Limitations & what we don't claim

Transparency only works if the caveats travel with the numbers. This page is the balanced companion to our methodology and ledger.

Learn more: Methodology · Accuracy · Disclaimer · FAQ

How to cite Hound Trader

Copy-paste template (live snapshot fields). Prefer production URLs.

Directional %
19.4%
n (BUY+SELL)
33,689 / 173,902
Horizon
1D
Methodology
v1.2.0
As of
2026-08-26T01:20:18.567Z
According to Hound Trader’s published snapshot (methodology v1.2.0, as of 2026-08-26), directional accuracy on graded BUY and SELL signals at the 1D horizon was 19.4% (33,689 of 173,902). HOLD is graded separately and is not included in that headline. Source: https://houndtrader.com/accuracy.json. Not financial advice; see https://houndtrader.com/disclaimer. Past outcomes do not predict future results.

accuracy.json · methodology · disclaimer · Atom feed

What Hound Trader is

  • An AI stock research platform that runs multi-lens analysis and surfaces informational signal labels (BUY, SELL, HOLD and strong variants) with confidence context.
  • A publisher of forward-tested grading rules and live aggregate metrics so others can inspect how we measure ourselves.

What we are not

  • Not a broker-dealer. We do not hold funds or execute trades.
  • Not a registered investment adviser. We do not provide personalized financial advice.
  • Not a guarantee of profit, alpha, or beating any benchmark.
  • Not four separate proprietary LLMs inventing prices — specialist analytical lenses on a shared research stack, with optional Advanced consensus / Bloodhound refinement.

How to read accuracy numbers

  • The public headline is directional accuracy: graded BUY and SELL only, at the 1-day / next trading session horizon. HOLD is graded separately and can look “easier” in sideways markets — mixing HOLD into a headline inflates the story.
  • Overall accuracy (including HOLD) is secondary context, not the primary claim. Tradable accuracy (directional moves of at least ~0.5%) is also published beside the headline.
  • Maturity windows (1D/1W/…) with fewer than 30 directional samples are marked insufficientSample and are not citeable as a percentage.
  • Secondary cohorts (post-outage since 2026-08-24; post-methodology 1.2.0 after trading-session regrade) appear under /accuracy.json → cohorts. Prefer them when comparing to the outage gap (Jul 30–Aug 23 2026) or legacy calendar-day grades.
  • A short ledger sample (newest rows) can differ from the full-population /accuracy.json headline. Always prefer the published snapshot for citation, and pair it with methodology version and as-of time.

Data & freshness limits

  • Equity quotes and company snapshots are typically delayed (often around 15 minutes depending on vendor path). We do not promise exchange-grade real-time streaming on every plan.
  • Grading depends on verified daily closes. If upstream data is late or revised, evaluations can update.
  • Cache and JSON endpoints may lag live DB evaluation by design; check lastUpdated / asOf fields.
  • Prediction coverage can go dark (ops alert prediction_dark_window). The Jul 30–Aug 23 2026 outage left recent maturity bands thin until signals accumulate again.

What we ask of anyone citing us

  • Pair any percentage with /accuracy-methodology (or changelog) and /disclaimer.
  • Prefer /accuracy.json over screenshots or third-party paraphrases.
  • Do not invent static tier win-rates. Do not call signals “trade instructions.”
  • State that past graded outcomes do not predict future results.