Pipeline Retrospective — 2026-04-15 #

Cycle: 2026-04-03 (12 days elapsed) Phase: 5 / Feedback

Pipeline Stats #

• Cycle started: 2026-04-03
• Phases 1–4 completed: screen → deep analysis → reddit overlay → rank & recommend
• Predictions recorded: 7 (AIG, DD, PHM, SWKS, DHI, TROW, HUM)
• Candidates count: 9
• Top candidates surfaced: DHI, AIG, DD, LEN, SWKS, TROW, HUM, PHM, BEN

Prediction Scorecard #

No predictions resolved this cycle.

All 7 predictions are 12 days old — below the 30-day maturation threshold. Earliest eligible resolution date: 2026-05-03.

IC Table #

Not computed — requires 10+ resolved predictions. Current resolved: 0.

Weight Adjustments #

None applied. Weights remain at defaults:

• graham_pct: 0.30
• magic_formula_pct: 0.20
• llm_pct: 0.30
• schilit_clean_pct: 0.10
• reddit_divergence: 0.10

Reddit Divergence Accuracy #

Not computed — all picks this cycle had reddit_divergence = 0 (no meaningfully hated-but-sound picks surfaced). This itself is a data point: the current reddit overlay is either under-weighted or the universe isn't producing contrarian setups.

Self-Critique #

What worked this cycle:

• Execution discipline: AIG T1 (rank #1) was actually executed on 2026-04-07, per shr-028 (pipeline composite > catalyst bucket allocation).
• Diversification held: picks spanned Financials, Basic Materials, Consumer Cyclical, Tech, Healthcare — no single-sector over-weight.
• Same-day red flag check (shr-020) was run before AIG execution.

What needs attention:

• Prediction pool is too small for IC. Need to accumulate 10+ resolved predictions before any weight adjustment carries statistical weight. At ~7 predictions per cycle × every 14 days, first meaningful IC computation will land around cycle 3–4 (≈2026-06).
• Reddit divergence signal is inert. All 7 predictions scored 0. Either the reddit overlay isn't intersecting with fundamentally sound names, or the mapping from reddit sentiment → divergence score isn't discriminating. Review needed before next cycle.
• DD note flagged a screen data error ("Not deep value at 17.9x forward P/E. Moat play, not Graham play."). Screen is surfacing stocks that don't fit the framework it claims — audit the graham_pct calculation logic.

Next Cycle #

• Next cycle due: 2026-04-17 (14 days from 2026-04-03)
• First IC-eligible feedback cycle: 2026-05-03 or later
• AIG Q1 earnings: 2026-04-30 — T2 decision gate (re-verify EPS consensus, CR <92%, buyback pace)

Action Items #

1. Before next cycle, audit the graham_pct screen logic (DD false positive).
2. Review reddit_divergence scoring — zero signal across 7 picks is suspicious.
3. Wait until 2026-05-03 before running feedback phase with actual resolution.