01How we score ourselves
Five sessions. One public outcome.
Every published call is written before the fact and resolved against market prices after the fact. The call, the model version and the timestamp are fixed before the outcome is known.
No backfill · no silent edits · public resolution
02Why this stack
Every sleeve is named, versioned and replaceable.
The 2026 open-source ranking for finance is clear. Evidence converges; no single sleeve owns the call.
- FinBERT — live
- 86–97% on Financial PhraseBank, blended with FinLex.
- World Monitor equity layer — live
- CII across 31 states, chokepoints and prediction markets. A 10% sleeve.
- FinGPT v3 — reserved
- LoRA-tuned LLM, needs a GPU.
- FinLex — always on
- Loughran–McDonald with VADER valence; the fallback when FinBERT is cold.
- TMX Global events — live
- Earnings, dividends, splits. A print inside the window is flagged, never rescored.
- Options shadow · options_v1
- Stored beside the ticket; never writes methodology_v1.
Evidence converges; no single sleeve owns the call.
035-day path
A CNN + LSTM shape, kept without a GPU.
The production hybrid keeps the shape of the original research on roughly 100 technical and news features.
- Weights
- 30% technical · 25% news · 14% sequence · 10% persistence · 11% direction tree · 10% geo, then ATR-scaled.
- Geo tilts
- Hormuz stress supports tankers and drags carriers; China–Taiwan stress drags the semi stack.
Input → thesis → call → five sessions → resolved
04Crypto book
crypto_v1 — a separate desk under the same rules of transparency.
Five calendar days, 24/7. It does not write methodology_v1.
- BUY / SELL
- Five-day USD return at or beyond ±3%.
- Dollar pegs
- USDT, USDC and DAI are always HOLD.
- Concentration
- Top-10 holders above 75% block a BUY.
Separate book · separate rules · same transparency
05Five Sessions (Studio)
A human rundown of the published top ten.
Live on Mondays at 09:00 GST from DIFC, public after the five sessions with a forensic post-mortem. The show reads the book — it does not write it.
Human explanation sits beside the public book.
06Signal-to-instrument map
Every World Monitor family is wired to the names it most plausibly leads.
A forecast that names its drivers is auditable.
- Chokepoint stress
- BNO and refiners; airlines inverse.
- GPS / jamming
- LMT, RTX, NOC, GLD.
- Cyber IOC language
- CRWD, PANW, FTNT.
Event → exposure → instrument → thesis tilt
07Buy, hold, sell
Three labels, versioned decision logic.
BUY — expected five-day return above +1.65% and sentiment not clearly negative. SELL — below −1.65% and sentiment not clearly positive. HOLD — everything else.
The label is concise; the evidence stays open.
08Scoreboard
Written before the fact, resolved after it.
The 1.0% dead band exists so HOLD is a real call, not a free win. Fewer than 10 resolved predictions shows “insufficient history”.
- BUY is correct
- When the return is at least +1.0%.
- SELL is correct
- When the return is at most −1.0%.
- HOLD is correct
- When the absolute return is under 1.0%.
Published call + fixed rule = public result
09What we claim — and refuse to claim
A real record, not predictive certainty.
Watstock claims exactly one thing: that its record is real. No accuracy marketing, no guaranteed returns.
Research and media — not financial advice.
10Free media
The book is free; the record is the product.
The ranking, the five-session scores, Studio and the public scoreboard are free. Sign in only for a watchlist or the paper simulator.
Open proof before depth.
11Paper book
Practice the decision without brokerage.
Each book starts with $100,000 of virtual cash. A mock fill uses the last session close.
Paper ticket ≠ brokerage ticket
12Honest limits
Every layer has failure modes.
Daily bars, not a streaming tape. The five-day path is a research ensemble, not a promise of alpha. Use the table as information — then do your own work.
Confidence is bounded. Invalidation is required.