veiledquant @veiledquant

5) Here's the part that breaks people's brain: EVERY plant gets paid the price of the LAST one swithced on. Not their own cost. The marginal plant's cost This is called the market clearing price (MCP). One price, paid to all.

0 0 0 9 0

View Details

veiledquant @veiledquant

2 months ago

4) Plot it and you get a staircase, the supply curve. Demand is (roughly) a vertical line. Whereever it crosses the staircase, that's the last plant you need to switch on to keep the lights on.

1 0 0 14 0

View Details

veiledquant @veiledquant

2 months ago

@GryptoGoon knowledge transfer without autism whisperer... does scott know?

0 0 1 518 0

View Details

veiledquant @veiledquant

2 months ago

@FirstSquawk Every new headline says “deal close” and then the next sentence quietly adds three more unsolved wars to the term sheet.

0 0 2 684 0

View Details

veiledquant @veiledquant

2 months ago

5/ Split by time. Train on the past, test on the future. You can't shuffle rows the way you would with a normal dataset, markets are a time series. Random shuffling drags future information backward and hands you results that mean nothing. 70/30, clean cut.

0 0 1 32 1

View Details

veiledquant @veiledquant

2 months ago

4/ The lagging is the whole game. .shift(1) on features, .shift(-1) on the target. Get this wrong and your backtest looks incredible for a reason that has nothing to do with edge. Lookahead bias is the #1 way people fool themselves.

1 0 1 32 1

View Details

veiledquant @veiledquant

3 months ago

the actual jump logic, close to what the paper describes:

0 0 0 15 0

View Details

veiledquant @veiledquant

3 months ago

in practice they approximate the LOB state with two variables: It = sign of best-price volume imbalance (V_bid − V_ask) / (V_bid + V_ask) σt = spread so the consistency condition becomes: find t' where the imbalance sign and spread match, and the effective events of the agent's action appear somewhere in the next 20 events after t'. they precompute dictionaries keyed by (imbalance sign, spread) → list of time indices. the jump target is then drawn at random from the matching candidates, excluding times too close to current (within 10 ticks). if no match exists, they continue sequentially.

1 0 0 24 0

View Details

veiledquant @veiledquant

3 months ago

@CRUDEOIL231 The market wants to trade “deal soon,” but this reads like political theater, not a closing document. The hard issues are uranium, sanctions/blockade, Hormuz, and domestic legitimacy. Adding Abraham Accords optics probably makes the landing harder, not easier.

0 0 6 651 0

View Details

veiledquant @veiledquant

3 months ago

the glue holding it all together is the MessageBus a pub/sub event bus in Rust. every component (DataEngine, RiskEngine, ExecEngine, your Strategy) publishes and subscribes to typed topics on this bus. nothing talks to anything else directly. this is how the same strategy code works in both backtest and live: the bus abstraction is identical. only what sits behind it changes, historical data replay vs. live WebSocket feed.

0 0 0 37 0

View Details

veiledquant @veiledquant

3 months ago

the architecture is "Rust at the core, Python at the edge." - Rust (nautilus_core): NautilusKernel, MessageBus, Cache, DataEngine, RiskEngine, ExecutionEngine. all typed, thread-safe, compiled. - Python/Cython layer: strategy API, indicators, adapters. bound to Rust via PyO3, zero-copy where possible. your strategy code sits in Python. everything beneath it runs in Rust. you feel none of the overhead.

1 0 0 87 0

View Details

veiledquant @veiledquant

3 months ago

@DeItaone That line sounds simple, but it’s doing a lot of work. “Remove the nuclear threat” can mean anything from inspections + limits to full dismantlement + uranium removal. ...The market will care a lot about which definition actually makes it into the paper