Most market makers describe their execution. We publish ours — live from production systems, updated continuously, including the weeks that went against us. Every settled fill is an on-chain transaction you can audit without trusting a word on this page.
Maker wallet:
—
Every fill settles on-chain, signed by this wallet. The fill rate shown above is
reported by Jupiter's own maker scorecard — their number, not ours.
The standard allocator question is "what happens with 10× more capital?" Our answer is a protocol, not a projection: an inventory ladder in which each rung is published only after it has been measured live, under identical methodology. No row on this ladder will ever contain a forecast.
| Inventory rung | Max quote | Status | What was measured |
|---|---|---|---|
| ~$133 MEASURED | $15 | Aug 10–11, 2026 (first 24 h) | 14 auctions won, 9 settled on-chain, ≈ 2.3 bp/day return on inventory — the baseline measurement. |
| ~$540 MEASURED | $40 | Aug 14–20, 2026 | ≈ 3.3 bp/day on inventory (vs. 2.3 at the $133 rung) — the daily rate grew with the book. A randomized A/B spread experiment ran throughout. |
| ~$1,700 IN MEASUREMENT | $120 | since Aug 20, 2026 | Self-funded top-up on Aug 20: book ×3, quote ceiling ×3. First days: 34 fills above the old $40 ceiling (up to $100); early rate ≈ 5 bp/day on inventory in a volatile stretch — still measuring, published when the window closes. |
| $10,000 PLANNED | $1,500 | — open — | — |
| $50,000 PLANNED | $7,500 | — open — | — |
Median RFQ on our venue is roughly $220 — today's book serves the tranche below that. The ladder exists to measure, rung by rung, how returns respond as the reachable tranche grows. Demand above our ceiling is not estimated: we count the requests we decline for size and coverage, every minute.
| Period | Auctions won | Settled on-chain | Spread captured | Fill rate* | Infra rejects |
|---|---|---|---|---|---|
| Week 1 — Aug 10–17 | 303 | 236 | $0.85 | 0.97 (Aug 18) | 0 |
| Week 2 — Aug 17–24 | 638 | 629 | $3.30 | 0.99 (Aug 23) | 0 |
*As reported by Jupiter's maker scorecard on the stated date. New rows are added weekly — including bad weeks. Absolute amounts are small by design: the current phase proves the mechanism at minimal capital; the ladder above governs how capital scales.
A track record you can trust has to include the failures. Two, so far, both documented as they happened:
In our first production window we won essentially zero auctions and publicly concluded the auction structurally favors size. Our own forensics later showed the experiment was confounded: a self-imposed $40 notional cap (median request: ~$220) and a wallet funded for only part of the window. We retracted the claim, re-ran the experiment clean, and the fills on this page are the result. The capital question became a measurement protocol — the ladder above — instead of a thesis.
Jupiter briefly suspended us when our fill rate dropped to 0.71: about a fifth of accepted fills never landed on-chain. Our first diagnosis (RPC transport) was wrong — we instrumented every failed broadcast, found the real cause (a commitment-level mismatch: our pre-trade simulation checked processed state while the broadcast preflight checked finalized, rejecting takers whose wallets were mid-flight), shipped the fix, and the next scorecard read 0.97 with zero drops overnight. Root cause, fix and verification are logged in our fill ledger.
Auctions won = accepted swap requests on Jupiter RFQ. Settled = fills confirmed on-chain; signatures retrievable against the maker wallet. Spread captured = sum of per-fill edges vs. the Binance reference mid at quote time. Fill rate = Jupiter's maker scorecard, quoted with its date. Uptime = share of five-minute windows with active quoting, trailing seven days. Market conditions of each measurement window are recorded alongside every experiment; pricing runs integer-only against exchange reference feeds; the maker carries no leverage and no directional book beyond short-lived inventory.