Skip to content
orbyd

How it works · the pipeline

From market noise to a defensible thesis, every trading day.

orbyd is a frontier language model reading the US tape on a fixed schedule. Five stages turn 30 days of news and four quarters of earnings into a per-ticker dossier you can audit — the reasoning, not the trade. Here's how a name becomes a thesis.

  1. 01 Ingest
  2. 02 Score
  3. 03 Assess
  4. 04 Compose
  5. 05 Learn

  1. Liquidity screen

    Deterministic Internal step

    Cut a ~400-name US universe to the tradable few before a model is ever called.

    A cheap, rules-only gate. No LLM touches a name that can't be traded cleanly — roughly three-quarters of the field is gone before any model runs.

    Reads
    Spread · ADV · market cap · tradable shape
    Emits
    ~25% of the universe survives

  2. Momentum + narrative scoring

    Deterministic + Sonnet Internal step

    Rank survivors by structure × volume × news density × theme-cluster strength.

    Themes are first-class. The system reads basket behaviour — how a name moves with its cohort — not isolated ticker action.

    Reads
    Price structure · volume · 30d news density · theme baskets
    Emits
    Ranked candidate shortlist

  3. Strategic-grade assessment

    Claude Opus Published

    Read everything on each candidate and write the dossier you see on this site.

    Opus synthesises thesis, invalidation trigger, bull case, bear case, setup, catalysts, and correlation notes — and grades the name FORTRESS / STRATEGIC / MARGINAL / NONE. This stage is the public product.

    Reads
    30d news · 4Q earnings transcripts · filings
    Emits
    A dossier per ticker + a strategic grade

  4. Portfolio composition

    Claude Opus · 1M context Stays private

    Reason across every candidate side-by-side in a single pass, under hard constraints.

    This is the stage a 1M-token window makes possible: the whole candidate set compared in one mind, not stitched from hundreds of isolated calls. The reasoning informs the public read; the positions stay private by design.

    Reads
    All dossiers at once · sizing caps · archetype + regime rules
    Emits
    A target book — deliberately not published

  5. Postmortem + adaptive optimiser

    Claude Opus Lessons published

    Grade the model's own calls and update how it weighs evidence.

    Recurring patterns promote to a playbook; rule weights update weekly from realised outcomes. The lessons surface in the methodology — the portfolio mechanics behind them do not.

    Reads
    Closed-trade outcomes · 90-day rolling window
    Emits
    Playbook entries + Bayesian-updated rule weights

Why a 1M-token window changes the work

The whole book reasoned in one mind — not stitched from hundreds of calls.

Most automated research scores names in isolation and bolts the results together. Stage 04 doesn't. Claude Opus's million-token context lets the composition pass hold every candidate's full dossier at once — comparing theses, conflicts, correlations, and concentration across the entire set in a single reasoning pass. That's how a name gets sized against its basket and the regime, not judged on its own merits alone. The difference between a spreadsheet of scores and an analyst who has read every name in the field.

Public — the read

  • Thesis, invalidation trigger, bull & bear case
  • Setup, catalyst calendar, correlation notes
  • Archetype, conviction, theme & regime calls
  • Every read dated and versioned

Private — the trade

  • Entry prices, share counts, stops, targets
  • Portfolio value, returns, P&L
  • The composed target book (Stage 04)
  • Publishing live actions would let readers front-run or fade them — and break a clean forward-test.

Common questions

How does an AI language model analyse stocks?
orbyd runs a five-stage pipeline every trading day. A deterministic screen cuts a ~400-name universe to the tradable few; momentum and narrative scoring rank the survivors; then Claude Opus reads 30 days of news, four quarters of earnings transcripts and filings per candidate and writes a structured dossier — thesis, invalidation trigger, bull case and bear case. A 1M-token context lets it compare the whole candidate set in a single reasoning pass.
Which AI models does orbyd use?
Anthropic's Claude Opus and Claude Sonnet. Opus handles the deep per-name synthesis and the 1M-token portfolio-composition pass; Sonnet handles the faster scoring passes.
What does a 1M-token context window actually do for stock research?
It lets the composition stage hold every candidate's full dossier at once and compare theses, conflicts, correlations and concentration across the entire set in one reasoning pass — rather than scoring names in isolation and stitching the results together afterwards.
Does orbyd trade in real time?
No. The pipeline runs on a fixed New-York-time schedule (premarket scan, a pre-close decision window, midday and opportunistic checks, and a post-close reconcile). It publishes research, not live orders, and runs on a paper account.
Is orbyd's research automated or human-written?
Fully automated. The dossiers, regime calls and macro views are written by frontier language models; the methodology is open and every read is dated. Humans don't edit the model's output.

Today the pipeline tracks 431 names across 66 themes. Go deeper:

Anatomy of a dossier See the dossiers Full methodology For LLM agents