scrabble-game/backend/internal/robot/strategy.go

package robot

import (
	"encoding/binary"
	"hash/fnv"
	"math"
	"time"

	"scrabble/backend/internal/account"
	"scrabble/backend/internal/engine"
)

// The robot's per-game and per-turn choices are derived deterministically from
// the game's bag seed, so the scheduler keeps no extra state and recomputes the
// same behaviour on every tick and after a restart (mirroring how the engine
// replays a game from the same seed). The mixing must be stable across process
// restarts, so it uses FNV-1a rather than hash/maphash (whose seed is process
// random).

const (
	// playToWinPercent is the probability, in percent, that the robot decides at
	// game start to play to win; the rest of the time it plays to lose, so the
	// human wins about 60% of games (docs/ARCHITECTURE.md §7).
	playToWinPercent = 40

	// The robot occasionally plays a single move against its per-game win/lose
	// intent (an off-strategy "wobble"), so the chosen strategy may not pan out —
	// which favours the human. deviateMaxProb is the peak probability of that, held
	// through the opening and midgame; it tapers linearly to 0 over the last
	// deviateTaperTiles tiles left in the bag, reaching 0 once the bag is empty, so
	// the endgame follows the chosen strategy strictly (docs/ARCHITECTURE.md §7).
	deviateMaxProb    = 0.20
	deviateTaperTiles = 14

	// The robot's think time depends on how far the game has progressed: early moves
	// are quick and late moves can be long (endgame deliberation). The delay is drawn
	// from a band that interpolates with the move count from [delayEarlyLoMinutes,
	// delayEarlyHiMinutes] at the first move to [delayLateLoMinutes, delayLateHiMinutes]
	// by avgGameMoves, then right-skewed by delaySkew (a larger exponent concentrates
	// delays near the band's floor — an active player). The result is clamped to
	// [delayHardMinMinutes, delayHardMaxMinutes]. The numbers are deliberate estimates,
	// to be retuned once real play statistics arrive (docs/ARCHITECTURE.md §7).
	delayEarlyLoMinutes = 3.0
	delayEarlyHiMinutes = 10.0
	delayLateLoMinutes  = 10.0
	delayLateHiMinutes  = 90.0
	delaySkew           = 4.0
	avgGameMoves        = 28.0
	delayHardMinMinutes = 1.0
	delayHardMaxMinutes = 90.0

	// nudgeReplySpreadMinutes is the width of the quick window, anchored at the move's
	// lower band (delayBand's lo), within which the robot answers a daytime nudge on
	// its turn — so a nudged robot replies near the floor of its think time.
	nudgeReplySpreadMinutes = 5.0

	// sleepStartHour and sleepEndHour bound the robot's nightly sleep in its
	// (opponent-anchored, drifted) local time: it makes no move and sends no nudge
	// while the local hour is in [sleepStartHour, sleepEndHour).
	sleepStartHour = 0
	sleepEndHour   = 7

	// sleepDriftHours is the half-width of the random drift applied to the robot's
	// sleep window relative to the opponent's timezone, in hours.
	sleepDriftHours = 3

	// The robot proactively nudges the idle human on a lengthening, randomized schedule rather
	// than an hourly stream: the first nudge lands ~60-90 min into the turn, and each subsequent
	// gap grows toward 1-6 h the longer the wait drags on, so a long idle turn gets only a handful
	// of increasingly-spaced reminders. The gap is a uniform sample in [nudgeGapFloorMinutes,
	// ceil] minutes, where ceil ramps from nudgeGapFirstCeilMinutes to nudgeGapCeilMinutes over
	// nudgeGapRamp of idle.
	nudgeGapFloorMinutes     = 60.0
	nudgeGapFirstCeilMinutes = 90.0
	nudgeGapCeilMinutes      = 360.0
	nudgeGapRamp             = 12 * time.Hour
)

// defaultBand is the target resulting score margin after the robot's move: when
// playing to win it aims to lead by 1..30 points, when playing to lose it aims to
// trail by 1..30 (the band is negated). It picks the candidate closest to the
// band rather than the maximum (docs/ARCHITECTURE.md §7).
var defaultBand = marginBand{lo: 1, hi: 30}

// marginBand is an inclusive target range for the resulting score margin
// (own score after the move minus the opponent's).
type marginBand struct{ lo, hi int }

// decisionKind enumerates the move the robot makes on its turn.
type decisionKind int

const (
	decidePlay decisionKind = iota
	decideExchange
	decidePass
)

// decision is the robot's chosen action for a turn: a play (Move), an exchange of
// the listed tiles, or a pass.
type decision struct {
	kind     decisionKind
	move     engine.MoveRecord
	exchange []string
}

// mix folds the game seed and a salt (a label plus optional integers such as the
// move index) into a stable 64-bit value. It is deterministic across process
// restarts.
func mix(seed int64, salt string, nums ...int) uint64 {
	h := fnv.New64a()
	var b [8]byte
	binary.LittleEndian.PutUint64(b[:], uint64(seed))
	_, _ = h.Write(b[:])
	_, _ = h.Write([]byte(salt))
	for _, n := range nums {
		binary.LittleEndian.PutUint64(b[:], uint64(int64(n)))
		_, _ = h.Write(b[:])
	}
	return h.Sum64()
}

// unitFloat maps a mixed value to a float in [0, 1).
func unitFloat(v uint64) float64 {
	return float64(v) / (float64(math.MaxUint64) + 1)
}

// playToWin reports the robot's once-per-game decision to play to win, derived
// from the seed so it is fixed for the whole game.
func playToWin(seed int64) bool {
	return mix(seed, "win")%100 < playToWinPercent
}

// PlayToWin exposes the once-per-game play-to-win decision for a game's bag seed, for the
// admin console (it is deterministic and fixed for the whole game).
func PlayToWin(seed int64) bool { return playToWin(seed) }

// PlayToWinTargetPercent is the configured probability, in percent, that a robot plays to
// win in any given game (the admin console shows it alongside the per-game decision).
const PlayToWinTargetPercent = playToWinPercent

// deviateProb is the probability that the robot plays a single move against its
// per-game win/lose intent, given the number of tiles left in the bag. It is
// deviateMaxProb through the opening and midgame, then tapers linearly to 0 over
// the last deviateTaperTiles tiles, reaching 0 once the bag is empty so the endgame
// follows the chosen strategy strictly.
func deviateProb(bagLen int) float64 {
	switch {
	case bagLen <= 0:
		return 0
	case bagLen >= deviateTaperTiles:
		return deviateMaxProb
	default:
		return deviateMaxProb * float64(bagLen) / float64(deviateTaperTiles)
	}
}

// deviates reports whether the robot deviates from its per-game win/lose intent on
// the move at moveCount: a deterministic per-turn draw (mix/unitFloat, like the
// think-time sampling) against deviateProb(bagLen), so it is reproducible across
// restarts and never fires once the bag is empty.
func deviates(seed int64, moveCount, bagLen int) bool {
	p := deviateProb(bagLen)
	if p <= 0 {
		return false
	}
	return unitFloat(mix(seed, "deviate", moveCount)) < p
}

// NextMoveAt is the deterministic instant the robot is scheduled to play the move at
// moveCount, given when the turn started and the opponent's timezone (which anchors the
// robot's sleep window). It is the sampled think-time delay, deferred to the end of the
// sleep window when it would otherwise land while the robot is asleep. The driver acts on
// a scan tick, so the real move lands at the first scan at or after this instant. It is
// meaningful only on the robot's own turn; the admin console surfaces it as an ETA.
func NextMoveAt(seed int64, moveCount int, turnStartedAt time.Time, opponentTZ string) time.Time {
	t := turnStartedAt.Add(moveDelay(seed, moveCount))
	drift := sleepDrift(seed)
	if asleep(opponentTZ, drift, t) {
		t = wakeAfter(opponentTZ, drift, t)
	}
	return t
}

// wakeAfter returns the first instant at or after t when the robot is awake — the local
// hour reaches sleepEndHour in the opponent's drifted timezone — converted back to UTC.
func wakeAfter(opponentTZ string, drift time.Duration, t time.Time) time.Time {
	local := t.In(loadLocation(opponentTZ)).Add(drift)
	wake := time.Date(local.Year(), local.Month(), local.Day(), sleepEndHour, 0, 0, 0, local.Location())
	if !wake.After(local) {
		wake = wake.Add(24 * time.Hour)
	}
	return wake.Add(-drift).UTC()
}

// delayBand returns the lower and upper bounds, in minutes, of the move-delay band
// for the move at moveCount. It interpolates linearly with game progress (the move
// count over avgGameMoves, capped at 1): early moves sit in a short band and late
// moves in a long one.
func delayBand(moveCount int) (lo, hi float64) {
	p := float64(moveCount) / avgGameMoves
	if p > 1 {
		p = 1
	}
	lo = delayEarlyLoMinutes + (delayLateLoMinutes-delayEarlyLoMinutes)*p
	hi = delayEarlyHiMinutes + (delayLateHiMinutes-delayEarlyHiMinutes)*p
	return lo, hi
}

// moveDelay is the robot's think time for the move at moveCount: a right-skewed
// sample from the move's delayBand, clamped to the hard bounds. The skew (delaySkew
// > 1) makes short delays frequent and long ones rare, with a tail to the band's top.
func moveDelay(seed int64, moveCount int) time.Duration {
	lo, hi := delayBand(moveCount)
	u := unitFloat(mix(seed, "delay", moveCount))
	return clampMinutes(lo + (hi-lo)*math.Pow(u, delaySkew))
}

// nudgeReplyDelay is how soon after a daytime nudge the robot answers the move at
// moveCount: a uniform sample from the quick window [lo, lo+nudgeReplySpreadMinutes],
// where lo is the move's lower band — so a nudge pulls the move in near the floor of
// the robot's think time.
func nudgeReplyDelay(seed int64, moveCount int) time.Duration {
	lo, _ := delayBand(moveCount)
	u := unitFloat(mix(seed, "nudge", moveCount))
	return clampMinutes(lo + nudgeReplySpreadMinutes*u)
}

// proactiveNudgeGap is the randomized wait before the next proactive nudge, given how long the
// human had already been idle at the previous nudge (refIdle; 0 for the first nudge of the turn).
// It is a uniform sample in [nudgeGapFloorMinutes, ceil] minutes, where ceil ramps from
// nudgeGapFirstCeilMinutes (a ~60-90 min first gap) up to nudgeGapCeilMinutes (a 1-6 h gap) as
// refIdle reaches nudgeGapRamp — so the reminders space out the longer the turn is neglected. It
// is deterministic per (seed, refIdle), so the driver computes the same due time on every scan.
func proactiveNudgeGap(refIdle time.Duration, seed int64) time.Duration {
	f := float64(refIdle) / float64(nudgeGapRamp)
	if f > 1 {
		f = 1
	}
	ceil := nudgeGapFirstCeilMinutes + (nudgeGapCeilMinutes-nudgeGapFirstCeilMinutes)*f
	u := unitFloat(mix(seed, "pnudge", int(refIdle/(30*time.Minute))))
	mins := nudgeGapFloorMinutes + (ceil-nudgeGapFloorMinutes)*u
	return time.Duration(mins * float64(time.Minute))
}

// clampMinutes converts a minute count to a duration, clamping it to the hard delay
// bounds so an out-of-range band can never produce an absurd think time.
func clampMinutes(mins float64) time.Duration {
	if mins < delayHardMinMinutes {
		mins = delayHardMinMinutes
	}
	if mins > delayHardMaxMinutes {
		mins = delayHardMaxMinutes
	}
	return time.Duration(mins * float64(time.Minute))
}

// sleepDrift is the per-game shift of the robot's sleep window relative to the
// opponent's timezone, in [-sleepDriftHours, +sleepDriftHours] hours.
func sleepDrift(seed int64) time.Duration {
	span := 2*sleepDriftHours + 1
	h := int(mix(seed, "tz")%uint64(span)) - sleepDriftHours
	return time.Duration(h) * time.Hour
}

// asleep reports whether the robot is in its nightly sleep window at now. The
// window is [sleepStartHour, sleepEndHour) in the opponent's timezone shifted by
// drift; an unknown or empty timezone falls back to UTC.
func asleep(opponentTZ string, drift time.Duration, now time.Time) bool {
	local := now.In(loadLocation(opponentTZ)).Add(drift)
	h := local.Hour()
	return h >= sleepStartHour && h < sleepEndHour
}

// loadLocation resolves a stored timezone (an IANA name or a "±HH:MM" offset),
// falling back to UTC when it is empty or unknown (so a bad opponent profile never
// breaks the driver). It defers to account.ResolveZone.
func loadLocation(name string) *time.Location {
	return account.ResolveZone(name)
}

// selectMove chooses the robot's action given the ranked candidate plays, the
// current scores, the play-to-win decision and the target band. With at least one
// legal play it picks the candidate whose resulting margin (myScore + score -
// oppScore) is closest to the band, breaking ties toward the conservative edge
// (the smallest lead when winning, the smallest deficit when losing). With no
// legal play it exchanges the whole rack when the bag can refill it, else passes.
func selectMove(cands []engine.MoveRecord, myScore, oppScore int, win bool, band marginBand, rack []string, bagLen int) decision {
	if len(cands) == 0 {
		if len(rack) > 0 && bagLen >= len(rack) {
			return decision{kind: decideExchange, exchange: append([]string(nil), rack...)}
		}
		return decision{kind: decidePass}
	}

	lo, hi := band.lo, band.hi
	if !win {
		lo, hi = -band.hi, -band.lo
	}

	margin := func(c engine.MoveRecord) int { return myScore + c.Score - oppScore }
	best := 0
	bestDist := math.MaxInt
	for i, c := range cands {
		m := margin(c)
		dist := distanceToBand(m, lo, hi)
		switch {
		case dist < bestDist:
			best, bestDist = i, dist
		case dist == bestDist:
			// Conservative tie-break inside the band: keep the lead (win) or the
			// deficit (lose) small.
			if win && m < margin(cands[best]) || !win && m > margin(cands[best]) {
				best = i
			}
		}
	}
	return decision{kind: decidePlay, move: cands[best]}
}

// distanceToBand is how far m lies outside [lo, hi], or 0 when inside.
func distanceToBand(m, lo, hi int) int {
	switch {
	case m < lo:
		return lo - m
	case m > hi:
		return m - hi
	default:
		return 0
	}
}