Proform metrics and chatgpt

Sol Ski · Feb 7, 2026

South 8.30pm hcp £2 10/1 coral (in shop)

Value = Model % – Market %

Horse	Skiena %	Market %	Edge
Street Life	17.6	7.7	+9.9%
Neptune Legend	12.9	2.4	+10.5%
Kitaab	11.4	2.9	+8.5%
Birkenhead	12.6	8.3	+4.3%
Popular Dream	10.6	12.5	-1.9%
Brazilian Belle	12.2	15.4	-3.2%
Ay Gee Ell	13.8	23.1	-9.3%
True Promise	17.5	40.0	-22.5%
Dapperling	3.4	2.9	+0.5%

Sol Ski · Feb 7, 2026

Brazilian belle won at 9/1.
So the modal was right about the Fav but was way out with the winner.

Sol Ski · Wednesday at 10:21 PM

12/02/26.
Ling 16.15 Hcp. Hitched to win at 9/2.

FINAL DATA-LED RANKING (Lingfield 16:15)

Hitched

Why he’s on top

Highest SCORE (1542)
One of the best efficiency profiles
Strong Lingfield (40%) + surface (36%)
Distance fine, field acceptable

The only horse where ratings, weight, and conditions all agree.

Foreseen

Why he’s second

Best score-per-lb efficiency
Lower OR than ability suggests
Excellent course (33%) + 1m (33%)
Lighter weight than main rivals

Field/Class samples thin, but handicap angle is strong.

Tribal Wisdom

Why he’s third

Competitive SCORE for OR
Good class (27%), distance (27%), field (28%)
Weight manageable

Weak surface record stops him challenging the top two.

gerry · Thursday at 12:55 AM

interesting its one of the races i have worked on my self tomorrow. but class 4 races are just training for me on new style of better races good luck.

Sol Ski · Thursday at 5:06 PM

1st Borgi 8/1
2nd miletus 3/1 fav
3rd Foreseenn 6/1

Sol Ski · Thursday at 5:35 PM

Chelmsford 18.oo Hcp.

Horse	Model Win %	Fair Odds	Market Odds	Market Implied %	Overlay?
Justcallmepete	24%	4.17 (≈ 13/4)	11/4	26.7%	Slightly Short
Dyrholaey	22%	4.55 (≈ 7/2)	11/4	26.7%	Overbet
Aberama Gold	20%	5.00 (4/1)	3/1	25.0%	Short
Gallant	18%	5.56 (≈ 9/2)	9/2	18.2%	✔ Fair
Coachello	10%	10.0 (9/1)	7/1	12.5%	Slightly Short
Thurso	6%	16.7 (≈ 16/1)	20/1	4.8%	✔ Small Value

Final Speed Hierarchy (All 6 Runners)

Horse	Peak	Repeatable
Justcallmepete	70	66–68
Dyrholaey	68	65–68
Gallant	68	63–66
Aberama Gold	66	64–66
Coachello	66	62–66
Thurso	67	62–64

Sol Ski · Thursday at 6:14 PM

Model vs Reality

Horse	Model Win %	Finish
Justcallmepete	24%	4th
Dyrholaey	22%
Aberama Gold	20%	3rd
Gallant	18%
Coachello	10%	6th
Thurso	6%	5th

Sol Ski · Thursday at 7:07 PM

Chelm 19.30 3yo Hcp.
This version will include:

Projected SR (with regression)
Variance (repeatability)
Pace probability weighting
Chelmsford 7f tactical bias
Field size compression (5-runner effect)
Surface transfer penalty (Wolv → Chelm)
Run-style suitability adjustment

Step — Weighted Final Model

Apply 60% steady / 40% honest weighting:

Horse	Final Win %
Crown The Future	45%
Crimson Rambler	28%
Oh So Perfect	17%
Mon Petit Frere	6%
Savvi Row	4%

Sol Ski · Thursday at 8:08 PM

Was The Model Wrong?

No — but refinement needed.

The model:

✔ Correctly opposed the favourite
✔ Identified the value horse
✔ Quantified mispricing

It slightly overestimated Crown’s conversion rate.

The Adjustment I’d Make

Increase weighting on:

Lowest OR horse with peak speed ability
In small fields where absolute ability matters more than grind

Basically:
In 5-runner C6,
raw ceiling sometimes beats profile reliability.

Sol Ski · 2026-02-14T13:51:53+0000

Ling 15.10 Hcp Back to win Heathcliff 4.42, Best rate 5.15.

MODEL vs MARKET (COMBINED)

Horse	Win %	Fair Odds	Market Odds	Edge	Verdict
Heathcliff	22.6%	4.42	5.50	1.24	VALUE
Best Rate	19.4%	5.15	6.50	1.26	VALUE
Wiltshire	16.8%	5.95	5.00	0.84	No value
Benevento	15.1%	6.62	2.88	0.44	False fav
Legal Reform	11.2%	8.93	8.00	0.90	No value
Tiger Crusade	9.1%	11.0	8.00	0.73	Underlay
Willem Twee	5.8%	17.2	17.0	0.99	Fair

Sol Ski · 2026-02-14T14:46:51+0000

Ling 15.45 Hcp.

ALGORITHM VERDICT (FINAL)

BET: UNCLE NADER – WIN @ 3/1
OPPOSE: WHO’S ON FIRST – FALSE FAV

No saver
No each-way
No second pick

This race triggers the algorithm cleanly.

Sol Ski · 2026-02-14T16:39:22+0000

15.10 heathcliff won
15.45 uncle nader bombed.

Sol Ski · 2026-02-14T18:30:46+0000

Newc 18.55 Hcp Value bet Penny mountin 11/1 sky bet.

CONTEXT (IMPORTANT)

In 4-runner handicaps, favourites are very often underlays
Penny Mountain only needs to win ~1 in 8 to justify the bet
Model says ~1 in 7.5

That’s exactly the profile you want.

Sol Ski · 2026-02-14T18:52:36+0000

Newc 19.25 Hcp Back Popty Ping win 9/2 sy bet.

MODEL vs MARKET (COMBINED TABLE)

Horse	Win %	Fair Odds	Market Odds	Edge	Verdict
Popty Ping	22.4%	4.46	5.50	1.23	VALUE
Silver State	20.8%	4.81	2.38	0.49	False fav
Archangel Josepi	17.9%	5.59	3.00	0.54	Underlay
Fickle McSelfish	11.3%	8.85	19.0	2.15	STRONG VALUE
Crown Inn To Win	10.4%	9.62	13.0	1.35	VALUE
The Caddy Master	8.4%	11.9	13.0	1.09	No bet
Riddikulus	5.2%	19.2	29.0	1.51	Value (thin)
Charcon	3.6%	27.8	29.0	1.04	No bet

Sol Ski · 2026-02-14T19:53:25+0000

Siver state won by 5.5 L. Popty ping was second.
5 pts out lay. 4.42 pts return so down 0.58 points on the day.

Sol Ski · 2026-02-14T21:03:11+0000

This is an over view of how the algorithm works.

The algorithm is trying to answer one question only:

“If this race were run many times under similar conditions, how often would each horse win?”

It does that by:

estimating each horse’s true ability from Card 2 data,
converting those abilities into win probabilities, and
comparing those probabilities to the market’s implied probabilities to find mispriced horses.

Everything else is deliberately ignored.

Step-by-step: what actually happens

Start with ability, not opinion

From Card 2, the algorithm uses:

OR (official ability anchor)
TS (what the horse has actually run to)
Score (recent performance proxy)
Age (regression / improvement risk)
Weight (performance tax)

These are treated as inputs, not conclusions.

No odds.
No “WIN%”.
No market influence.

Detect regression vs improvement

A key idea:

If TS > OR → horse may be ahead of its mark
If TS < OR → horse may be overrated / regressing

The algorithm penalises horses that need to improve just to justify their rating, and slightly rewards horses already running to or above it.

This is where many favourites quietly fall apart.

Build a performance score

All inputs are normalised and combined into a single internal score per horse.

Important:

No horse is assumed “special”
Scores are relative within the race
Tight races stay tight; weak races stay weak

This prevents runaway favourites in handicaps.

Bring in the market (only now)

Only at this point do we look at odds.

For each horse:

Market odds → implied probability
Model odds → true probability

Then we calculate:

Edge = Market Odds ÷ Fair Odds

This tells us how wrong the market is, not who will win.

Proform metrics and chatgpt

Yearling

Value = Model % – Market %​

Yearling

Yearling

FINAL DATA-LED RANKING (Lingfield 16:15)​

Hitched​

Foreseen​

Tribal Wisdom​

Mare

Yearling

Yearling

Final Speed Hierarchy (All 6 Runners)​

Yearling

Model vs Reality​

Yearling

Step — Weighted Final Model​

Yearling

Was The Model Wrong?​

The Adjustment I’d Make​

Yearling

MODEL vs MARKET (COMBINED)​

Yearling

ALGORITHM VERDICT (FINAL)​

Yearling

Yearling

CONTEXT (IMPORTANT)​

Yearling

MODEL vs MARKET (COMBINED TABLE)​

Yearling

Yearling

Step-by-step: what actually happens​

Start with ability, not opinion​

Detect regression vs improvement​

Build a performance score​

Bring in the market (only now)​

Value = Model % – Market %

FINAL DATA-LED RANKING (Lingfield 16:15)

Hitched

Foreseen

Tribal Wisdom

Final Speed Hierarchy (All 6 Runners)

Model vs Reality

Step — Weighted Final Model

Was The Model Wrong?

The Adjustment I’d Make

MODEL vs MARKET (COMBINED)

ALGORITHM VERDICT (FINAL)

CONTEXT (IMPORTANT)

MODEL vs MARKET (COMBINED TABLE)

Step-by-step: what actually happens

Start with ability, not opinion

Detect regression vs improvement

Build a performance score

Bring in the market (only now)