The conviction score is the primary decision signal. It blends earnings quality, thematic positioning, valuation attractiveness, and catalyst credibility into a single number with a buy/hold/sell label.
| Dimension | Default Weight | Focus |
|---|---|---|
| Fundamental | 25% | Earnings quality, estimate direction, financial health |
| Thematic | 25% | AI resilience, theme momentum, peer standing |
| Valuation | 25% | Relative P/E, leverage ratios, PE momentum |
| Catalyst | 25% | Thesis conviction, analyst momentum, insider activity, institutional momentum, consistency |
Measures earnings quality, estimate direction, and overall financial health. Simple average of available components.
| Component | Formula | Scale |
|---|---|---|
| Beat Rate | EPS beats in last 4 quarters / 4 * 100 | 0 = never beat, 100 = always beat |
| Surprise Quality | 50 + avg_surprise * 5 |
+10% avg surprise = 100, 0% = 50, -10% = 0 |
| Sentiment | (avg_sentiment + 1) * 50 |
Score -1..+1 mapped to 0..100 |
| Estimate Trajectory | 50 + revision_pct * 10 |
+5% NTM revision = 100, -5% = 0 |
| Guidance | Most recent guidance direction | Raised = 80, Maintained = 50, Lowered = 20 |
| Financial Strength | Composite signal (see Financial Strength) | 0–100 from snapshot + trends |
Measures structural positioning: AI exposure, theme health, and competitive standing. Uses adaptive weighting — AI resilience weight scales with signal strength so it doesn't dominate for companies where AI impact is ambiguous.
| Component | Source | Weight | Details |
|---|---|---|---|
| AI Resilience | AI Resilience composite | Adaptive0.3 + 0.7 × |score−50|/50Floor 0.3, ceiling 1.0 |
5-dimension LLM-scored composite (0–100). Weight self-adjusts: neutral scores (~50) are dampened; extreme scores (fortress/vulnerable) carry full weight. |
| Theme Health | theme_momentum table | 1.0 (fixed) | 50 + accel_ratio - decay_ratioMeasures % of themes accelerating vs decaying |
| Peer Standing | peer_rankings table | 1.0 (fixed) | Percentile composite across earnings quality, AI exposure, valuation, momentum |
Without dampening, AI resilience dominates thematic for the majority of tickers (especially those missing theme momentum or peer data). A uranium miner scored 87 on AI resilience (data center power demand) — that signal is real and should carry weight. A defense contractor at 50 is neither helped nor hurt by AI — its thematic score should be driven by theme momentum and peer standing, not a neutral AI score.
Lower valuations = higher scores. Uses TTM (trailing twelve month) aggregates for flow-based metrics and sector-aware leverage thresholds.
| Component | Formula | Thresholds |
|---|---|---|
| PE Percentile | % of peer group with higher forward P/E | Cheapest in group = 100 |
| PEG Ratio | (3 - PEG) / 2 * 100 |
PEG 1 = 100, PEG 3 = 0 |
| Price Strength | price / 52w_high * 100 |
Near highs = high momentum |
| Net Debt / TTM OCF | (cap - ratio) / cap * 100 |
Utilities/Energy/Industrials: cap=15x Others: cap=10x |
| EV / TTM FCF | (cap - ratio) / (cap - 15) * 100 |
Utilities/Energy/Industrials: cap=80x Others: cap=60x |
| PE Momentum | Velocity + acceleration from PE snapshots | Compressing PE = high score, expanding = low. See Signals |
Capital-intensive sectors (Utilities, Energy, Industrials) get wider thresholds for ND/OCF and EV/FCF because structural leverage is normal in those businesses. A utility at 9x ND/OCF isn't distressed—it's a regulated asset base financed at utility-grade rates.
Measures forward-looking conviction quality, analyst sentiment, ownership signals, and thesis coherence. Simple average of available components.
| Component | Formula / Source | Details |
|---|---|---|
| Thesis Conviction | Calibrated + penalty-adjusted (see below) | Raw LLM 0–10 score stretched via piecewise scaling |
| Rating Momentum | 50 + net_upgrades * 5 |
+10 net upgrades = 100, -10 = 0 |
| Estimate Momentum | EPS/revenue revision trajectory | From 30d revision composite |
| Guidance Trajectory | Management guidance direction | Raised = 80, Maintained = 50, Lowered = 20 |
| Thesis Consistency | 100 minus penalties | See Signals → Thesis Consistency |
| Insider Activity | Insider signal (0–100) | SEC Form 4: C-suite/director purchases vs baseline selling. Cluster buys boost score; elevated selling penalizes. |
| Institutional Momentum | Institutional signal (0–100) | WhaleWisdom 13F: net new positions, increased/decreased ratio, net share change QoQ. |
| Contrarian Signal | Technical oversold + fundamental divergence | Fires when price action conflicts with improving fundamentals |
The raw LLM conviction score (0–10) is stretched and penalty-adjusted:
When an analyst sets analyst_dcf_mode = 'override' on a thesis, the system treats the analyst's DCF-derived conviction as the primary signal rather than blending with the LLM thesis conviction. In 'signal' mode (default), the analyst DCF informs the thesis prompt as additional context but doesn't override.
Applies only to software-adjacent companies where either revenue_catalyst or pricing_resilience < 70:
| Tier | Trigger | Effect |
|---|---|---|
| Existential | Both AI dims < 50 | Up to 20pt penalty; hard cap at 55 unless growth evidence is strong |
| Prove It | Either AI dim 50–69 | Up to 12pt penalty scaled by vulnerability and growth proof deficit |
| Label | Score Range |
|---|---|
| Strong Buy | ≥80 |
| Buy | ≥65 |
| Hold | ≥50 |
| Underweight | ≥35 |
| Sell | <35 |
Conviction scores also map to portfolio construction tiers:
| Tier | Conviction | Role |
|---|---|---|
| Anchor | ≥70 | Core holdings, largest position sizes |
| Ballast | 60–69 | Stable positions, moderate sizing |
| Contrarian | 50–59 | Thesis-dependent, smaller sizing |
| High-Risk | <50 | Watch/trim candidates |