count, average mint price realized, and last-24h/7d inflow. Always-on; renders an empty state until two snapshot rows are available.

Retired 2026-07-28: the §54 JKP detector

Yoki Arcade entered maintenance mode; the JKP detector stack was retired to stop the cron noise. The code and the accumulated data are kept, not deleted; only the workflow that drove them is gone. Final state at shutdown: 12,077 matches, 3,358 addresses, 4,183 pairs, zero addresses flagged and zero suspicious pairs across the full run.

• JKP event walker cron (2.2): RETIRED. Walked MatchCreated/Joined/Revealed/Resolved/Draw/Cancelled/Swept events from YokiJKP, maintained data/jkp-matches.json keyed by matchId, recomputed data/jkp-aggregate.json (per-address W/L/D + volume) and data/jkp-pairs.json (canonical pair counts). These files are now frozen.
• Always-winning trigger cron (2.4): RETIRED. Flagged any address with ≥75% win-rate over ≥5 matches, posted Slack CRITICAL to #04-yoki-arcade-alerts, persisted state in data/flagged.json. Never fired.
• Suspicious-pairs trigger cron (2.5): RETIRED. Flagged pairs that had played ≥10 matches against each other AND where those matches were ≥50% of either participant's total. Persisted state in data/suspicious-pairs.json. Never fired.
• Public leaderboard tile (2.3): HIDDEN. VITE_LEADERBOARD_ENABLED was flipped to false at retirement: with the walker gone, jkp-aggregate.json no longer updates, and a public leaderboard that silently freezes is worse than no leaderboard.

To revive any of this, restore .github/workflows/jkp-events.yml from git history (last present at de6f4540) and flip VITE_LEADERBOARD_ENABLED back to true. The scripts under scripts/ are untouched and still work.

Local development

npm install
npm run dev

By default, the Vite config serves under base path /yoki-monitor/ (matching the gh-pages deploy URL). To preview at / locally:

VITE_BASE_PATH=/ npm run dev

Configuration

SPA env vars (build-time, in Vercel/Amplify-style)

All optional. Falls back gracefully:

Var	Default	Purpose
VITE_SONEIUM_RPC_URL	Soneium public RPC	Alchemy/dRPC paid endpoint if public RPC is throttling.
VITE_BASE_PATH	/yoki-monitor/	Override for non-gh-pages hosting paths.
VITE_TREASURY_BALANCE_JSONL_URL	unset	URL to the public treasury-balance.jsonl from Phase 1.2 cron. Page hides 24h-delta when absent.
VITE_TREASURY_INFLOWS_JSONL_URL	unset	URL to the public treasury-inflows.jsonl from Phase 1.3 cron. Drives the Phase 2.6 KPI tiles.
VITE_JKP_AGGREGATE_JSON_URL	unset	URL to the frozen jkp-aggregate.json. Only relevant if the 2.2 cron is revived.
VITE_LEADERBOARD_ENABLED	false	Set to "true" to surface the leaderboard tile. Held at false since the 2.2 cron was retired 2026-07-28; the underlying data no longer updates.
VITE_REVIEW_KEY	unset	Secret token enabling the operator review view at /?review=<token>. See "Review mode" below.
VITE_FLAGGED_JSON_URL	unset	Override for the flagged.json raw URL (review view only).
VITE_SUSPICIOUS_PAIRS_JSON_URL	unset	Override for the suspicious-pairs.json raw URL (review view only).

Cron secrets + variables (set in repo Settings → Secrets and variables → Actions)

Name	Type	Used by	Purpose
SONEIUM_RPC_URL	secret	all crons	Alchemy/dRPC URL. Falls back to viem's built-in Soneium public RPC if unset.
SLACK_WEBHOOK_URL_ALERTS	secret	1.2 / 1.3	Slack incoming-webhook URL for the operator alerts channel.
TREASURY_OUTFLOW_DRY_RUN	variable	1.3	Set to false on 2026-07-28, outflow alerts are now LIVE. Was true from 2026-05-11, during which five real outflows (totalling 184,500 ASTR) were recorded to treasury-outflows.jsonl without ever alerting.
TREASURY_HEARTBEAT	variable	1.2	true posts a Slack balance heartbeat at most once a week (see "Weekly heartbeat" below). Clear or set false to mute.

Deployment

Production deploys to https://astarnetwork.github.io/yoki-monitor/ via .github/workflows/deploy-pages.yml using the official actions/deploy-pages action (no gh-pages branch).

Pushing to main triggers a build + deploy automatically, EXCEPT when the only changes are under data/** or **/*.md — those are cron-driven and don't affect the SPA. Manual rebuilds: Actions → "1.4 Deploy SPA to GitHub Pages" → Run workflow.

Optional build-time variable:

Name	Type	Purpose
VITE_SONEIUM_RPC_URL	repo variable	Paid Alchemy/dRPC URL baked into the SPA. Falls back to Soneium public RPC if unset.

Cron workers

GitHub Actions workflows under .github/workflows/:

Three scheduled workflows remain, all daily and deliberately staggered:

Workflow	Cadence (UTC)	What it does
treasury-balance.yml (1.2)	daily 09:05	Read treasury ASTR balance, append to data/treasury-balance.jsonl. Silent unless TREASURY_HEARTBEAT=true, and then at most weekly.
treasury-flow.yml (1.3)	daily 09:25	Walk ASTR Transfer events to/from treasury since last checkpoint, append to data/treasury-{inflows,outflows}.jsonl, Slack CRITICAL on any outflow.
bot-balances.yml (1.7)	daily 09:45	Read ETH + ASTR balance for the 5 matchmaking bot wallets (funding + bots 1–4) in parallel, append to data/bot-balances.jsonl. Silent; the bot Lambda owns its own auto-disable alerts.

All workflows commit the JSONL/JSON deltas back to main from a bot identity.

Why the 20-minute stagger matters. Each workflow's concurrency group only serialises it against itself, never against its siblings, and the commit step is a bare git commit && git push with no rebase or retry. When two crons ran at the same minute the second push died with ! [rejected] main -> main (fetch first) and failed the whole run. At the old cadences (5 min / 15 min / hourly) they collided constantly at :00, :15 and :30. Daily runs 20 minutes apart cannot overlap, each job finishes in well under a minute, so the race is eliminated structurally rather than patched. If you ever re-tighten a cadence, add git pull --rebase to the commit step first.

Schedules are offset from the top of the hour on purpose: GitHub drops scheduled runs under load, and :00 is the worst slot.

Retired: minter-balance.yml (1.1, disabled 2026-06-03 when MINTER_ROLE was revoked and the hot wallet went inert), daily-champions.yml (2.5a, campaign ended 2026-06-02 and its backing endpoint was deleted 2026-06-18), and jkp-events.yml (2.2/2.4/2.5, see Phase 2 above).

Weekly heartbeat

With TREASURY_HEARTBEAT=true, 1.2 posts a balance line to Slack at most once every 6.5 days, tracked via data/.heartbeat-treasury.json. The gate is on elapsed time, not on a weekday+hour match: with a single daily tick, a fixed-slot gate would either miss a dropped run and go silent for two weeks, or drift a day later every week. The state file is written only after Slack accepts the post, so a failed post retries on the next daily tick.

Data note: timestamps in the flow JSONLs

Rows in treasury-inflows.jsonl / treasury-outflows.jsonl written before 2026-07-28 carry the time the cron happened to observe the transfer, not the time it occurred. The 2026-06-09 outflow, for example, landed on-chain at 14:30:29Z but is recorded as 16:47:20Z. From 2026-07-28 the walker stamps the block timestamp, which the daily cadence made essential: walk time would otherwise be up to 24h off. blockNumber was always accurate and is the field to trust for anything older.

Review mode

Frozen since 2026-07-28. The 2.4 and 2.5 triggers were retired, so both files below are static. Both ended their run empty: zero flagged addresses, zero suspicious pairs. The view still renders, it just no longer updates.

The public dashboard surfaces balances, growth, and engagement only — it does NOT list flagged addresses or suspicious pairs. Those live in Slack and in two JSON files in data/.

A casually-gated operator surface is available at /?review=<VITE_REVIEW_KEY>. It replaces the public layout with two tables:

• Always-winning flags — pulled from data/flagged.json (written by the 2.4 trigger).
• Suspicious pairs — pulled from data/suspicious-pairs.json (written by the 2.5 trigger).

Each row carries the trigger evidence (W/L/D, win-rate, pair count, share %) and a status badge (Active / Cleared). The filter bar at the top toggles between Active (default), Cleared, and All.

Security model — important to understand:

• VITE_REVIEW_KEY is compiled into the JS bundle as a plain string at build time. Anyone who downloads the bundle and greps it can find the key in seconds.
• The underlying JSON files are already public at raw.githubusercontent.com/AstarNetwork/yoki-monitor/main/data/. The review URL hides the rendered view from a casual visitor; it does NOT hide the data from anyone who knows where to look.
• This is "don't put it on the front page" gating, not authentication. The §54 signal-leak tradeoff still applies — the heuristics themselves are documented in the spec.

If real concealment is required, the two trigger output files would need to live in a separate private repo with the cron pushing there.

What this is NOT

• Not preventive — surfaces signals, does not pause matches or freeze accounts.
• Not real-time: since 2026-07-28 the crons run once a day, so an outflow alert can be up to ~24h behind the event. Accepted tradeoff for maintenance mode; tighten treasury-flow.yml if the game becomes active again.
• Not watching JKP any more: the §54 detector was retired 2026-07-28. See Phase 2.
• Not an in-page alerts surface — alerting lives in the GitHub Actions cron workers and posts to an internal Slack channel.