Second Order is an onchain Alpha Crash Test. If I copy this wallet, what do I actually receive once delay, size, competing flow and the source's own exit are priced in? It runs on Mobula's wallet, trade, market and security data and returns a private pre-trade verdict: ALLOW · RESIZE · BLOCK.
Project Brief | Quickstart | Demo Video | Architecture | Docs | Submission | Slides |
Live Demo | Stream API health | Stream API reference
Copy-trading products rank wallets by their realized return. A wallet that made +186% is shown as something to follow. But the follower is not the source: they arrive seconds later, they pay the price impact of everyone who arrived before them, they exit after the source has already sold into the same pool, and if a hundred other followers had the same idea, the trade's edge is gone long before the last of them fills.
A profitable source wallet is not necessarily profitable to copy.
Second Order replays the source trade, runs a shadow-follower simulation across delays and sizes against observed quotes and competing flow, drains a Remaining Alpha meter as capacity is consumed, and evaluates your intended size and delay against what survives.
The utility tells you:
- the source's realized return, as a leaderboard would
- how much scenario-compatible follower capital remains at your delay, and how fast it drained
- your own scenario-adjusted outcome at your size
- the maximum scenario-compatible size, if any
- whether the source exited while followers were still holding
- what every number rests on, and how confident the model is
Second Order is a two-part system:
- A stream service that turns Mobula observations into normalized, versioned events
- A browser utility that reduces those events into a capacity surface and a verdict, locally!
Alpha Capacity Surface
C(delay, crowd AUM)= the maximum aggregate follower capital for which the trade's scenario-adjusted follower expected value remains positive.
Remaining Alpha is C at your delay given the flow already observed. The CrowdGuard verdict compares your intended size against it.
- Price the follower, not the leader. Every number on the primary surface describes what a follower would receive.
- Evidence before verdict. F9 opens the Evidence Log: provenance, model inputs, data quality, the latest quote column, the security snapshot, the competing-flow ledger and the assumptions.
- Fail conservative. Missing or stale market, quote or security data can only move a verdict toward
RESIZEorBLOCK. - Name the provenance.
DEMO SCENARIO,ESTIMATED RECONSTRUCTIONandLIVE WITNESSEDare visibly different states everywhere a number appears. - Private by default. Size, delay and policy stay in
localStorage.
At a high level, Second Order estimates this statement:
A follower of size
SenteringDseconds behind this trade, behind the flow that actually arrived before them, exiting after the source's remaining position is sold, would receive this scenario-adjusted outcome.
From Mobula:
- the source wallet's trades (
/api/2/wallet/trades) → the anchoring buy and later exits - the wallet's trading analysis (
/api/2/wallet/analysis) → the leaderboard number - the pool's trade history (
/api/2/token/trades) → competing same-direction flow - 5-second candles (
/api/2/market/ohlcv-history) → the price path after the trade - market details and token security (
/api/2/market/details,/api/2/token/security) → depth, taxes, honeypot flags - with a Growth-plan key: the quoting and fast-trade WebSocket streams → live witnessed quotes and flow
Every observation becomes a
DomainEventwith an explicit provenance kind. Duplicate id's are dropped. Every frame is validated at the boundary.
- Quote grid. Observed quotes over (delay, size) become a continuous entry-price function: linear in delay, log in size, constant-product extrapolation marked as such beyond the observed envelope.
- Entry. The follower pays the average price over their slice of the impact curve at their delay.
- Exit. The target is the source's typical gain until a source exit is witnessed, then the current observed spot: no further upside is assumed once the source has left. The source's unsold remainder is sold first; exit depth is the lower of reported liquidity and the depth the latest quotes imply.
- Costs. Buy/sell taxes from the security snapshot, a proportional platform fee and fixed gas per side.
- Capacity. EV rises then falls in size; a golden-section search finds the peak and a bisection finds the upper root. That root is
C. - CrowdGuard.
ALLOWif your size is at or belowCand every input is fresh and complete.RESIZEif a smaller size survives.BLOCKotherwise, or on a critical security flag.
The 100 shadow followers are 100 sampled (delay, size) scenarios re-evaluated as flow arrives.
Press F5. A wallet with +186% realized return has just bought. Shadow followers board along the delay axis, green at first. Competing flow enters, execution depth thins, the source exits 55% of its position, Remaining Alpha collapses from $13,925 to $84. The CrowdGuard dialog reads: RESIZE do not copy at $1,000. Scenario-adjusted outcome −12.4%. Maximum scenario-compatible size $84. → Copy anyway is greyed out.
That run is a calibrated synthetic fixture and says so. Then switch REPLAY to a real one:
- Fake CBBTC honeypot on Base. A real wallet bought a token named CBBTC in a $250k pool; Mobula's static analysis flags it as a honeypot with a 100% sell tax; the wallet itself sold 86% four minutes in. Verdict: SECURITY BLOCK.
- FLOCK on Base. 109 same-direction trades in five minutes, pool deep enough that a $1,000 copy stays scenario-compatible.
Or press F2, type any wallet, and press F6: the utility reconstructs the minutes after its latest buy from Mobula history, in about fifteen seconds.
- pre-trade checks in copy-trading UIs
- wallet leaderboards that show follower capacity next to source return
- position sizing for signal groups
- a prospective Crowdproof credential, earned only when a wallet's signals survive witnessed follower conditions
| Label | Means | How it is produced |
|---|---|---|
| DEMO SCENARIO | Synthetic fixture, nothing from a market | Seeded generator, calibrated to the brief's numbers, committed with a manifest and disclosure |
| ESTIMATED RECONSTRUCTION | Real Mobula history fetched after the fact; quotes inferred from the price path and current depth | Stream service, any wallet, keyless demo API or your key |
| LIVE WITNESSED | Captured from Mobula streams while it happened | Growth-plan key; quoting + fast-trade WebSockets |
| Component | Technology | Purpose |
|---|---|---|
| Data | Mobula REST + WebSocket | Wallet trades, pool history, candles, security, quotes, flow |
| Contracts | Zod | Versioned event envelope, API payloads, raw Mobula validators |
| Model | TypeScript (pure) | Quote grid, capacity solver, reducer, CrowdGuard; unit and property tests |
| Stream | Fastify + SSE | Providers (replay, reconstruction, live), dedupe, fan-out, capture |
| Persistence | PostgreSQL + Drizzle | Events, capacity snapshots, replay manifests, errors; sessions rebuilt after restarts |
| Web | Next.js 16 + React 19 + Tailwind 4 | The setup utility; VT323 + Archivo self-hosted |
| Dialogs | Radix (behaviour only) | Focus, Escape, aria for the Win95 windows |
| Tests | Vitest + Playwright | 55 unit/integration tests, 15 e2e at 1440×900, 1280×800, 390×844 |
| Hosting | Vercel + Railway | Web on Vercel; stream + Postgres on Railway |
- Utility: https://second-order-crash-test.vercel.app
- Stream API: https://stream-production-900a.up.railway.app (
/health,/api/replays,/api/capabilities,/api/sessions) - Repository: https://github.com/MihRazvan/second-order
corepack enable && pnpm install
pnpm dev # web :3000 · stream :4010 (keyless Mobula demo API unless MOBULA_API_KEY is set)
pnpm test # unit + integration
pnpm test:e2e # Playwright, needs pnpm devMore in the Quickstart.
Built with <3 during the Crypto Canal's Common Sense hackathon, 2026 in Amsterdam.