One market.
Calibrated on it.

What a tick is worth

A market maker posts a bid and an ask and waits to be hit. Whoever queues first at a price gets filled first, and on a busy book that queue is long. The cheapest way to the front is to improve the best price by one tick — the smallest increment the venue allows. You give up one tick of edge and you buy priority.

Whether that trade is good depends entirely on what a tick is worth on your market. On Ondo gold, one basis point is 46 ticks. So one tick is about 0.02 bp — priority is very nearly free. On silver the same single tick is 1.47 bp: roughly sixty-eight times more expensive, and enough to hand back a large share of the spread you are trying to earn.

This is why Vulcan is a market-specific mode and not a preset with a gold label on it. A symbol guard in the engine zeroes the tick improvement on silver even if the mode is selected there. Same code, different arithmetic, opposite decision.

Two levers, measured separately

Vulcan is the general-purpose gold mode plus exactly two changes. We ran each one on its own before running them together, because two changes at once tell you nothing about either.

• One tick inside

  Step in front of the queue instead of behind it. Shadow run: +9 CPM. Independently, a 24-hour live measurement over 282 367 frames landed on 8.5 — the numbers agreed.

• Requote at 0.2 bp

  The old threshold of 1.0 bp is 46 ticks: the quote deliberately sat still while the market walked away from it. At 0.2 bp it is 9 ticks, and the quote follows. Shadow run: +8 CPM, holding across three independent windows.

• Together

  +18 CPM. The two levers add up rather than cancelling — they fix different halves of the same problem: getting filled, and being filled at a price that still makes sense.

CPM here is cost per million of turnover, in dollars. Lower is better; it is what a market maker actually pays to run. We report it as a delta because the absolute level swings with the day — hour-to-hour noise on this book is about ±29.

Patient with time, impatient with direction

Underneath those two levers sits the gold inventory policy, and it is unusual in both directions at once. It will sit in a position far longer than a general-purpose bot. It will cut a position that keeps moving against it far sooner.

• Holds up to 90 min

  Against 12 minutes on the generic mode. Time to leave on its own terms instead of at whatever price is available right now.

• 30 min maker grace

  Against 3 minutes. Nearly all of the loss in this strategy lives in forced taker exits, so the exit is given room to work passively first. The taker is the last resort, not the first.

• Cuts drift at 6 bp

  Against 14. The old threshold effectively never fired: typical adverse drift on gold runs 0.4–1.8 bp, so 14 was a ceiling above the sky. Six is a real limit.

• Inventory lock

  Once a position exists, the side that would grow it goes dark and only the reducing side keeps quoting. Measured effect: taker exits fell from 1.42/h to 0.89/h, and runs of three or more clips from 7.5% to 4.1%.

The lock has a real price: turnover drops, because half the time only one leg is quoting. We turned it off for four days in August on the strength of two measurements that showed nothing, then put it back — the measurements were correct but were measuring the wrong thing. Its true cost is around 7.7 CPM, which is invisible against ±29 of hourly noise. Three independent signals showed the mechanism works.

It warms up, and it refuses to guess

• First 120 seconds

  No inventory is built at all. It watches the book before committing. A session inspected inside this window looks idle because it is — that is the design, not a fault.

• Fail-closed entry

  If the engine cannot read your position, it stops opening. It does not assume you are flat and trade on the assumption.

• 60-second freshness

  A position reading older than a minute is treated as no reading. Without this rule, “we checked an hour ago” is indistinguishable from “we just checked”.

Give it a stop it can live with

This is the one setting worth thinking about, and we learned it the expensive way rather than from theory. Vulcan’s entire mechanism is holding inventory for up to ninety minutes. While it holds, the unrealised P&L on that inventory swings. That is normal and temporary.

A stop-loss carried over from a faster mode cuts it off mid-swing. In the first days live, every session running a five-dollar stop was halted within hours — each one breaching the threshold by a few cents — while sessions with a wider stop kept running for days. Nothing was broken. The protection did exactly what it was told.

Set the stop against the size you actually trade, not against a number that made sense for a mode which closed positions in twelve minutes. A stop far smaller than your inventory ceiling will fire on ordinary movement.

Four steps