Files
Pete/internal/games/holdem/betting.go
prosolis e6c1bd3b54 games: the poker table opens, and the bots go back to school
Phase 4. Hold'em, and it's the only table in the casino that is a session
rather than a game: you buy in, play as many hands as you like, and leave with
what's in front of you. So the live row spans hands and chips cross the border
exactly twice. Everything in between is inside the engine.

The bots move inside ApplyMove, as UNO's do, which is what keeps poker off a
socket: shove all-in and the flop, turn, river, showdown and payout all come
back in one response, as a script the felt plays back.

The CFR policy the plan called "the single highest-value asset in either repo"
was never read. Not once, in the whole life of the game: the trainer wrote its
info-set keys under IP/OOP and the runtime looked them up under BTN/SB/BB, so
every lookup missed and fell silently through to a pot-odds heuristic. Nothing
looked broken, because a policy miss is not an error. And it was the wrong
policy anyway — ten big blinds deep, trained on a tree where a call always ends
the street, which is not poker. So the trainer is rewritten to play the real
engine through the real reducer, at every stack depth the table deals, and the
trainer and the table now build the key with the same function so they cannot
drift apart again. A test fails if the bots stop finding themselves in it.

Three money bugs, and the tests earned their keep. Chip conservation across a
hundred sessions caught an uncalled bet that minted chips. A var-init ordering
trap meant every card was identical, every showdown tied and every bot believed
it held exactly 50% equity. And the browser caught the rake being silently
zero — the tier said 5 meaning percent, the casino handed it 0.05 meaning a
fraction, and integer division took the house's cut down to nothing.

Claude-Session: https://claude.ai/code/session_013M5nD7PgUboJXoDcYHzpuJ
2026-07-14 09:08:59 -07:00

280 lines
7.3 KiB
Go

package holdem
import "sort"
// The betting rules. These are the fiddly ones — min-raise, short all-ins that
// don't reopen the action, side pots — and they came over from gogobee, where
// they had no tests at all. They have some now.
// blinds posts the small and the big. Heads-up is the exception every poker
// implementation gets wrong once: with two players the button *is* the small
// blind and acts first before the flop, and last after it.
func (s *State) blinds(evs *[]Event) (bb int) {
var sb int
if s.dealt() == 2 {
sb, bb = s.Button, s.nextIn(s.Button)
} else {
sb = s.nextIn(s.Button)
bb = s.nextIn(sb)
}
s.post(sb, s.Tier.SB, "small", evs)
s.post(bb, s.Tier.BB, "big", evs)
s.Bet = s.Tier.BB
s.MinRaise = s.Tier.BB
s.Aggressor = bb // the big blind has the option to raise their own blind
return bb
}
// post puts a blind up. A player too short to cover it is all-in for what they
// have, which is legal and is why the amount is clamped rather than refused.
func (s *State) post(seat int, amount int64, which string, evs *[]Event) {
p := &s.Seats[seat]
if amount > p.Stack {
amount = p.Stack
}
p.Stack -= amount
p.Bet = amount
p.Total = amount
if p.Stack == 0 {
p.State = AllIn
}
*evs = append(*evs, Event{Kind: "blind", Seat: seat, Amount: amount, Text: which})
}
// firstPreFlop is under the gun: the seat after the big blind, or the button
// itself when the table is heads-up.
//
// The button only gets it if the button can still act. A short stack can be
// all-in on its own blind — post a small blind of 1 with 1 chip left and you are
// in the hand with no chips and no say — and handing the action to a seat that
// cannot act wedges the table.
func (s *State) firstPreFlop(bb int) int {
if s.dealt() == 2 && s.Seats[s.Button].State == Active {
return s.Button
}
if s.dealt() == 2 {
return s.nextCanAct(s.Button)
}
return s.nextCanAct(bb)
}
// firstPostFlop is the first seat left of the button, on every street after the
// flop. The button acts last from here on, which is the whole point of it.
func (s *State) firstPostFlop() int { return s.nextCanAct(s.Button) }
// ---- the five things a seat can do ----------------------------------------
func (s *State) fold(seat int, evs *[]Event) {
p := &s.Seats[seat]
p.State = Folded
p.Acted = true
s.History += "f"
*evs = append(*evs, Event{Kind: "action", Seat: seat, Text: "fold"})
}
func (s *State) check(seat int, evs *[]Event) error {
p := &s.Seats[seat]
if p.Bet < s.Bet {
return ErrCantCheck
}
p.Acted = true
s.History += "c"
*evs = append(*evs, Event{Kind: "action", Seat: seat, Text: "check"})
return nil
}
func (s *State) call(seat int, evs *[]Event) error {
p := &s.Seats[seat]
owed := s.Bet - p.Bet
if owed <= 0 {
return ErrNothingToCall
}
if owed > p.Stack {
owed = p.Stack // a call for less than the bet is a call all-in
}
p.Stack -= owed
p.Bet += owed
p.Total += owed
p.Acted = true
text := "call"
if p.Stack == 0 {
p.State = AllIn
text = "allin"
s.History += "a"
} else {
s.History += "c"
}
*evs = append(*evs, Event{Kind: "action", Seat: seat, Text: text, Amount: owed, Total: p.Bet})
return nil
}
// raise raises *to* a total, not *by* an amount. Every poker interface in the
// world means the total, and a browser that means the other thing bets wrong.
func (s *State) raise(seat int, to int64, evs *[]Event) error {
p := &s.Seats[seat]
most := p.Bet + p.Stack
if to > most {
return ErrTooBig
}
if to < s.Bet+s.MinRaise && to < most {
return ErrTooSmall // only a shove may be smaller than a legal raise
}
added := to - p.Bet
over := to - s.Bet
p.Stack -= added
p.Bet = to
p.Total += added
p.Acted = true
if over > 0 {
s.MinRaise = over
}
s.Bet = to
s.Aggressor = seat
text := "raise"
if p.Stack == 0 {
p.State = AllIn
text = "allin"
s.History += "a"
} else {
// The policy was trained against a tree with two raise sizes in it, so the
// history it reads has to say which one this was: R for a pot-sized raise
// or bigger, r for anything smaller.
if pot := s.inPlay(); pot > 0 && float64(over) >= float64(pot)*0.75 {
s.History += "R"
} else {
s.History += "r"
}
}
*evs = append(*evs, Event{Kind: "action", Seat: seat, Text: text, Amount: added, Total: to})
return nil
}
// allin pushes the lot.
//
// A short all-in does not reopen the betting. If a player shoves for less than
// a full raise over the current bet, players who have already acted may call it
// but may not raise again — otherwise a tiny stack could be used to reopen the
// action for a partner, which is the oldest collusion trick there is.
func (s *State) allin(seat int, evs *[]Event) error {
p := &s.Seats[seat]
if p.Stack <= 0 {
return ErrNoChips
}
added := p.Stack
to := p.Bet + added
p.Stack = 0
p.Bet = to
p.Total += added
p.State = AllIn
p.Acted = true
if to > s.Bet {
if over := to - s.Bet; over >= s.MinRaise {
s.MinRaise = over
s.Aggressor = seat
}
s.Bet = to
}
s.History += "a"
*evs = append(*evs, Event{Kind: "action", Seat: seat, Text: "allin", Amount: added, Total: to})
return nil
}
// ---- when is a street over ------------------------------------------------
// streetDone reports whether the betting round is finished, given the seat the
// action would pass to next.
//
// The "has acted" check is the load-bearing half. The big blind has money in
// front of them without having chosen to put it there, so a round where
// everybody merely limps in has all bets matched while the blind has never had
// a say. Without this, they never get their option.
func (s *State) streetDone(next int) bool {
if s.canActCount() == 0 {
return true
}
for i := range s.Seats {
p := &s.Seats[i]
if p.State != Active {
continue
}
if p.Bet != s.Bet || !p.Acted {
return false
}
}
// The last aggressor being all-in means the action can't get back to them:
// everyone left has matched the bet above, so there is nothing more to do.
if s.Seats[s.Aggressor].State == AllIn {
return true
}
return next == s.Aggressor
}
// ---- side pots -------------------------------------------------------------
// sidePots slices the pot into layers, one per distinct all-in level. A player
// can only win the part of the pot they could have lost, so each layer is
// contested by exactly the players who paid into it.
//
// Folded players' chips stay in the pot — they paid for the right to fold — but
// they are eligible for nothing.
func (s *State) sidePots() {
s.collect()
var levels []int64
for i := range s.Seats {
p := &s.Seats[i]
if p.State == Folded || p.State == Out || p.Total == 0 {
continue
}
levels = append(levels, p.Total)
}
if len(levels) == 0 {
return
}
sort.Slice(levels, func(i, j int) bool { return levels[i] < levels[j] })
var pots []Pot
var prev int64
for _, level := range levels {
if level <= prev {
continue
}
var amount int64
var eligible []int
for i := range s.Seats {
p := &s.Seats[i]
paid := p.Total - prev
if paid > level-prev {
paid = level - prev
}
if paid > 0 {
amount += paid // folded money counts toward the pot...
}
if p.State != Folded && p.State != Out && p.Total >= level {
eligible = append(eligible, i) // ...but wins no part of it
}
}
if amount > 0 {
pots = append(pots, Pot{Amount: amount, Eligible: eligible})
}
prev = level
}
if len(pots) > 0 {
s.Side = pots
s.Pot = 0
}
}