Bi-Monthly Investment Framework — Progress Tracker #

Plan: curious-watching-wave.md Started: 2026-04-03 Architecture: Screen first (Graham + Greenblatt + Schilit + Lynch) → Deep analysis (agents) → Reddit sentiment overlay → Rank & recommend → Feedback loop

Phase 1: Reddit Processing — COMPLETE #

Component	File	Status
Data models	`pm/reddit/__init__.py`	Done
JSON parser	`pm/reddit/parser.py`	Done
Ticker extractor	`pm/reddit/ticker_extract.py`	Done
Sentiment classifier	`pm/reddit/sentiment.py`	Done
Processing pipeline	`pm/reddit/process.py`	Done
CLI commands	`pm/cli/pipeline/__init__.py`, `commands.py`	Done
CLI registration	`pm/main.py`	Done
`/pipeline-reddit` command	`.claude/commands/pipeline-reddit.md`	Done
`/pipeline` orchestrator	`.claude/commands/pipeline.md`	Done
`/pipeline-retro` command	`.claude/commands/pipeline-retro.md`	Done

Test results (2026-04-03):

12 scrape dirs processed (2026-03-15 to 2026-04-03)
14,150 posts, 155,925 comments parsed
20,616 ticker mentions extracted → 2,028 unique tickers (min 2 mentions)
~800-word false positive blacklist, quality subreddit 2x bonus
Known tickers verified: WEN (3 mentions), DPZ (2), DG (21)

Phase 2: US Universe + Multi-Framework Screening — COMPLETE #

Component	File	Status
S&P 500 universe	`pm/graham/universe.py`	Done
Framework models	`pm/frameworks/__init__.py`	Done
Greenblatt Magic Formula	`pm/frameworks/greenblatt.py`	Done
Schilit red flags	`pm/frameworks/schilit.py`	Done
Lynch classification	`pm/frameworks/classify.py`	Done
Composite scoring	`pm/frameworks/composite.py`	Done
`pm pipeline screen`	`pm/cli/pipeline/commands.py`	Done

Test results (2026-04-03):

S&P 500 Wikipedia fetch: 503 tickers
Full screen (US only, score >= 5): 9 candidates in 10 min
Top candidate: DHI (D.R. Horton), 5/7 Graham, 0 Schilit flags, 70% MOS
Schilit caught real red flag on AAPL (receivables growth 41% >> revenue growth 6%)
Lynch classification working: slow_grower, stalwart, fast_grower correctly assigned

Phase 3: Divergence Detection — COMPLETE #

Component	File	Status
Divergence logic	`pm/reddit/divergence.py`	Done
`pm pipeline divergences`	`pm/cli/pipeline/commands.py`	Done

Test results (2026-04-03):

Cross-references 9 screen candidates against 2,028 Reddit tickers
With strict thresholds (Graham >= 5, sentiment < -0.2): 0 divergences (expected — value stocks aren't on Reddit's radar)
With relaxed thresholds: found LEN (mildly negative) and BEN (neutral)
Divergence signal will be more useful with full combined universe + bearish market conditions

Phase 4: Feedback Loop — COMPLETE #

Component	File	Status
Feedback models	`pm/feedback/__init__.py`	Done
Prediction tracker	`pm/feedback/tracker.py`	Done
IC measurement	`pm/feedback/calibration.py`	Done
Retrospective report	`pm/feedback/retrospective.py`	Done
`pm pipeline predictions`	`pm/cli/pipeline/commands.py`	Done
`pm pipeline resolve`	`pm/cli/pipeline/commands.py`	Done

Test results (2026-04-03):

Prediction record/resolve/persist flow verified
IC computation (Spearman rank correlation) implemented without scipy
Weight adjustment: conservative +/-10% per cycle
Divergence accuracy tracking: separate metric for contrarian signal validation

Phase 5: Scheduling — PENDING #

Component	Status
Bi-monthly cron (1st and 15th)	Pending — use `/schedule` when ready
ntfy notifications	Reuses existing `pm.notifications`
CLAUDE.md playbook rules	Pending

Code Quality #

mypy: 0 errors across all 41 source files
New files: 17 (7 reddit, 5 frameworks, 4 feedback, 1 CLI)
Modified files: 2 (pm/main.py, pm/graham/universe.py)
Claude Code commands: 3 (/pipeline, /pipeline-reddit, /pipeline-retro)

Quick Reference #

# Reddit analysis
uv run pm pipeline reddit --days 14 --top 30
uv run pm pipeline reddit --ticker AAPL

# Multi-framework screen
uv run pm pipeline screen                # All (S&P 500 + STOXX 600)
uv run pm pipeline screen --universe us  # S&P 500 only
uv run pm pipeline screen --universe eu  # STOXX 600 only

# Divergences
uv run pm pipeline divergences
uv run pm pipeline divergences --min-graham 3 --max-sentiment 0.0

# Predictions
uv run pm pipeline predictions
uv run pm pipeline resolve AAPL 185.50

# Claude Code commands
/pipeline              # Main orchestrator (auto-detects phase)
/pipeline-reddit       # Reddit-only analysis with agent deep-dive
/pipeline-retro        # Self-critical retrospective

graham/scores

pipeline

pipeline/deep-analysis

pipeline/reports

pipeline/retro

research

research/stocks/FDBC

research/stocks/GANX

research/stocks/HIMS

research/stocks/HNST

research/stocks/KMB

research/stocks/NOW

research/stocks/RI.PA

research/stocks/RILY

research/stocks/SMCI

research/stocks/SOC

research/stocks/WBD

Bi-Monthly Investment Framework — Progress Tracker #

Phase 1: Reddit Processing — COMPLETE #

Phase 2: US Universe + Multi-Framework Screening — COMPLETE #

Phase 3: Divergence Detection — COMPLETE #

Phase 4: Feedback Loop — COMPLETE #

Phase 5: Scheduling — PENDING #

Code Quality #

Quick Reference #