ClaritX forward-tested AI stock-selection record

An append-only public archive of the forward test published at claritx.ai/research.

Pre-registered AI stock-selection strategies, measured forward — never backfilled — against a total-return S&P 500. The rules for each arm were frozen before its first measured day. Losing arms are published alongside winning ones; a record that keeps only its good months is not a record.

Why this repository exists

The numbers live on a website that rebuilds every night, so a figure quoted last month is not the figure on the page today. This repository fixes that. Twice a day a workflow copies the published files here and commits them if anything changed. That gives every snapshot:

• a timestamp and a hash that the publisher cannot rewrite quietly — a restated number shows up as a visible diff in this history, not as a silent edit;
• a stable address to cite — link a commit, or a monthly edition (edition-YYYY-MM releases), rather than a page that will change.

Files

File	One row per
data/claritx-forward-test-daily.csv	strategy arm per trading day — the arm's index and the total-return S&P 500, both rebased to 1.0 at the arm's inception
data/claritx-forward-test-arms.csv	strategy arm — inception, total return, S&P 500 over the same window, alpha, information ratio, Sharpe, volatility, max drawdown
data/claritx-score-bands.csv	AI score band — every completed analysis scored 0–100 at publication, measured forward
data/claritx-cohort-history.csv	cohort per nightly snapshot — the same measurement re-run and archived each night
data/claritx-evidence.json	the manifest: snapshot date, temporal coverage, row counts, citation

Returns and alpha are decimals (0.0916 = 9.16%). Alpha is the arm's return minus the S&P 500 total return over the same window.

Reading it honestly

• It is a forward test of a few months, not a multi-year backtest. Short windows are dominated by the regime they happened to cover.
• Arms are paper portfolios, not traded accounts.
• The score-band table is the one to read first if the question is "does a higher AI score mean a better stock?" — and its history in cohort-history.csv shows how that answer has changed night to night.

Methodology and the live charts: https://www.claritx.ai/track-record · https://www.claritx.ai/ai-rank-report

Known gaps

Gaps stay in the history; they are documented here rather than rewritten.

• 2026-09-24 (commit bc0c3c4): the nightly run at the source skipped every strategy arm after a database slowdown. The site published a manifest listing only the score-band and cohort-history files, and this archive committed it. The daily and per-arm files in that commit are unchanged from 2026-09-23 — they were not re-published that night. Since then the archive refuses any snapshot that stops listing a file it already holds, and the source serves the last complete snapshot instead of an empty one.

How to cite

ClaritX Research Engine, "Forward-tested AI stock-selection record", snapshot <snapshot_date> (commit <sha>). https://github.com/Orcules/claritx-forward-test

See CITATION.cff. Not investment advice.

License

Data: CC BY 4.0 — reuse freely, credit "ClaritX Research Engine" with a link to https://www.claritx.ai/research. Scripts: MIT. See LICENSE.