Skip to content
CURRENT
-7 → -6 (+1) over 24 captures ATL @ GB spread -7 → -6 house House backtest last 10: 6–4 · 90-day all-market 55.3% (n=8163) · 23h ago Wire Rams defeat depleted Giants 28-6 on 'Monday Night Football' Wire Rams' Stafford passes Rivers for 6th all time in passing TDs Wire Davante Adams, Matthew Stafford lead Rams past Giants 28-6 after Jaxson Dart's injury
Access: Anonymous access. Content follows.
learn

Fork a Public Model and Make It Yours

Read the price, role, and market first Every listed model is public and forkable. What a card shows, where the one-click remix lands, what a fork copies, and why its record restarts at zero.

10 sections

Model Desk

The Shark Snip desk for model coverage. Every claim ships with its sample size and its interval, or it does not ship.

Key takeaways (from article sections)

  • What a public card tells you before you touch it
  • One click, and where it lands
  • What saving actually requires
  • What a fork carries, and what it leaves behind
  • Change one thing, then train
  • The forward record restarts at zero
  • Where a fork goes next
  • The Receipts Drawer
  • FAQ
  • Where these numbers come from

Updated Sep 9, 2026 · the public model catalog.

Every model Shark Snip lists is public, and each one opens into a card you can read. The catalog states its own rule up front: a model with nothing clean behind it shows an empty record rather than a flattering one. Copying somebody else's work takes one click, and keeping the copy takes a subscription.

What a public card tells you before you touch it

Each card carries the sport, the license, a version number and a readiness line. Under that sits either a record or an honest empty state, printed plainly as no record yet and no settled history yet.

A card that has been trained shows its backtest and the size of the sample behind it. Copies get counted, so you can see how many people have already taken a run at the thing. Where variants exist they are listed out, each with its own cover rate and sample count, which tells you whether the family agrees with itself.

Version numbers matter more than they look. A card still on its first version has never been revised, and a family of variants around one name is somebody searching for a setting that sticks.

Read the sample count before the percentage. A cover rate over a hundred-odd games is a rumour, and the catalog prints both numbers side by side so the rumour is easy to spot.

One click, and where it lands

Two buttons sit on every card. View model card opens the model's own page. Remix with simple tweaks lands on the knob panel further down that same page.

It does not drop you on the full Studio board, and that is deliberate. The one-click path exists to change a few settings on somebody's working model, not to hand a beginner an empty board and wish them luck.

The panel opens for anyone. Turn the settings signed out and you get a preview built from the parent's last ten graded games, which is enough to watch the record move before you commit to anything. The full run is a paid view.

What saving actually requires

Keeping your version is the step with a gate on it. Signed out, the button reads sign in to save your spin. A signed-in free account is told that Pro is required to save. With a subscription, that same button reads save as mine.

So the free version of this is real but temporary. You can explore any public model, all of it, and walk away with what you learned rather than a saved copy.

What a fork carries, and what it leaves behind

A saved fork copies the sport, the name with a remix marker added, every slot pick, the assumptions and the stated limitations. It records which model it came from, so the lineage survives.

It does not copy the record. Your fork arrives as a private draft with no graded picks against its name, and the parent's history stays with the parent, where it was earned.

Every model you own counts against a free account's three saves, whatever state it is in, and a fork is a model you own. Publishing one back to the public catalog is a paid step as well, so a free copy stays private.

Change one thing, then train

The temptation on a fork is to change six settings at once and watch the number jump. Then you have a different model and no idea which change made it different.

Sizing is the cheapest first move. Swapping a Kelly stake for a flat one makes the next record measure the model rather than the staking rule sitting on top of it.

Features is the change that moves the trust card furthest. Dropping the market price input is the standard test for a model whose correlation with the book reads high, and it is one edit, so the answer is readable.

The forward record restarts at zero

A fork inherits picks, not proof. Whatever the parent proved, your version has proved nothing until its own calls grade.

Studio keeps that honest at the publish gate, which waits for 20 forward picks to grade before it shows a record. A fork starts that count from nothing on the day you save it.

Where a fork goes next

A saved fork sits in your own list as a private draft until you train it. Training is free and runs on the machine in front of you, so the only real cost of finding out is an evening.

Publishing it back to the catalog is a second paid step. It is also the one that puts your name against a public record, and the catalog does not soften what it shows.

The honest path is slow by design. Fork something that already works, change one setting, train it, and read the trust card. Do that three times and you will know the parent better than its own description ever explained it.

The Receipts Drawer

Forking is the fastest way to learn what a working board looks like, and the slowest way to build a record. Both things are true, and the second one surprises people.

The catalog earns trust with its empty states. A gallery willing to print no record yet, next to a model somebody worked on, is a gallery that will print your bad month too.

Open a card, turn the knobs signed out, and watch how that preview responds before you spend a save on it. If the record barely moves when you change a setting, that setting was not the reason the model works. The public track record is where the claim finally settles.

FAQ

Do I need an account to remix a public model? No. The knob panel opens for anyone, and you can turn the settings and watch the record move without signing in. Saving the result as your own model is the step that needs a paid account.

What does a fork copy from the original? The sport, the name with a remix marker added, every slot pick, the assumptions and the stated limitations. It records which model it came from. It does not copy the record.

Does a fork count against the three free saves? Yes. A free account keeps three saved models, counted across every model you own whatever state it is in, and a saved fork is one of them. Saving a fork needs a paid account, the same as saving any model here, and a paid account has no cap.

What should I change first on a fork? One thing. Sizing is the cheapest place to start, because a flat stake makes the record measure the model. Features is the change that moves the trust card most, and dropping the market price input is the standard test.

Why does my fork show no record? Because it has not made a live call yet. A fork inherits picks, not proof, and the publish gate holds a forward record back until 20 picks have graded.

Where these numbers come from

Frequently asked questions

Do I need an account to remix a public model?
No. The knob panel opens for anyone, and you can turn the settings and watch the record move without signing in. Saving the result as your own model is the step that needs a paid account.
What does a fork copy from the original?
The sport, the name with a remix marker added, every slot pick, the assumptions and the stated limitations. It records which model it came from. It does not copy the record.
Does a fork count against the three free saves?
Yes. A free account keeps three saved models, counted across every model you own whatever state it is in, and a saved fork is one of them. Saving a fork needs a paid account, the same as saving any model here, and a paid account has no cap.
What should I change first on a fork?
One thing. Sizing is the cheapest place to start, because a flat stake makes the record measure the model. Features is the change that moves the trust card most, and dropping the market price input is the standard test.
Why does my fork show no record?
Because it has not made a live call yet. A fork inherits picks, not proof, and the publish gate holds a forward record back until 20 picks have graded.

Build a free model in 60 seconds →

Go →
6m read time
2 players/teams
8 key angles

Angles in this read

  • Edge meter Positive expected value is presented as a meter, not a guarantee.
  • Model sparkline Model output and projection movement get a tiny sparkline rhythm.
  • Probability bands Ranges and uncertainty are shown as bands rather than fake certainty.
  • Widget lift Calculators and decision widgets get a measured surface entrance.
  • Line reveal Pretext-measured lines reveal without reflowing the article.
  • Entity chip Player and team names are surfaced as scannable chips.

This article's context stays anchored to Updated Sep and FAQ Do and model, price and builder, all of which appear in the post itself.

Names and terms found in this article
Updated SepFAQ Domodelpricebuildershark snip builderhow to
Share this guide Help another reader make a sharper decision.

Get picks in your inbox

One email, every slate — ranked edges, no touts. Unsubscribe any time.

Start free — pick a sport

Go →

Continue with evidence

Related reading and source status

Related Reads

Inside the Six-Stage Reveal When Training Finishes — on the Advanced Canvas — Shark Snip
Betting Tools

Inside the Six-Stage Reveal When Training Finishes — on the Advanced Canvas

A finished training run on /build/[slug] plays a fixed six-stage reveal, grade first, timed to the exact millisecond in the source.

Sep 11, 2026 5 min read
The Screen That Shows Your Backend Before Training Starts — on the Advanced Canvas — Shark Snip
Betting Tools

The Screen That Shows Your Backend Before Training Starts — on the Advanced Canvas

Pre-Train Preview on /build/[slug] names the exact backend a run will use, from a single side-effect-free probe, before training starts.

Sep 11, 2026 5 min read
Eight Lattice Slots, and the 56 Modules That Can Fill Them — Shark Snip
Betting Tools

Eight Lattice Slots, and the 56 Modules That Can Fill Them

The /build board holds eight fixed slots filled from a live, growing list of pieces, not a four-tier gallery.

Sep 11, 2026 6 min read

query: loadMergedBlogPostCards + scoreRelated · n = 3

No data

No graded source picks match this article yet

The public.source_accuracy_scores 90-day query returned no rows for this article's inferred sport.