Start with provenance.
Issuer facts, transparent calculations, editorial interpretation, missing sources, and unknowns remain visibly distinct.
Product case · Market research + trader training
Discovery: test a thesis. Train the decision.
Discovery turns one market question into an inspectable workflow: evidence, candles and volume, deterministic rule graphs, bounded parameter search, foreground signal receipts, and future-blind decision practice. Its immediate public path uses clearly labeled offline fictional fixtures and an official-source, price-free NVIDIA evidence case. A QA-backed 3:32 browser walkthrough shows the public workflow end to end.
- Available now
- Offline market lab, source-linked NVDA case, editable strategy DAG, bounded search, foreground signals, future-blind trainer, and product tour
- Deployment-gated
- Authenticated Alpaca U.S. history, server-authoritative trainer sessions, private model briefs, cloud history, and paid enrollment
- Not live
- Global-market coverage, background alert delivery, model training, broker connection, order execution, or autonomous trading
Recorded product proof · 03:32
Watch the public workflow run end to end.
A silent, QA-backed screen recording moves through evidence, candles and volume, the editable strategy graph, bounded search, foreground signals, and future-blind trader training. Use the chapter controls to jump directly to a capability.
Interactive visual transcript
One workflow, seven chapters.
Select a chapter to seek the embedded recording. The visible descriptions mirror its scene-caption track.
Recording boundary. This walkthrough shows deterministic public research and simulation paths only. It does not demonstrate protected market-data access, continuous background delivery, brokerage connection or execution, personalized investment advice, or a performance promise.
01 · Market need
People want a useful financial view—not another confident answer box.
Individual investors need help catching confirmation bias and naming what would change a thesis. Repeat professionals need a consistent, exportable research object. Teams and governed enterprises need source boundaries, permissions, retention, model controls, and human approval.
Discovery uses one provenance and receipt contract across those jobs. Every public price row is labeled as an offline fixture; every rule edit can be traced to its graph; every hypothetical fill states when the signal was observed and when it could execute; every unavailable service fails closed instead of silently substituting data or capability.
02 · Connected workflow
One question. Six connected, inspectable jobs.
The public workspace keeps the selected symbol or fictional instrument attached as a person moves from evidence review to chart inspection, rule construction, bounded testing, foreground monitoring, and deliberate practice.
Inspect the exact bar.
Mouse and keyboard inspection connect candles, volume, indicators, replay cutoffs, execution markers, and row-level context.
Edit the rule itself.
Typed sockets, weighted sums, safe rewiring, removal, undo, redo, and auto-layout keep the deterministic graph understandable.
Bound the experiment.
A cancellable browser worker searches declared parameter grids with progress, holdout diagnostics, benchmark context, and rejection receipts.
Keep signals local and explicit.
Foreground rules evaluate loaded completed bars, suppress duplicate fingerprints, and retain browser-local receipts while the page is open.
Reveal tomorrow only after deciding.
A seeded challenge advances one daily candle at a time, records BUY, HOLD, SELL, or target decisions, and returns a complete scorecard and ledger.
03 · Inspectable public product
See what runs—and what each result means.
Public release · reviewed August 28, 2026
Offline data, executable rules, visible receipts, and future-blind practice.
The immediate market lab provides four clearly labeled fictional instruments with 660 deterministic business-day OHLCV rows each. It supports daily and client-derived UTC weekly/monthly candles, volume, indicators, replay, prominent hypothetical BUY/SELL markers, exact fill receipts, and synchronized multi-instrument overlay, lane, and table views.
The trainer draws a seeded instrument and hidden window from the loaded set, reveals one daily bar at a time, and applies disclosed costs under a decision-at-close, hypothetical-fill-at-next-open policy. It offers untimed review or 1, 5, 15, 30, and 60-second browser-assisted turns. In the immediate local drill, an expired turn attempts at most one race-guarded automatic HOLD; browser throttling can delay the deadline. The final ledger exposes total and benchmark return, win rate, profit/loss ratio, reward/risk, drawdown, and the composite score.
Open the public market lab · Open the trading trainer · Inspect the source-linked NVDA case
Market demo4 fictional instruments · 660 deterministic business-day rows each
NVDA evidence3 official sources · 5 reported facts · 2 transparent calculations
Strategy studio25 deterministic block types · weighted DAG · immediate chart, log, and basket recomputation
Trading trainerUntimed + 1 / 5 / 15 / 30 / 60 seconds · race-guarded local auto-HOLD
04 · Product paths
Built for individual usefulness and governed organizational work.
Inspect, test, and practice.
The offline market lab, NVIDIA evidence case, product walkthrough, browser-local signals, parameter search, and fictional future-blind trainer require no account.
Bring approved private capacity.
Authenticated Alpaca U.S. history, server-authoritative trainer sessions, private evidence briefs, cloud history, and paid enrollment remain fail-closed pending production approval.
Standardize review.
Roles, versioning, disagreement, approval receipts, and shared evidence policy remain product-design work—not a live collaboration claim.
Govern approved sources.
Licensed news and social feeds, read-only database patterns, tenant boundaries, model policy, retention, alert delivery, and broker connectivity require separately scoped implementation.
05 · Automation boundary
Inspectable automation is not autonomous trading.
Bounded parameter search runs in a cancellable browser worker. Foreground signals evaluate only the completed bars already loaded while the page processes them. The decision timer is browser-assisted. None is a server scheduler, delivery guarantee, whole-market ranker, brokerage connection, or order executor.
The launch-gated private brief path is a constrained structured model call. Its receipts can record identity and origin checks, cost reservation and reconciliation, schema validation, and output-integrity checks. These are browser and server operational receipts—not hidden model chain-of-thought, proof that a source is true, or an instruction to trade. No server-side background runner, queue, daemon, external alert channel, or local coding-agent relay is enabled in the public release.
06 · Limitations
A structured answer can still be wrong.
Offline fixture results demonstrate mechanics, not real-market performance. Historical tests can still contain selection, survivorship, look-ahead, overfitting, corporate-action, liquidity, tax, and market-microstructure limitations. A visually strong backtest or trainer score does not establish future profitability or suitability.
Discovery does not execute orders, take custody of assets, continuously monitor portfolios, or consider a user’s complete circumstances. Material facts require verification with current primary sources. Regulatory treatment depends on actual product features and conduct, audience, jurisdiction, and applicable law.
Decision boundary. No personalized investment or trade advice, price target, expected return, trade execution, allocation, or position-size output. No claim promises accuracy, suitability, profit, or protection from loss.