power policy: cap power by range and energy, gate 3.0 on above-average chances

Implements the user's energy management request: "firing from more than 200px
should be a 'not good chances zone' so faster bullets and more chances to hit
matters more than single hit damage with low chances. When we are lower than 50
health, same thing. I would like to use 3.0 power only when the chances of
hitting are higher than average."

Design: a CAP on top of the existing `bestPower`, not a rewrite. `bestPower`
still answers "which bin does this gun's own data prefer"; the policy caps it:

  ramming                                       -> 3.0  (reason ram, exempt)
  dist > TR_POWER_FAR_DIST (200)                -> 1.0  (far)
  elif selfEnergy < TR_POWER_LOW_ENERGY (50)    -> 1.0  (lowEnergy)
  elif pEst <= pRef                             -> 2.0  (belowAvg)
  else                                          -> 3.0  (full)
  power = min(gunPreferredBinPower, cap)   # can only LOWER power

p=1.0 is the right "low" tier on the measured mechanics: bullet speed 20-3p so
p=1.0 gives speed 17 vs 11 at p=3.0 (55% faster = less lead error), fire
interval 10+2p so 12 ticks vs 16 (33% more shots), and drain 0.083/turn vs
0.1875 (2.25x slower). All three things the user asked for at long range.
pEst = the chosen bin's virtual rate (gun aggregate when the bin is empty);
pRef = the gun's aggregate mean unless TR_POWER_REF > 0. No-data guns are
vacuously below-average -> cap 2.0 (conservative, documented).

Control arm: TR_POWER_POLICY=0 = uncapped = today's behaviour exactly.
Knobs: TR_POWER_POLICY, TR_POWER_FAR_DIST, TR_POWER_LOW_ENERGY,
TR_POWER_FAR_CAP, TR_POWER_MID_CAP, TR_POWER_REF, TR_POWER_LOG.
TR_POWER_MID_CAP exists because the user did not specify the middle case
(close + healthy + not-above-average); 2.0 is the default, flippable to 1.0.

Seam: the cap lives in a pure `applyPowerPolicy` and is applied only in
`selectShot` (the single place real shots are chosen), so the logic is testable
without a battle. Ram is wired from `shouldRam` - the same value the movement
dispatch uses for the (0,50) band.

CORRECTION TO AN ASSUMPTION IN THE TASK: `offline_range.nim` does NOT call
`bestPower`/`selectShot` - it only replays virtual-bullet spawn/resolve across
all power bins, independent of the real shot's power. So there is no offline
power-selection path that could diverge from the live one, and the acceptance
test guards the metric, not the policy. Policy coverage therefore comes from the
new unit test.

Verification: test_power_policy 26/26 in BOTH modes (default and TR_POWER_POLICY=0
control arm); test_gun_harness 39, test_vbullet_metric 11, test_power_selection 3,
test_adaptive_radar 41, test_tfil_ring_weights 24; acceptance_offline_vs_online
12/12 VERDICT PASS (live battle). ModularBot compiles.

UNVERIFIED: the live effect on damage/survival/score. No A/B has run.
This commit is contained in:
2026-09-21 23:59:56 +02:00
parent 9caf1d3728
commit c9825dfb0b
4 changed files with 356 additions and 5 deletions
+34 -4
View File
@@ -66,10 +66,40 @@ proc shouldFire*(currentGunDir, targetAngle, gunHeat, distPx: float): bool =
elif delta < -180.0: delta += 360.0
abs(delta) <= aimToleranceDeg(distPx) and gunHeat <= 0.0
proc selectShot*(t: var VirtualTracker, targetId: int = -1, tick = 0): (GunId, int, float) =
proc selectShotPolicy*(t: var VirtualTracker, targetId = -1, tick = 0,
dist = 0.0, selfEnergy = 100.0,
ramming = false): (GunId, int, float, PowerCap) =
## `selectShot` plus the energy-aware power-policy decision, so a caller can
## log the cap and its reason (see `applyPowerPolicy` in virtual_bullets).
##
## `dist` is the current distance (px) to the target and `selfEnergy` our own
## energy; `ramming` exempts the caps (the movement code's `shouldRam` is the
## single source of truth). The policy is applied identically wherever this is
## called, so live and any offline caller cannot diverge.
let gunId = t.selectGun(targetId, tick)
let (prefBin, preferred) = t.bestPower(gunId, targetId)
# pEst / pRef mirror `bestPower`'s own fitness source (per-target when data
# exists, else the deterministic aggregate). An empty bin carries no rate of
# its own, so it borrows the gun's aggregate — the same "no data" case the
# policy documents.
let fit = t.fitnessFor(targetId)
let pRef = if PowerRefFixed > 0.0: PowerRefFixed
else: gunRate(fit[gunId], pooled = true)
let pEst =
if fit[gunId].bins[prefBin].count == 0: pRef
else: fit[gunId].bins[prefBin].hitRate()
let dec = applyPowerPolicy(preferred, dist, selfEnergy, pEst, pRef, ramming)
result = (gunId, binIndexForPower(dec.power), dec.power, dec)
proc selectShot*(t: var VirtualTracker, targetId = -1, tick = 0,
dist = 0.0, selfEnergy = 100.0,
ramming = false): (GunId, int, float) =
## Returns (gunId, powerBinIdx, power) — the shot to take this tick.
## Pass targetId to pick the best gun for that specific enemy. `tick` drives
## the minimum-dwell hysteresis (see `selectGun`).
let gunId = t.selectGun(targetId, tick)
let (binIdx, power) = t.bestPower(gunId, targetId)
## the minimum-dwell hysteresis (see `selectGun`). `dist`/`selfEnergy`/`ramming`
## feed the energy-aware power cap (`TR_POWER_POLICY`); defaults keep every
## existing caller compiling, and `TR_POWER_POLICY=0` reproduces the uncapped
## `bestPower` preference. Use `selectShotPolicy` when the cap/reason is needed.
let (gunId, binIdx, power, _) =
t.selectShotPolicy(targetId, tick, dist, selfEnergy, ramming)
result = (gunId, binIdx, power)
@@ -599,6 +599,103 @@ proc bestPower*(t: VirtualTracker, gunId: GunId, targetId: int = -1): (int, floa
if fw.count == 0 or fw.hitRate() >= bar:
return (binIdx, PowerBins[binIdx])
# ── energy-aware power policy (TR_POWER_*) ───────────────────────────────────
#
# `bestPower` answers "which power bin does this gun's own virtual data prefer?"
# and is deliberately left untouched. The policy below CAPS that preference using
# only cheap, always-available state — range, our own energy, and the gun's own
# rate — so a long-range or low-energy shot trades single-hit damage for a
# faster bullet (speed = 20-3p, so LOW power is FASTER and needs less lead) and a
# shorter fire interval (10+2p, so LOW power = MORE shots). It never RAISES
# power, so the shipped behaviour is exactly the `cap = 3.0` case, which is also
# the control arm (`TR_POWER_POLICY=0`).
#
# Measured basis (real shots vs DrussGT, 8-16 runs): hit rate 21.6% at 0-100px,
# 27.1% at 100-200, then 19.3% at 200-300, 10.9% at 300-400, 6.8% at 400-600 and
# 5.4% at 600-800. Energy math: E[dE] = p(3P-1), so the break-even hit
# probability is 1/3 INDEPENDENT of power; at range/low energy the extra speed
# and shots of p=1.0 dominate. Damage is 4p (p<=1) / 6p-2 (p>1).
const
PowerPolicyEnvVar* = "TR_POWER_POLICY" ## 0 = control arm (uncapped)
PowerFarDistEnvVar* = "TR_POWER_FAR_DIST" ## px; beyond this = bad-chances zone
PowerLowEnergyEnvVar* = "TR_POWER_LOW_ENERGY" ## self energy below this = conserve
PowerFarCapEnvVar* = "TR_POWER_FAR_CAP" ## cap for far / low-energy
PowerMidCapEnvVar* = "TR_POWER_MID_CAP" ## cap when close+healthy but not above avg
PowerRefEnvVar* = "TR_POWER_REF" ## 0 = gun's own mean; >0 = fixed P_ref
let PowerPolicyEnabled* = envBool(PowerPolicyEnvVar, true)
let PowerFarDist* = envFloat(PowerFarDistEnvVar, 200.0)
let PowerLowEnergy* = envFloat(PowerLowEnergyEnvVar, 50.0)
let PowerFarCap* = envFloat(PowerFarCapEnvVar, 1.0)
let PowerMidCap* = envFloat(PowerMidCapEnvVar, 2.0)
let PowerRefFixed* = envFloat(PowerRefEnvVar, 0.0)
type
PowerReason* = enum
prFull ## above-average chances, close, healthy -> full power
prFar ## beyond TR_POWER_FAR_DIST -> bad-chances zone
prLowEnergy ## self energy below TR_POWER_LOW_ENERGY -> conserve
prBelowAvg ## chances not above the gun's own average -> no power 3.0
prRam ## ramming: exempt (at contact P->1, so 3.0 is correct)
PowerCap* = object
power*: float ## the power to fire (<= the gun's preference)
cap*: float ## the cap applied (3.0 = uncapped)
reason*: PowerReason ## why
proc powerReasonName*(r: PowerReason): string =
case r
of prFull: "full"
of prFar: "far"
of prLowEnergy: "lowEnergy"
of prBelowAvg: "belowAvg"
of prRam: "ram"
proc binIndexForPower*(power: float): int =
## Index of `power` in `PowerBins`; if it is not an exact bin value, the
## highest bin whose power does not exceed it (0 if none). Keeps the returned
## bin index consistent with a capped power.
result = 0
for i in 0..<len(PowerBins):
if abs(PowerBins[i] - power) < 1e-9: return i
if PowerBins[i] <= power: result = i
proc applyPowerPolicy*(preferredPower, dist, selfEnergy, pEst, pRef: float,
ramming: bool,
enabled = PowerPolicyEnabled,
farDist = PowerFarDist,
lowEnergy = PowerLowEnergy,
farCap = PowerFarCap,
midCap = PowerMidCap): PowerCap =
## PURE cap core — no tracker, no battle. `power = min(preferredPower, cap)`,
## so the result can only ever LOWER the gun's own preference. Order of
## precedence: ram (exempt) > far > low energy > below average > full.
##
## `pEst` is the gun's rate for the bin it chose (or its aggregate when that
## bin is empty); `pRef` is the gun's aggregate mean (or the fixed
## `TR_POWER_REF`). A cold gun has no data, so `pEst <= pRef` is vacuously
## true and it gets the mid cap — deliberately conservative until it has
## evidence its chances are above average.
if ramming:
return PowerCap(power: preferredPower, cap: 3.0, reason: prRam)
if not enabled:
return PowerCap(power: preferredPower, cap: 3.0, reason: prFull)
var cap: float
var reason: PowerReason
if dist > farDist:
cap = farCap
reason = prFar
elif selfEnergy < lowEnergy:
cap = farCap
reason = prLowEnergy
elif pEst <= pRef:
cap = midCap
reason = prBelowAvg
else:
cap = 3.0
reason = prFull
PowerCap(power: min(preferredPower, cap), cap: cap, reason: reason)
proc chooseFromFit*(fit: seq[GunFitness], diag: ptr SelectorDiag = nil,
mode: SelectorMode = smAbsolute,
referenceRate = -1.0,
+207
View File
@@ -0,0 +1,207 @@
## Unit guard for the energy-aware power policy (TR_POWER_*).
##
## The policy is a CAP on the gun's own preferred bin: at long range or low
## energy it trades single-hit damage for a faster bullet and more shots, and it
## withholds power 3.0 unless the gun's own chances for the chosen bin are above
## that gun's average. It must NEVER raise power, and `TR_POWER_POLICY=0` must
## reproduce the uncapped preference exactly (the control arm).
##
## Pure: no Java, no battle. Run:
## nim c -r common_libs/tests/test_power_policy.nim
## and once with the flag off:
## TR_POWER_POLICY=0 nim c -r common_libs/tests/test_power_policy.nim
import std/[strformat, math, tables]
import gun_harness/virtual_bullets
import gun_harness/selector
var failures = 0
proc check(name: string, ok: bool) =
if ok: echo "PASS: ", name
else: echo "FAIL: ", name; inc failures
proc recordHit(fw: var FitnessWindow, hit: bool) =
fw.hits[fw.head] = hit
fw.head = (fw.head + 1) mod WindowSize
inc fw.count
proc seedWindow(t: var VirtualTracker, targetId, gunId, binIdx, hits, misses: int) =
if targetId notin t.fitness:
t.fitness[targetId] = newSeq[GunFitness](t.numGuns)
var fw = addr t.fitness[targetId][gunId].bins[binIdx]
for _ in 0..<hits: recordHit(fw[], true)
for _ in 0..<misses: recordHit(fw[], false)
# ── pure cap core ────────────────────────────────────────────────────────────
proc testFar() =
let d = applyPowerPolicy(preferredPower = 3.0, dist = 250.0, selfEnergy = 100.0,
pEst = 0.9, pRef = 0.1, ramming = false, enabled = true)
check "distance > TR_POWER_FAR_DIST -> cap 1.0 (far)",
d.power == 1.0 and d.cap == 1.0 and d.reason == prFar
proc testLowEnergy() =
let d = applyPowerPolicy(3.0, 100.0, 30.0, 0.9, 0.1, false, enabled = true)
check "self energy < TR_POWER_LOW_ENERGY -> cap 1.0 (lowEnergy)",
d.power == 1.0 and d.cap == 1.0 and d.reason == prLowEnergy
proc testBelowAverage() =
# pEst == pRef is "not above average": withhold power 3.0 -> cap 2.0.
let d = applyPowerPolicy(3.0, 100.0, 100.0, 0.2, 0.2, false, enabled = true)
check "chances not above average -> cap 2.0 (belowAvg)",
d.power == 2.0 and d.cap == 2.0 and d.reason == prBelowAvg
# Strictly below also caps.
let e = applyPowerPolicy(3.0, 100.0, 100.0, 0.1, 0.2, false, enabled = true)
check "chances strictly below average -> cap 2.0 (belowAvg)",
e.power == 2.0 and e.reason == prBelowAvg
proc testAboveAverageFull() =
let d = applyPowerPolicy(3.0, 100.0, 100.0, 0.5, 0.2, false, enabled = true)
check "above average + close + healthy -> cap 3.0 (full)",
d.power == 3.0 and d.cap == 3.0 and d.reason == prFull
proc testRamExempt() =
# Far, low energy, no chance data: still full power because we are ramming.
let d = applyPowerPolicy(3.0, 500.0, 5.0, 0.0, 0.9, true)
check "ramming exempts the caps even far + low energy",
d.power == 3.0 and d.reason == prRam
# Ram beats far even with the policy on.
let e = applyPowerPolicy(3.0, 500.0, 5.0, 0.0, 0.9, true, enabled = true)
check "ram exemption takes precedence over the far cap", e.power == 3.0
proc testCapNeverRaises() =
# Preferred below every cap must pass through untouched.
let a = applyPowerPolicy(1.0, 500.0, 5.0, 0.0, 0.9, false, enabled = true)
check "far cap does not raise a preferred p=1.0", a.power == 1.0
let b = applyPowerPolicy(1.5, 100.0, 100.0, 0.1, 0.2, false, enabled = true)
check "mid cap does not raise a preferred p=1.5", b.power == 1.5
let c = applyPowerPolicy(1.5, 100.0, 100.0, 0.9, 0.2, false, enabled = true)
check "full cap does not raise a preferred p=1.5", c.power == 1.5
proc testPrecedence() =
# far beats low energy, and low energy beats belowAvg.
let a = applyPowerPolicy(3.0, 500.0, 5.0, 0.0, 0.9, false, enabled = true)
check "far takes precedence over low energy", a.reason == prFar
let b = applyPowerPolicy(3.0, 100.0, 5.0, 0.0, 0.9, false, enabled = true)
check "low energy takes precedence over belowAvg", b.reason == prLowEnergy
proc testDisabledUncapped() =
# enabled=false is the code path TR_POWER_POLICY=0 drives.
for p in [1.0, 1.5, 2.0, 3.0]:
let d = applyPowerPolicy(p, 500.0, 5.0, 0.0, 0.9, false, enabled = false)
check fmt"policy off reproduces the uncapped preference p={p:.1f}",
d.power == p and d.cap == 3.0
proc testEnvFlag() =
# The flag is read once at module init, so the active branch is selected by
# the process environment. Run the test twice (default and TR_POWER_POLICY=0).
if PowerPolicyEnabled:
check "TR_POWER_POLICY default (on): far shot is capped to 1.0",
applyPowerPolicy(3.0, 300.0, 100.0, 0.5, 0.2, false).power == 1.0
else:
check "TR_POWER_POLICY=0: far shot is NOT capped (control arm)",
applyPowerPolicy(3.0, 300.0, 100.0, 0.5, 0.2, false).power == 3.0
proc testBinIndex() =
check "binIndexForPower maps the shipped bins exactly",
binIndexForPower(1.0) == 0 and binIndexForPower(1.5) == 1 and
binIndexForPower(2.0) == 2 and binIndexForPower(3.0) == 3
check "binIndexForPower snaps a non-bin cap to the highest bin not above it",
binIndexForPower(2.5) == 2 and binIndexForPower(0.5) == 0
proc testReasonNames() =
check "reason names match the documented log vocabulary",
powerReasonName(prFar) == "far" and
powerReasonName(prLowEnergy) == "lowEnergy" and
powerReasonName(prBelowAvg) == "belowAvg" and
powerReasonName(prFull) == "full" and
powerReasonName(prRam) == "ram"
# ── integration through the tracker (per-target fitness + empty bins) ─────────
# These depend on the process-wide flag, so each branch asserts the behaviour of
# the mode it is actually running in.
proc testTrackerAboveAverage() =
var t = initTracker(1)
# bin3 is far above this gun's aggregate (50% vs 20%): above average.
seedWindow(t, 7, 0, 0, 10, 90)
seedWindow(t, 7, 0, 1, 10, 90)
seedWindow(t, 7, 0, 2, 10, 90)
seedWindow(t, 7, 0, 3, 50, 50)
let (g, _, p, d) = t.selectShotPolicy(7, tick = 0, dist = 100.0,
selfEnergy = 100.0, ramming = false)
if PowerPolicyEnabled:
check "tracker: above-average bin + close + healthy -> power 3.0",
g == 0 and p == 3.0 and d.reason == prFull
else:
check "tracker (control): above-average bin still fires power 3.0",
g == 0 and p == 3.0
proc testTrackerBelowAverage() =
var t = initTracker(1)
for b in 0..<len(PowerBins):
seedWindow(t, 7, 0, b, 20, 80) # all bins equal: chosen bin not above mean
let (_, _, p, d) = t.selectShotPolicy(7, tick = 0, dist = 100.0,
selfEnergy = 100.0, ramming = false)
if PowerPolicyEnabled:
check "tracker: flat gun -> cap 2.0 (belowAvg)", p == 2.0 and d.reason == prBelowAvg
else:
check "tracker (control): flat gun keeps its uncapped preference (power 3.0)",
p == 3.0
proc testTrackerFarAndLowEnergy() =
var t = initTracker(1)
seedWindow(t, 7, 0, 3, 50, 50)
let (_, _, pFar, dFar) = t.selectShotPolicy(7, 0, dist = 250.0,
selfEnergy = 100.0, ramming = false)
let (_, _, pLow, dLow) = t.selectShotPolicy(7, 0, dist = 100.0,
selfEnergy = 30.0, ramming = false)
if PowerPolicyEnabled:
check "tracker: far -> power 1.0", pFar == 1.0 and dFar.reason == prFar
check "tracker: low energy -> power 1.0", pLow == 1.0 and dLow.reason == prLowEnergy
else:
check "tracker (control): far does NOT cap (uncapped preference)", pFar == 3.0
check "tracker (control): low energy does NOT cap (uncapped preference)", pLow == 3.0
proc testTrackerColdAndEmptyBin() =
var t = initTracker(1)
let (_, _, pCold, dCold) = t.selectShotPolicy(7, 0, dist = 100.0,
selfEnergy = 100.0, ramming = false)
if PowerPolicyEnabled:
# Cold gun: no data at all -> pEst <= pRef vacuously -> mid cap, but the
# preferred bin is already 1.0, so the fired power stays 1.0.
check "tracker: cold gun gets the mid cap but keeps its p=1.0 preference",
pCold == 1.0 and dCold.reason == prBelowAvg
else:
check "tracker (control): cold gun keeps its p=1.0 preference", pCold == 1.0
proc testTrackerRamExempt() =
var t = initTracker(1)
seedWindow(t, 7, 0, 3, 50, 50)
let (_, _, p, d) = t.selectShotPolicy(7, 0, dist = 500.0,
selfEnergy = 5.0, ramming = true)
check "tracker: ramming exempts the caps", p == 3.0 and d.reason == prRam
# ── driver ───────────────────────────────────────────────────────────────────
testFar()
testLowEnergy()
testBelowAverage()
testAboveAverageFull()
testRamExempt()
testCapNeverRaises()
testPrecedence()
testDisabledUncapped()
testEnvFlag()
testBinIndex()
testReasonNames()
testTrackerAboveAverage()
testTrackerBelowAverage()
testTrackerFarAndLowEnergy()
testTrackerColdAndEmptyBin()
testTrackerRamExempt()
if failures > 0:
echo "\n", failures, " check(s) FAILED"
quit(1)
echo "\nAll power-policy checks passed."