Skip to article content

Revised 23 October 2026 · v1.0.0 · what changed

Code: github.com/herrrickshaw/us-stock-picks

Watch the walkthrough

Watch on YouTube: youtu.be/-xmjl_d2xLg

Why it is useful

• Self-directed researchers and students who want to see how Darvas breakout, Piotroski, Coffee Can, Magic Formula and momentum screens can be combined into one score get a readable, auditable implementation.
• It replaces juggling separate screens: one run gives a sector-capped shortlist, a full ranked CSV and a paper-portfolio file with transparent screen hits for each name.
• Evidence from the run: a fresh scan of the 300 most liquid US stocks was ranked end to end in under three seconds: 256 passed the liquidity gate, 181 were ranked (24 CORE, 31 QUALITY, 50 MOMENTUM, 76 WATCH).
• It also shows how fragile a rank is: an earlier attempt with fundamentals missing for 162 of the 300 names produced a different top 15 (12 CORE names, not 24). Missing data silently becomes 'screen not passed', so check coverage before reading the list.

What it is for

us-stock-picks (package us_stock_picks) reads the newest us_full_scan_*.xlsx workbook, applies a liquidity gate (price at least $5, turnover at least $2 million a day), scores ten screens with fixed weights, assigns each stock to a sleeve (CORE, QUALITY, MOMENTUM, WATCH), caps names per sector, and writes HTML, CSV and JSON outputs.

The scan workbook comes from a separate pipeline that is not in this repo. For this run I built one with the Darvas, golden-cross, five fundamental screens and machine-learning stages of the author's sibling BazaarTalks scanner, fed with cached prices (to the 21 October 2026 session) for the 300 most liquid US non-ETF stocks and live fundamentals from Yahoo Finance. A small adapter added turnover, a liquidity tier and an above-EMA50 flag, and I named the file with the session date so the as-of label is right.

Not available in my workbook: the Quality score and grade, the 'Actionable' flag, so those two screens never fire. The full 7,500-stock universe was not scanned.

Who it is for

• Students learning multi-factor screening
• Researchers who want a transparent, reproducible ranking recipe
• Developers building a screening report from tabular scan data

What you need

Get the code and an AI assistant

You do not have to be a programmer. Get the repository, give it to an AI assistant, and ask in plain language. Any of these routes works:

git clone https://github.com/herrrickshaw/us-stock-picks.git
cd us-stock-picks
claude      # or: gemini

Install

1. Clone the repository

   git clone https://github.com/herrrickshaw/us-stock-picks.git
   cd us-stock-picks

2. Create an environment and install dependencies

   python3 -m venv .venv && source .venv/bin/activate
   pip install -r requirements.txt

Quick start: commands and what you should see

See all options

python -m us_stock_picks --help

Usage text listing --scan, --top, --max-per-sector, --portfolio-size, --min-price, --min-turnover, --refresh and --open.

Rank a scan workbook

python run_us_picks.py --scan path/to/us_full_scan_YYYYMMDD_HHMM.xlsx --top 15 --max-per-sector 3 --out outputs

Prints a summary (as of, universe, liquid, ranked, sleeves) and the shortlist; writes HTML, CSV, JSON and paper-portfolio files. On 23 Oct 2026: as of 2026-10-21, universe 300, liquid 256, ranked 181. Note: python -m us_stock_picks works only if the repo root is on the path.

Inspect the committed outputs without any data pull

python3 -c "import json;print(json.load(open('outputs/us_picks_latest.json'))['summary'])"

A dictionary with as_of, universe_n, liquid_n, ranked_n and sleeve_counts.

Sample outputs

Real output from a run on the date shown in the sources line; trimmed to fit.

Run summary from the 23 Oct 2026 re-run (as of 2026-10-21)

python run_us_picks.py --scan us_full_scan_20261021_0000.xlsx
as of:     2026-10-21 00:00
universe:  300
liquid:    256
ranked:    181
sleeves:   {'WATCH': 76, 'MOMENTUM': 50, 'QUALITY': 31, 'CORE': 24}

Old committed run (24 Jul 2026): universe 5,814, liquid 1,623, ranked 721, CORE 103 / QUALITY 82 / MOMENTUM 330 / WATCH 206. The universes differ by design (300 versus 5,814), so the counts are not comparable.

Shortlist format sample (illustrative mechanical output, not advice)

outputs/us_investor_picks_latest.csv
rank  ticker (exchange)  sleeve  score  screens
1     GD   (NYSE)        CORE    62.3   4
2     TT   (NYSE)        CORE    61.0   4
3     TEAM (NASDAQ)      CORE    60.7   5
4     ADSK (NASDAQ)      CORE    60.7   5
5     INTU (NASDAQ)      CORE    60.3   4
6     LLY  (NYSE)        CORE    55.7   4

Mechanical screen output on a 300-stock universe, shown only to illustrate the file format. Not investment advice and not a recommendation. Quality and Actionable screens were unavailable.

Screens passed per ranked stock (quant file, 100 rows)

value_counts on screens_passed
2 screens: 20
3 screens: 55
4 screens: 22
5 screens: 3

No stock passed six or more screens, versus six for the top names in the committed July file, which had the Quality and Actionable screens.

Config: conviction weights (us_stock_picks/config.py)

breakout 18 | piotroski_strong 16 | coffee_can 14 | magic_formula 10 | bull_cartel 10
golden_cross 10 | quality_ab 8 | actionable 6 | ml_bullish 5 | above_ema50 3
CORE needs score >= 48 and >= 2 screens

Weights are the author's choices and are not backtested in this repo.

Ways to use it

Sample prompts for your AI assistant

Copy, paste and edit the parts in [brackets].

• Explain how conviction_score is computed in us_stock_picks/engine.py and config.py, with a worked example for one row of the ranked CSV.
• Using outputs/us_quant_ranked_latest.csv, show how many names pass each screen and which pairs of screens overlap most.
• Which names in the shortlist have a BEARISH ML_Direction but a high conviction score? What does that disagreement suggest about the weights?
• Suggest a backtest design to test whether the CORE sleeve beats an index, listing the biases to avoid.
• Draft a plain-English disclaimer and data-freshness note for sharing this output with a study group.

Limits and honest caveats

Sources: the us-stock-picks README, requirements.txt, us_stock_picks/config.py and code, run on 23 October 2026 in a scratch clone and throwaway virtual environment on a scan workbook built from cached US prices (last session 21 October 2026) and Yahoo Finance fundamentals. Only derived aggregates and format samples shown.

About this article: Researched, written and edited by Umashankar Triplicane Dwarakanathan, with AI research assistance; every figure is meant to trace to the primary source cited. See the Editorial Policy for how sourcing, AI use and corrections work.