Signals & Trends Estimate revisions, PE momentum, theme health, peer rankings, and consistency checks

Signals are intermediate computations stored in a unified signal store with TTL-based expiration (typically 7 days). Each signal produces a 0–100 score and a categorical label. Signals feed into the four conviction dimensions as components.

Signal Store

All signals are written to ticker_signals with a valid_days TTL. Expired signals are excluded from conviction scoring. Most signals refresh on a 7-day cycle.

Estimate Revisions

Tracks how consensus NTM (next twelve month) estimates have shifted across three lookback windows. Uses annual-period estimates only.

Lookback Windows

Window	Lookback	Source
7-day	today − 7d	Nearest `estimate_snapshots` row on or before cutoff
30-day	today − 30d	Same
90-day	today − 90d	Same

Delta Calculation

rev_pct = (current_estimate − old_estimate) / old_estimate × 100
Computed separately for EPS and Revenue in each window

Composite & Score

composite = average(eps_pct_30d, rev_pct_30d) — whichever are available
score = max(0, min(100, 50 + composite × 10))
+5% revision → 100, 0% → 50, −5% → 0

Label	Condition
Rising	composite > +1%
Stable	−1% to +1%
Falling	composite < −1%

Feeds into Fundamental → Estimate Trajectory component.

PE Momentum

Measures the velocity and acceleration of P/E ratio changes to detect valuation compression or expansion trends. Uses COALESCE(forward_pe, trailing_pe) from PE snapshots at four time points.

Velocity (First Derivative)

velocity_30d = (PE_now − PE_30d) / PE_30d × 100
velocity_90d = (PE_now − PE_90d) / PE_90d × 100

Acceleration (Second Derivative)

velocity_prior = (PE_30d − PE_60d) / PE_60d × 100
acceleration = velocity_30d − velocity_prior

Scoring

Base: vel_clamped = clamp(velocity_30d, −30, +30)
vel_score = 50 − vel_clamped × 1.5
Compression (−30%) → 95, flat → 50, expansion (+30%) → 5

Acceleration adjusts the base score based on the regime:

Regime	Velocity	Acceleration	Adjustment	Interpretation
Decelerating compression	< 0	< 0	+up to 10 pts	Bottoming signal
Accelerating expansion	> 0	> 0	−up to 10 pts	Getting frothy
Decelerating expansion	> 0	< 0	+up to 8 pts	Cooling off (mild positive)
Accelerating compression	< 0	> 0	−up to 8 pts	Bearish

Labels

Label	Condition
Compressing	velocity < −3% and accel > 0
Bottoming	velocity < −3% and accel ≤ 0
Expanding	velocity > +3% and accel > 0
Peaking	velocity > +3% and accel ≤ 0
Stable	−3% to +3%

Feeds into Valuation → PE Momentum component.

Theme Momentum

Classifies each theme detected for a ticker by its lifecycle stage, then aggregates to produce a health score reflecting the balance of accelerating vs decaying themes.

Theme Classification

Each (ticker, theme) pair is evaluated in priority order:

Label	Condition	Meaning
Decaying	Last seen > 30 days ago	Theme fading from narrative
New	First seen ≤ 14 days ago	Emerging theme
Accelerating	≥2 sources OR ≥3 total occurrences	Multi-source reinforcement
Stable	Default	Mentioned but not intensifying

Health Score

accel_ratio = accelerating_count / total_themes × 100
decay_ratio = decaying_count / total_themes × 100
momentum_health = max(0, min(100, 50 + accel_ratio − decay_ratio))
All accelerating → 100, all decaying → 0, balanced → ~50

Feeds into Thematic → Theme Health component.

Rating Momentum

Tracks net analyst upgrades and downgrades from rating_changes across 30-day and 90-day windows.

Computation

net_30 = upgrades_30d − downgrades_30d
net_90 = upgrades_90d − downgrades_90d
bullish_pct_90d = upgrades_90d / total_90d × 100

Labels

Label	Conditions (any trigger)
Bullish	net_30 ≥ 2, OR net_90 ≥ 3, OR upgrades > 0 with zero downgrades
Bearish	net_30 ≤ −2, OR net_90 ≤ −3, OR downgrades > 0 with zero upgrades
Neutral	Otherwise

Scoring

score = max(0, min(100, 50 + net_90d × 5))
+10 net upgrades → 100, 0 → 50, −10 → 0

Feeds into Catalyst → Rating Momentum component.

Peer Rankings

Ranks companies within their peer group across five weighted dimensions. Each dimension is a percentile rank (0–100) computed among peers with ≥3 members.

Dimension	Weight	Raw Metric	Direction
Earnings Quality	30%	`blended_beat * 0.5 + min(blended_surprise, 15) * (100/15) * 0.5` beat = eps_beat_rate * 0.6 + rev_beat_rate * 0.4	Higher = better
AI Exposure	25%	AI composite score (prefers composite > revenue_catalyst > revenue_exposure)	Higher = better
Valuation	20%	Forward P/E	Lower = better
Estimate Momentum	15%	`50 + (eps_pct_30d * 0.6 + rev_pct_30d * 0.4) * 5`	Higher = better
Price Momentum	10%	`ytd_pct * 0.4 + qtd_pct * 0.4 + rating_net * 2 * 0.2`	Higher = better

composite = Σ(weight_i × percentile_rank_i) / Σ(available weights)
Missing dimensions excluded, remaining weights renormalized

Alert

A peer_rank_bottom alert fires when a ticker's composite falls below the 25th percentile of its peer group.

Feeds into Thematic → Peer Standing component.

Thesis Consistency

Starts at 100 and deducts penalties for contradictions between thesis conviction and observed signals. Ensures high-conviction calls are supported by data.

Check	Trigger	Penalty	Severity
Estimate contradiction	Conviction ≥ 7 AND estimates falling	−25	High
Price below bear case	Price < bear_EPS × bear_PE × 0.9	−20	High
Rating divergence	Conviction ≥ 7 AND rating bearish	−15	Medium
Price above bull case	Price > bull_EPS × bull_PE × 1.1	−15	Medium
Peer rank divergence	Conviction ≥ 7 AND peer rank < 30th pct	−15	Medium
Stale thesis opportunity	Conviction ≤ 4 AND estimates rising	−10	Medium
Rating opportunity	Conviction ≤ 4 AND rating bullish	−10	Medium
Thesis staleness	Thesis age > 30 days	−1 per 15d, max −15	Low (<60d) / Medium (≥60d)

Labels

Label	Score
Healthy	≥80
Warning	≥50
Critical	<50

Feeds into Catalyst → Thesis Consistency component. Consistency flags also modify the thesis conviction score via cross-signal penalties (see Conviction Score).

AI Theme Concentration

Measures what fraction of a company's detected themes are AI-related, as an indicator of narrative concentration risk or AI exposure depth.

concentration = ai_theme_count / total_theme_count × 100

Label	Threshold	Interpretation
High	>30%	AI-dominant narrative; deep exposure but concentration risk
Moderate	15–30%	Meaningful AI presence balanced with other themes
Low	≤15%	AI is peripheral to the investment narrative

Concentration Risk

High AI theme concentration triggers a theme_concentration risk flag (medium severity) in the risk module. This doesn't directly penalize conviction but surfaces in risk reports and alerts.

Used in AI deep-dive reports and risk monitoring. Indirectly influences the Thematic dimension through theme health calculations.