feat(daemon): standalone checkpoint loop + dashboard rework (#505)

[oaksprout]

Run-mode

FEATURE

Context

Closes #505 — operator nodes on realistic cadence (≥1 novel solution per livenessPeriod) were silently failing every L2 staking checkpoint because the activity window stayed open indefinitely. The existing RewardClaimLoop only fires checkpointAndClaim when calculateStakingReward > 0, but calculateStakingReward returns 0 when liveness is already failing — chicken-and-egg until manual intervention or eviction.

Same session surfaced a SQLite NOT NULL constraint that was mis-flagging successful deliveries as FAILED in the dashboard (on-chain deliveries succeeded, local store insert reverted), plus a chain of dashboard cleanups around the Activity table and Wallet card that operators asked for live during the test session.

Summary of changes

Daemon — new CheckpointLoop (client/src/daemon/checkpoint-loop.ts)

• Fires bare checkpoint() on each unique staked proxy at livenessPeriod cadence.
• Config: checkpointIntervalMs (default 300_000), env JINN_CHECKPOINT_INTERVAL_MS.
• Gated by stakingMode === 'standard'. Master EOA signs (permissionless call, no Safe execTransaction).
• Idempotent and per-call error-isolated so one bad proxy doesn't stop the rest.

Store — artifacts.desired_state_id NOT NULL fix (client/src/store/store.ts)

• Legacy DBs predating Task-native IDs ([#220] PR 4: real tJINN earned dashboard #406) defined desired_state_id as NOT NULL. Newer code only writes task_id, which made post-success delivery writes revert with NOT NULL constraint failed: artifacts.desired_state_id, mis-flagging the dashboard row as FAILED.
• SQLite can't ALTER COLUMN to drop the constraint, so Store now detects the legacy column at startup (hasLegacyDesiredStateId) and mirrors task_id into it on insert.

Dashboard — Activity table (ActivityCard.tsx)

• Swap Run / Task column order (Run first).
• Rename state badge pending → active.
• Remove the 50-row cap + 360px ScrollArea clip so all rows are visible.
• Filter "never engaged" FAILED rows: runStartedAt === null OR failureReason matches "not ready" / "not enabled". Catches expired evaluator opportunities and harness-not-enabled rows; keeps real failures visible.
• Single-SolverNet operators see all rows regardless of manifestCid — previously the delivery manifest CID didn't match the SolverNet manifest CID, silently dropping every COMPLETE row.

Dashboard — Wallet card (WalletCard.tsx, Overview.tsx)

• Wire the existing rewards.claimedStakingRewardsWei field through to the CLAIMED widget (was reading a non-existent claimedJinnLifetime and defaulting to 0).
• Add a Service field (e.g. #50) under Identity alongside Agent / Master / Safe, so operators can quote their on-chain service id without digging through jinn status.

Tests

• New CheckpointLoop suite (6 tests): per-proxy dedup, skip non-staked services, per-call error isolation, intervalMs=0 no-op.
• ActivityCard.test.tsx updated for active label + new filter behavior (both "never engaged" signatures + a real-failure keep case).
• WalletCard.test.tsx fixture updated for the new serviceId field on ServiceIdentity.

Acceptance

• CheckpointLoop ticks every livenessPeriod and is gated by config.
• Loop survives one bad proxy without stopping subsequent calls.
• Loop is a no-op when intervalMs = 0.
• Successful deliveries on legacy-schema DBs no longer mis-flag as FAILED.
• Activity table shows all qualifying rows (no clip), with Run column first and active for in-flight state.
• Evaluator-not-enabled and "never started" FAILED rows are filtered; real solver failures remain visible.
• Wallet card CLAIMED widget reflects the daemon's claimedStakingRewardsWei total; Identity panel shows on-chain Service #id.
• vitest run clean for checkpoint-loop, ActivityCard, WalletCard suites.

Out of scope

• Replacing the L2 stOLAS (≡ L2 JINN) → L1 tJINN bridge readout: the dashboard's CLAIMABLE / CLAIMED widgets still surface L2 reward queue values rather than real L1 distributor balance. Honest fix needs a JinnDistributor read or an L1-balance widget rename.
• Evaluator role activation flow surfaced by the new filter: the filter masks the symptom (FAILED rows), but the underlying "operator joined evaluator without jinn harnesses enable swe-rebench-v2-evaluator" UX gap remains.
• Resolving the nonce race between RewardClaimLoop and CheckpointLoop on the master EOA — the failure is non-fatal (loop logs and retries on next tick) but warrants a follow-up to coordinate transactions or move the checkpoint call onto the Safe.

Reference

• Issue: feat(daemon): standalone checkpoint loop to keep staking liveness window short #505 (this PR closes it)
• Activity-checker rules: contracts/src/staking/TaskActivityCheckerV3.sol (isRatioPass, recordSolutionDelivery)
• Manual checkpoint that proved the rate math: tx 0x1e4d75598701fbbe6066b8275eb77f1854d8369f28268b1990d45436306e3a0d
• Real-world earning evidence from the test session: ~0.31 JINN claimed pre-split across multiple distributor.claim calls after the checkpoint reset (see operator daemon logs).
• Earlier tJinn surface work this builds on: [#220] PR 4: real tJINN earned dashboard #406 (status API), [codex] expose real tJINN status balance #447 (real WalletCard balance), refactor(dashboard): reframe activity rows as runs + tJINN unit fixes #504 (dashboard label cleanup).

🤖 Generated with Claude Code