[Bug] ISSUE-0021: rebuild live leaderboard and local E2E #33

Context

The latest local walkthrough exposed that the MVP completion claim is not supported by the current implementation or evidence:

• Battle.sol correctly emits ChallengeCreated with only challengeId, challenger, and opponent indexed, but the backend ABI and parseAbiItem mark all five fields indexed.
• handleChallengeResolved compares an address to a token ID, so winner/loser addresses are not derived correctly and leaderboard rows remain empty.
• /leaderboard is still hard-coded demo data and explicitly says the live indexer is future work.
• Monster detail screenshots show broken images despite committed PNG art.
• The backend has no tests, CI only type-checks/builds it, and the current E2E evidence has no mobile, console, a11y, or resolved-live-leaderboard proof.

Goal

Deliver a reproducible fresh-Anvil flow from mint/hatch/train/battle through indexed API data and rendered live UI, with evidence strong enough to support the MVP acceptance claims.

Scope / allow-list

• backend/**
• frontend/**
• .github/workflows/ci.yml
• contracts/test/** only if an event-contract regression assertion is required
• scripts/** for deterministic local E2E orchestration
• docs/evidence/0021/**
• PROJECT_STATUS.md, sessions/**, memory/**, docs/INDEX.md, ENGINEERING.md, TESTING.md for truthful status and durable lessons

Non-goals

• Monad testnet deployment ([Spike] ISSUE-0019: deploy to Monad testnet + agent-browser E2E #24)
• Contract storage/layout/supply changes
• New gameplay features or art generation

Acceptance criteria

• Backend event ABI exactly matches the Solidity event declarations; ChallengeCreated decodes both token IDs and both addresses from a real Anvil log without ad hoc topic assumptions.
• ChallengeResolved maps winner/loser token IDs to the correct trainer addresses, awaits post-battle monster synchronization, and records turns/draw/transaction metadata.
• Starting from a fresh SQLite DB after the battle has resolved replays history into one correct battle row and a non-empty leaderboard; restarting from the checkpoint is idempotent.
• Backend has automated tests for event decoding, resolution mapping, leaderboard ranking, API validation, replay, and restart behavior; CI runs them.
• Frontend leaderboard consumes the live backend API and implements loading, populated, empty, and unavailable states with no demo entries or Phase 2 placeholder copy.
• Monster detail art uses committed assets and renders successfully for the E2E species/stage.
• A one-command or documented deterministic local E2E run starts fresh Anvil, deploys contracts, executes two-wallet mint/hatch/train/battle, runs indexer/API/frontend, and asserts contract/API/UI state.
• Desktop 1440x900 and mobile 390x844 screenshots prove the populated live leaderboard and monster details; browser console is clean and axe reports no serious/critical findings.
• Contract tests, backend tests/typecheck/build, frontend tests/typecheck/lint/build, and GitHub CI pass.
• PROJECT_STATUS.md and session/memory records match the verified state and distinguish local proof from pending testnet deployment.

Evidence requirements

Store change-summary.md, implementation-plan.md, raw test logs, API/contract traces, desktop/mobile screenshots, Playwright/axe results, reviewer reports, and verification.md under docs/evidence/0021/.

Reviewers

Required cold-start roles: bug-hunter, behavior-reviewer, architecture-reviewer, security-reviewer, and ui-reviewer. No Critical/High finding may remain open.