Scoring model

Every weight, every metric, and the assumptions behind the ranking

How a score is built

Each instrument is measured on 29 raw metrics computed from its own price history and filings.
Every continuous metric is converted to a percentile rank within its own market — not scored against an absolute threshold. A P/E of 30 means different things on the S&P 500 and the KLCI, and any fixed cut-off that suits one market empties or floods the other.
Boolean signals score 0 or 100. RSI is banded rather than ranked, so an overbought 95 scores below a healthy 60 instead of winning.
Metrics roll up into six dimensions, then into one weighted total. A metric with no data drops out and its weight is redistributed, rather than counting as zero.

What this does not tell you

The weights are unvalidated. They are initial research values. No backtest has been run against them — that is Phase 3. Until then the ranking is a shortlist for research, not evidence that the model works.
A percentile is relative. In a uniformly expensive, uniformly deteriorating market the top-ranked name still scores near 100. High means “best available here”, never “good”.
No position sizing, stops or execution. This is Phases 1–2 only: data infrastructure and screening.

Dimensions and weights

Fundamental quality

20% of total

Growth, profitability, leverage and valuation from the most recent trailing-twelve-month figures the market had seen as of the run date.

Metric	Direction	Weight in dimension
Revenue growth YoY	Higher is better	10%
Earnings growth YoY	Higher is better	10%
EPS growth YoY	Higher is better	10%
Return on equity	Higher is better	15%
Net margin	Higher is better	10%
Free cash flow margin	Higher is better	10%
Debt to equity	Lower is better	15%
P/E	Lower is better	10%
P/B	Lower is better	10%

Technical trend

25% of total

Where price sits against its own moving averages, and whether the shorter-term structure confirms the longer-term one.

Metric	Direction	Weight in dimension
Price above 50-day MA	Yes / no	15%
Price above 200-day MA	Yes / no	20%
50-day above 200-day	Yes / no	20%
MACD above signal	Yes / no	15%
RSI (14)	Best between 45 and 70	15%
Position in 20-day range	Higher is better	15%

Momentum & structure

25% of total

Trailing performance across three horizons, weighted toward the medium term, plus how steadily that performance was delivered.

Metric	Direction	Weight in dimension
20-day return	Higher is better	15%
60-day return	Higher is better	25%
120-day return	Higher is better	25%
Trend consistency	Higher is better	20%
Near 52-week high	Higher is better	15%

Relative strength

15% of total

Performance against the market benchmark. The one dimension that still discriminates when every absolute return is negative.

Metric	Direction	Weight in dimension
Excess return vs index, 60d	Higher is better	40%
Excess return vs index, 120d	Higher is better	35%
RS line slope, 60d	Higher is better	25%

Volume confirmation

10% of total

Whether the move is backed by participation, and whether the name is liquid enough to actually trade.

Metric	Direction	Weight in dimension
Volume expansion (20d vs 60d)	Higher is better	50%
Avg daily turnover	Higher is better	50%

Risk profile

5% of total

Volatility, historic drawdown and gap exposure. Every metric here is inverted: a calmer instrument scores higher.

Metric	Direction	Weight in dimension
Annualised volatility	Lower is better	30%
Max drawdown, 1y	Lower is better	30%
Largest overnight gap	Lower is better	20%
ATR as % of price	Lower is better	20%

Gates applied before scoring

Gate	Rule	Why
Minimum history	200 daily bars	The 200-day moving average is the longest window the model reads.
Liquidity, US	5,000,000 USD/day	Average daily traded value over 20 sessions. A name that cannot be entered and exited at size is not an opportunity, and leaving it in distorts every percentile around it. The floors differ because Bursa is an order of magnitude thinner — one shared floor would empty it.
Liquidity, Bursa	500,000 MYR/day

Weights are stored in trading.weight_profiles and read at run time, so changing how the model ranks needs no deploy and every past run stays reproducible against the weights it actually used.