Overview
Is the book healthy, and did it actually trade? Everything below is live from the broker.
Signals · did the machine do its job?
Book health · continuous review
Performance
The track record since inception — is this actually making money, and how?
Track record
Closed trades · realized P&L
Positions
Everything held right now, largest exposure first.
Strategies
Which sleeve is contributing — and whether that contribution is real alpha or factor beta.
Per-sleeve P&L
Real-alpha oversight · factor attribution
Risk & sizing
Exposure against every hard limit, the books reconciliation, and the sizing tools applied this step.
Limits
Per-strategy exposure gross / net · #pos
Sizing · tools applied
Activity
What the bot has actually done — orders that reached the broker, and recent ops events.
Recent orders
Ops timeline
Event log
Every strategy, risk and lifecycle decision as it happens — the hedge being sized or skipped, a sleeve going on or off, exposure climbing toward a cap, a gate refusing a step. Newest first.
Go-live readiness
Would real capital be justified yet? Every row must be GREEN before funding.
Validation & replays
The gate verdict behind each sleeve, and the vectorbtpro replay charts that back it.
Full-stack verdict per sleeve
Backtest evidence · vectorbtpro replay
uv run python scripts/build_dashboard_visuals.pyData catalog what a study can be written against
Before an idea is worth writing down: is the data here, from whom, and how far back? Spans come from one probe file per store — an indication of coverage, not a promise that every symbol reaches the same date.
ttengine pipeline stream → Redis → aggregator → QuestDB
Recent additions what changed since the previous scan
Vendors
On disk
Research pipeline where ideas enter, and where they die
This is not an account. It is the programme that decides what an account is ever allowed to trade. Attrition is the product: a falsification ships the same artifacts as a confirmation.
Waiting on you a machine proposed this; only you can confirm it
The funnel
Research steps every step is a notebook you can re-run
Replay coverage two engines, independently, on the same weight matrix
Verdict mix gate status across the sleeve population
Multiple-testing position the denominator every result is deflated at
Backtest runs the evidence under the verdicts
Every strategies backtest writes a run to the registry: params, config, metrics,
span and an equity curve. The gating panel shows what those runs concluded; this shows
the runs themselves, because a PASS over 150 sessions is not the same object as a PASS over 2,500.
Registry
Recent runs
Trials ledger the multiple-testing denominator
Every Deflated Sharpe in this repo is deflated at N, the total trials this programme
has spent. Since the 2026-08-30 migration every row carries a machine-readable
spend, so canonical_n() returns the number instead of refusing to
guess — and still fails closed if a cell ever stops parsing, because a wrong N reads as
rigour while deflating nothing.
Reconciliation
Adjudication queue rows no parser may infer
Research gating
Every sleeve's position on the G0–G7 ladder, each verdict shown with the gate's own measured error rate. A §21 PASS is evidence, not proof — these numbers say how much.
The gate ladder what clearing each rung actually proves
The §21 bar what a verdict is measured against — configurable, and stamped on every report
Gate calibration what a PASS is worth
Ladder · per-sleeve evidence
Declaration gaps what is not frozen yet
Allocator pre-flight ASSP-09-R4 · conflict C1
Controls
Bounded, audited changes to live parameters. Out-of-bounds values are refused, never clamped.
Functions
Brain
The LLM investing loop on paper-2 — what it is allowed to do, whether it is doing it, and the reason it gives when it does not. Read-only.
Why proposals are not landing
Limits · what can block a trade
Services
Reviewer guidance
—
Log
—
Fomo book
The LLM decision book on paper-2 — orchestrator-fed; strategies decision new to add manually.