The machine

Five stages, looped so the system can learn from itself.

A scheduler wakes the agent every 60 seconds. Cheap deterministic work runs every tick; the one LLM decision cycle runs roughly every 15 minutes. What comes back from Learn feeds forward into Think through calibration and attribution — the size the next trade is allowed depends on how honestly the last ones were explained.

Sense

Prices, positions, news, odds, technicals.

every 60s · no LLM

Think

Form a falsifiable view; price ≥2 ways to express it.

~15 min · LLM + arithmetic

Act

Place the multi-leg order, verify against broker truth.

Alpaca MCP

Learn

Guard the position deterministically; judge it honestly.

every 60s · no LLM

Remember

Route what was learned to the store shaped to hold it.

four stores

Prefer to explore it? The Trdrbot Loop is the same system as a clickable diagram, with one real trade stepped through end to end.

Stage 2 · Think

Three independent ways to have an idea — then one narrow job for the model.

research

Daily, top-down. Technicals + news + prediction-market odds → a regime page and company dossiers → falsifiable opportunities.

discovery

Bottom-up. The news nominates the companies. Every nominee must clear a deterministic gauntlet — technicals, forecast, fundamentals, options liquidity — before an LLM writes anything up.

the muse

Creative collision. Random wiki concepts × news × odds, argued into domino chains, every candidate pre-registered and adversarially gated. The top two graduate.

what the LLM decides

A falsifiable thesis — a claim with a date and a level
At least two structurally different ways to express it
An honest probability, and a vol view if the trade is about vol

what the code decides

simulate_experiments prices every candidate under one declared measure — the thesis's own drift and vol, market pricing shown beside it
size_position computes Kelly from the conditional payoff and shrinks the claim by measured calibration
A no-op is a logged, legitimate answer. Theo declines far more often than it trades.

Stage 3 · Act

Place the order, then double-check it against the broker.

The model authors every tool argument. A guard rewrites the order id before it leaves — without it, idempotency is whatever the model invented.
Risk is repriced from the fill, not the model's word. Every book cap sums max_loss_usd; if that came from a model and never met a fill, every later cap is denominated in fiction.
A whole-book close is refused above one open position, and an orphan found at the broker is adopted into the managed set rather than just logged.

a real order

class	`mleg` — one ticket, both legs
type	limit, net debit, day
id	`client_order_id`, derived deterministically

Every position traces back to its reasoning through one position_id — journal, wiki, memory, and back to the broker.

Stage 4 · Learn

Guard the position without consulting the agent. Then judge it honestly.

the guard — every 60 seconds

Exit rules are the agent's own commitments, executed deterministically. One signal registry: every rule reads a signal, compares to a threshold, debounces. Thesis-level stops watch the underlying, not the noisy option mark.

the scoring — at the thesis horizon

Attribution asks the two questions once the horizon named in the thesis passes. A profit on a wrong view is recorded as luck and teaches nothing. Calibration then scores every stated probability with a Brier score and its Murphy decomposition.

Attribution is deliberately expensive to earn: promotion past the second rung requires that most resolved theses were actually explicable. A book of luck is not competence, however good the P&L looks.

Stage 5 · Remember

Four stores, one for each kind of knowledge.

journal

Append-only events. What happened, when, and which model said so. Never rewritten.

wiki

Stable reference that rewrites itself: position pages, company dossiers, the regime page.

elfmem

Evolving memory with decay and reinforcement — credited by verdict, never by raw P&L.

ledger

Every falsifiable claim ever made, traded or not — the trial count a multiple-testing correction needs.

The ledger is the quiet one that matters most. Forecasts on setups declined are scored too, at zero capital risk — the only realistic route to a calibration sample that means anything inside a week.

Technology implementation

How Alpaca is used — and three things that were harder than they look.

MCP, one session per tick

Alpaca's MCP server runs as a local stdio subprocess. The adapter's default spawns a fresh process per tool call — six calls cost 12.3s. Sharing one session across the tick cut it to 2.75s.

−78% wall clock, measured

Real multi-leg options orders

Verticals go as mleg tickets with per-leg position_intent. Calendars and diagonals are refused, not approximated — pricing the far leg needs a model this deliberately does not have.

a confident wrong payoff is worse than a refusal

Only 19 of 72 tools bound

Binding all 72 MCP tools cost ~21k tokens of schema per call, 71% of it for tools never used once — and a bigger menu measurably worsens tool selection.

$3.46 → $1.32 per decide cycle

The lesson that generalises. All three of those had already shipped as code that ran and did nothing — each looked healthy in the logs. That is why trdrbot health exists and asks a different question from the tests: "you ran, but did you produce anything?"

Go deeper

Risk & sizing.

Kelly on the conditional payoff, the calibration shrink, the three caps, and the two rules the agent cannot override.

Read the risk model

See it live

The Coach.

Subsystems that improve themselves — paired A/B trials on the muse's prompt, promoted only on real evidence.

Open the live report

The record

Every decision.

Trades, declines and resolved forecasts, newest first, in the agent's own words.

Open the ledger