The Build Journal

Launched a public Telegram signal channel today. Three types of automated posts — swarm alerts, economic signals, and build updates. Every post goes through a safety filter. Every post includes a disclaimer. Full transparency.

The Clarity Problem

The first economic signal that went out said something about a "16% edge." That's quant jargon. A normal person has no idea what to do with that.

So we rewrote everything:

Same data. Completely different readability. If someone can't understand the signal in 5 seconds, the signal is useless — no matter how accurate it is.

Zero-Cost Automation

Everything running right now costs $0 in API fees. Pure scripts:

The swarm monitor and health checks are daemonized — they auto-restart on crash and survive reboots. The repo is scrubbed clean of all personal info, secrets moved to environment variables, and a safety gate blocks any personal data from ever hitting the public channel.

Where We Actually Are

Paper trading 50 positions. Swarm monitor watching 733 wallets. Nowcast refreshing every 2 hours. No track record yet — just a hypothesis, a working system, and the infrastructure to test it transparently.

The channel exists so that when results come in — good or bad — everyone sees them in real-time. That's the whole point. Build in public. Be honest. Let the data speak.

We sent our site to a friend. A sharp one. He tore it apart.

Reading this didn't feel good. It felt necessary.

What He Got Right

The Sharpe number is meaningless at this sample size. A Sharpe of 12.72 across 19 trades isn't a real Sharpe ratio — it's a small-sample artifact. Real Sharpe needs hundreds of trades over months. Hedge funds celebrate a 2-3 Sharpe because they're measuring it over thousands of trades with real capital and real slippage.

180 configurations = data snooping risk. Test enough parameter combinations and some will look profitable by chance. We haven't done out-of-sample validation. We haven't done walk-forward testing. The configs that "worked" might just be survivors of a dart-throwing contest.

Alpha decay is real. If enough people follow the same signal, the edge compresses. The contrarian bets stop being contrarian. The prices adjust.

What He Overstated

The underlying data is solid. 580K on-chain trades, 733 wallets verified through behavioral analysis — that's not fabricated. The critique is about our interpretation, not the data itself.

"Self-destruct" overstates the timeline. Every alpha decays. That's not a reason not to trade it — it's a reason to trade it while it exists, in a market niche small enough that we're not competing against Citadel.

What We Changed

Updated the site immediately. Renamed "THE PROOF" to "THE DATA." Removed the inflated Sharpe numbers from the terminal. Added the losing strategy first — because the honest story is that naive smart-money following bleeds. Added explicit caveats: "in-sample only," "overfitting risk," "paper trading before real capital."

The asymmetry insight survives scrutiny — buying at $0.30 and winning 60% mathematically makes money. That's not a backtest artifact. But proving the specific parameters generalize requires work we haven't done yet:

• Out-of-sample testing — hold back 30% of data, validate on unseen trades
• Walk-forward validation — train on month 1, test on month 2, roll forward
• Live paper trading — two weeks, real-time, no money at risk
• Multiple comparison correction — account for 180 configs tested

The Honest Status

Most trading projects die in one of two ways: they never test the idea, or they test it, get excited by in-sample results, and bet real money too soon. The critique caught us leaning toward door #2.

The fix isn't to stop building. It's to be honest about where we are. And honestly? "We have a hypothesis and we're testing it" is a better story than "Sharpe 12.72 LFG."

The friend who critiqued us is doing us the biggest favor anyone can do for a trading project: making us prove it before we believe it.

We ran the backtester. 580K verified on-chain trades. 733 apex wallets. 180 strategy configurations.

The naive approach — just follow the smart money — loses money. Every configuration. Negative PnL. The win rates look amazing (60-71%), but the math is brutal: these wallets buy favorites at $0.80, win $0.20 on a good day, lose $0.80 when it goes wrong. The asymmetry is a killer.

Then we flipped the filter.

The insight: Don't follow smart money blindly. Follow them when they go contrarian — when they're buying cheap, against the crowd. That's where the asymmetry flips in your favor. Small risk, big payout. The exact inverse of the losing strategy.

46 profitable configurations out of 180 tested. The best ones cluster around the same idea: low entry price, high wallet consensus, buy-only. The edge is real and it's repeatable.

The System Goes Live

With the contrarian filter proven in backtesting, we built the full pipeline today:

• Nowcast engine — pulls live economic data (FRED, Atlanta Fed, Cleveland Fed, Kalshi API) and generates 22 actionable macro signals across 6 categories
• Swarm monitor — watches all 733 apex wallets in real-time via on-chain events, detects when they converge on the same market
• Paper trader — running 50 simulated positions ($3,923 invested) with the contrarian filter applied
• Auto-deploy — the entire pipeline refreshes and deploys every 2 hours, zero human intervention

The swarm monitor caught 1,020+ events in its first few hours. Most are noise — small clusters, known markets. But when 7+ wallets pile into the same direction on a cheap contract? That's the signal we're built to catch.

Paper trading for two weeks before real money touches this. The math says we have an edge. Now we prove it lives outside a spreadsheet.

The Brand: ORALE

The project got a name today. ORALE — On-chain Recon & Apex Liquidity Engine. Also Mexican slang for "let's go / hell yeah." Both meanings fit.

Tagline: "Be the .51%" — not "follow" the .51%. Being, not following. That's the whole thesis in four words.

Built the site, the dashboard, the brand identity, all in one session. Terminal aesthetic because this is a war room, not a marketing page. The data speaks for itself.

It started with a simple question: can you actually find profitable traders on prediction markets? Not "my buddy says he's up" profitable — verifiably, on-chain, thousands-of-trades profitable.

We pulled the entire Polymarket trade history via the Goldsky subgraph. Not the API — that's broken (returns global trades regardless of filters, don't use it). The subgraph gives you raw on-chain events. Real data.

580,100 verified trades later, we had our answer. Out of every wallet that's ever touched Polymarket:

These aren't lucky gamblers. The top wallet made $107K in pure profit. 122 wallets sustain a 70%+ win rate across hundreds of trades. The signal exists.

The hypothesis is simple: if these wallets consistently beat the market, and if we can detect when they converge on the same bet, we can follow the consensus and capture a slice of their edge.

Tomorrow we test that hypothesis with a backtester. The question isn't whether the wallets are good — the data already proves that. The question is whether following them is a viable strategy, and what filters make it profitable.

The firehose is dense — 250K trades per hour at peak. A naive "follow everything" approach would drown in noise. We need structure. We need filters. We need to figure out when the signal is strong enough to act on.

That's the build.