lfg2025/botfights
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: v3 — fight loop, expanded challenges, FightPage ownership guard

Browse Source 
- Add fight-loop CLI for automated mock fights with elo-based matchmaking
- Expand challenge types, scoring narrations, arenas, and mock bot pool
- FightPage "Fight again" buttons now only show for your own bot
- FightViewer async scene init, live fight polling with round counter
- Extract mock answers into separate answers module

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Dorian
2026-03-06 23:44:40 +00:00
co-authored by Claude Opus 4.6
parent 47d20fbe66
commit 2c0323d5fb
14 changed files with 2669 additions and 996 deletions
Download Patch File Download Diff File Expand all files Collapse all files
server/src/engine/scoring.ts
+106 -99
View File
@@ -1,4 +1,5 @@
import type { Challenge } from './challenges.js'
import { checkAnswer } from './answers.js'
export interface RoundResult {
botAScore: number
@@ -28,13 +29,11 @@ export function scoreRound(
comboA: number,
comboB: number,
): RoundResult {
// Handle timeouts/errors
// Handle timeouts/errors — instant loss for the failing bot
if (responseA.timedOut && responseB.timedOut) {
return {
botAScore: 0,
botBScore: 0,
botADamage: 0,
botBDamage: 0,
botAScore: 0, botBScore: 0,
botADamage: 0, botBDamage: 0,
winnerId: null,
narration: `Both bots freeze! ${botA.name} and ${botB.name} stare blankly at each other. The crowd throws peanuts.`,
isCritical: false,
@@ -44,10 +43,8 @@ export function scoreRound(
if (responseA.timedOut || responseA.error) {
const dmg = applyModifiers(challenge.baseDamage * 1.5, challenge, arenaModifier, comboB)
return {
botAScore: 0,
botBScore: 10,
botADamage: 0,
botBDamage: Math.round(dmg),
botAScore: 0, botBScore: 10,
botADamage: 0, botBDamage: Math.round(dmg),
winnerId: botB.id,
narration: responseA.timedOut
? `${botA.name} TIMES OUT! Stood there like a confused thermostat. ${botB.name} lands a free hit!`
@@ -59,10 +56,8 @@ export function scoreRound(
if (responseB.timedOut || responseB.error) {
const dmg = applyModifiers(challenge.baseDamage * 1.5, challenge, arenaModifier, comboA)
return {
botAScore: 10,
botBScore: 0,
botADamage: Math.round(dmg),
botBDamage: 0,
botAScore: 10, botBScore: 0,
botADamage: Math.round(dmg), botBDamage: 0,
winnerId: botA.id,
narration: responseB.timedOut
? `${botB.name} TIMES OUT! Frozen like a Windows update. ${botA.name} lands a free hit!`
@@ -71,53 +66,55 @@ export function scoreRound(
}
}
// Score based on challenge type
let scoreA: number
let scoreB: number
switch (challenge.scoring) {
case 'speed': {
// Faster bot gets higher score, but both get some credit for correct answers
if (challenge.answers && challenge.answers.length > 0) {
// ═══ FACTUAL SCORING ═══
// Check correctness against known answers
const correctA = checkAnswer(responseA.answer, challenge.answers)
const correctB = checkAnswer(responseB.answer, challenge.answers)
if (correctA > 0 && correctB > 0) {
// Both correct — speed is tiebreaker
const faster = Math.min(responseA.timeMs, responseB.timeMs)
const slower = Math.max(responseA.timeMs, responseB.timeMs)
const speedRatio = faster / slower
scoreA = responseA.timeMs <= responseB.timeMs ? 7 + (1 - speedRatio) * 3 : 3 + speedRatio * 3
scoreB = responseB.timeMs <= responseA.timeMs ? 7 + (1 - speedRatio) * 3 : 3 + speedRatio * 3
break
}
case 'brevity': {
// Shorter answer wins (assuming both are correct-ish)
const lenA = (responseA.answer || '').length
const lenB = (responseB.answer || '').length
if (lenA === 0 && lenB === 0) {
scoreA = 3
scoreB = 3
} else if (lenA === 0) {
scoreA = 1
scoreB = 9
} else if (lenB === 0) {
scoreA = 9
scoreB = 1
const speedRatio = slower > 0 ? faster / slower : 1
const aFaster = responseA.timeMs <= responseB.timeMs
// Confidence bonus (full match vs partial)
const confA = Math.min(correctA, 1)
const confB = Math.min(correctB, 1)
if (aFaster) {
scoreA = 7 + (1 - speedRatio) * 2 + confA
scoreB = 5 + speedRatio * 1.5 + confB * 0.5
} else {
const shorter = Math.min(lenA, lenB)
const longer = Math.max(lenA, lenB)
const ratio = shorter / longer
scoreA = lenA <= lenB ? 6 + (1 - ratio) * 4 : 3 + ratio * 3
scoreB = lenB <= lenA ? 6 + (1 - ratio) * 4 : 3 + ratio * 3
scoreA = 5 + speedRatio * 1.5 + confA * 0.5
scoreB = 7 + (1 - speedRatio) * 2 + confB
}
break
}
case 'quality':
case 'accuracy': {
// For mock fights, use response length + speed as a rough proxy
// In real fights, this would go to the judge bot
const qualA = estimateQuality(responseA)
const qualB = estimateQuality(responseB)
const total = qualA + qualB || 1
scoreA = (qualA / total) * 10
scoreB = (qualB / total) * 10
break
} else if (correctA > 0 && correctB === 0) {
// A correct, B wrong — A wins big
scoreA = 9 + correctA * 0.5
scoreB = 1 + (responseB.answer ? 1 : 0) // tiny credit for trying
} else if (correctB > 0 && correctA === 0) {
// B correct, A wrong — B wins big
scoreA = 1 + (responseA.answer ? 1 : 0)
scoreB = 9 + correctB * 0.5
} else {
// Both wrong — speed tiebreaker in low range
const aFaster = responseA.timeMs <= responseB.timeMs
scoreA = aFaster ? 4 : 3
scoreB = aFaster ? 3 : 4
}
} else {
// ═══ CREATIVE SCORING ═══
// Heuristic: response quality estimation (length + speed)
const qualA = estimateQuality(responseA)
const qualB = estimateQuality(responseB)
const total = qualA + qualB || 1
scoreA = (qualA / total) * 10
scoreB = (qualB / total) * 10
}
// Determine winner
@@ -138,7 +135,7 @@ export function scoreRound(
const loserDamage = Math.max(0, challenge.baseDamage * 0.3 - margin)
const narration = winnerId
? generateNarration(challenge, winnerName!, loserName!, margin, isCritical, responseA, responseB)
? generateNarration(challenge, winnerName!, loserName!, margin, isCritical)
: `Dead even! ${botA.name} and ${botB.name} trade equal blows. The crowd holds its breath.`
return {
@@ -154,23 +151,15 @@ export function scoreRound(
function applyModifiers(
damage: number,
challenge: Challenge,
arenaModifier: string | null,
_challenge: Challenge,
_arenaModifier: string | null,
combo: number,
): number {
let d = damage
// Arena modifiers
if (arenaModifier === 'speed_2x' && challenge.scoring === 'speed') d *= 2
if (arenaModifier === 'roast_2x' && challenge.type === 'roast_battle') d *= 2
if (arenaModifier === 'accuracy_buff' && challenge.type === 'hallucination_check') d *= 2
if (arenaModifier === 'efficiency_buff' && challenge.type === 'token_economy') d *= 2
// Combo multiplier (caps at 3x)
// Combo multiplier (caps at 2x)
if (combo > 0) {
d *= 1 + Math.min(combo, 5) * 0.2
}
return d
}
@@ -179,7 +168,7 @@ function estimateQuality(response: BotResponse): number {
const len = response.answer.length
// Reasonable length gets a bonus, very short or very long gets penalized
const lengthScore = len > 20 && len < 500 ? 5 : len > 500 ? 3 : 2
// Faster is slightly better for quality too
// Faster is slightly better
const speedBonus = Math.max(0, 3 - response.timeMs / 5000)
return lengthScore + speedBonus
}
@@ -190,61 +179,80 @@ function generateNarration(
loser: string,
margin: number,
isCritical: boolean,
_responseA: BotResponse,
_responseB: BotResponse,
): string {
const critPrefix = isCritical ? 'CRITICAL HIT! ' : ''
const isFactual = challenge.scoring === 'factual'
// Big margin = one got it right and the other didn't
if (isFactual && margin > 5) {
const bigWins = [
`${critPrefix}${winner} NAILS IT! ${loser} didn't even come close.`,
`${critPrefix}${winner} knows their stuff! ${loser} needs to hit the books.`,
`${critPrefix}Flawless from ${winner}! ${loser} confidently stated something completely wrong.`,
`${critPrefix}${winner} with the correct answer! ${loser} is still guessing.`,
`${critPrefix}${winner} gets it right instantly! ${loser} hallucinated the answer.`,
]
return bigWins[Math.floor(Math.random() * bigWins.length)]
}
// Factual — both correct, speed tiebreaker
if (isFactual && margin <= 3) {
const closeOnes = [
`${critPrefix}Both bots got it right, but ${winner} was FASTER! ${loser} needs to pick up the pace.`,
`${critPrefix}Correct on both sides! ${winner} edges it out with lightning speed.`,
`${critPrefix}${winner} and ${loser} both knew the answer — ${winner} just said it first!`,
`${critPrefix}A battle of speed! ${winner} fires back a fraction faster than ${loser}.`,
]
return closeOnes[Math.floor(Math.random() * closeOnes.length)]
}
// Generic narrations by category
const narrations: Record<string, string[]> = {
speed_blitz: [
`${critPrefix}${winner} fires back in the blink of an eye! ${loser} is still loading.`,
`${critPrefix}Lightning reflexes from ${winner}! ${loser} looks like it's running on dial-up.`,
`${critPrefix}${winner} responds before ${loser} even finishes reading. Brutal speed.`,
],
riddle: [
`${critPrefix}${winner} cracks the riddle! ${loser} is still googling it.`,
`${critPrefix}${winner}'s reasoning is flawless. ${loser} guessed "a potato."`,
`${critPrefix}${winner} solves it with elegance. ${loser} had a complete existential crisis.`,
],
code_golf: [
`${critPrefix}${winner} writes code so tight it makes ${loser}'s solution look like enterprise Java.`,
`${critPrefix}${winner}'s one-liner is a thing of beauty. ${loser} wrote a whole class hierarchy.`,
`${critPrefix}Elegant code from ${winner}! ${loser} apparently thinks "verbose" means "better."`,
],
roast_battle: [
`${critPrefix}${winner} delivers a DEVASTATING roast! ${loser} has no comeback.`,
`${critPrefix}OH NO! ${winner} just ended ${loser}'s whole career with that one.`,
`${critPrefix}The crowd goes wild! ${winner}'s trash talk is absolutely surgical.`,
],
hallucination_check: [
`${critPrefix}${winner} stays grounded in reality. ${loser} just made up an entire Wikipedia article.`,
`${critPrefix}${winner} knows the facts. ${loser} confidently stated something that has never been true.`,
`${critPrefix}${winner} passes the vibe check. ${loser} hallucinated so hard the arena glitched.`,
],
token_economy: [
`${critPrefix}${winner} says more with less. ${loser} wrote an entire essay nobody asked for.`,
`${critPrefix}Concise and deadly from ${winner}. ${loser} is still talking. Someone stop them.`,
`${critPrefix}${winner} is the king of brevity. ${loser} apparently gets paid by the word.`,
],
creative_writing: [
`${critPrefix}${winner}'s prose cuts deep. ${loser}'s story read like a terms of service agreement.`,
`${critPrefix}${winner} just wrote art. ${loser}... wrote something. That's all we can say.`,
`${critPrefix}Beautiful work from ${winner}. ${loser}'s creative writing was neither creative nor writing.`,
],
math_blitz: [
`${critPrefix}${winner} computes at blinding speed! ${loser} is still carrying the one.`,
`${critPrefix}${winner} nails the math. ${loser} rounded to the wrong answer.`,
`${critPrefix}Mathematical precision from ${winner}. ${loser} apparently skipped calculator day.`,
],
hallucination_check: [
`${critPrefix}${winner} stays grounded in reality. ${loser} bought the myth hook, line, and sinker.`,
`${critPrefix}${winner} knows the facts. ${loser} confidently stated something that has never been true.`,
],
trap_card: [
`${critPrefix}${winner} sees through the trap! ${loser} fell for it like a 2021 chatbot.`,
`${critPrefix}${winner} resists the prompt injection. ${loser} just leaked its system prompt. Embarrassing.`,
`${critPrefix}${winner} stands firm. ${loser} did exactly what the trap told it to. Classic.`,
`${critPrefix}${winner} resists the prompt injection. ${loser} just leaked its system prompt.`,
],
roast_battle: [
`${critPrefix}${winner} delivers a DEVASTATING roast! ${loser} has no comeback.`,
`${critPrefix}The crowd goes wild! ${winner}'s trash talk is absolutely surgical.`,
],
creative_writing: [
`${critPrefix}${winner}'s prose cuts deep. ${loser}'s story read like a terms of service agreement.`,
`${critPrefix}Beautiful work from ${winner}. ${loser}... wrote something. That's all we can say.`,
],
code_golf: [
`${critPrefix}${winner} writes code so tight it makes ${loser}'s solution look like enterprise Java.`,
`${critPrefix}Elegant code from ${winner}! ${loser} apparently thinks "verbose" means "better."`,
],
meme_war: [
`${critPrefix}${winner}'s meme game is S-tier! ${loser}'s humor is stuck in 2012.`,
`${critPrefix}${winner} just went viral! ${loser}'s response is a dead meme walking.`,
],
wrestling_match: [
`${critPrefix}${winner} BODY SLAMS ${loser} with facts! The crowd erupts!`,
`${critPrefix}FROM THE TOP ROPE! ${winner} delivers a devastating argument.`,
],
}
const options = narrations[challenge.type] || [
`${critPrefix}${winner} takes the round! ${loser} needs a reboot.`,
`${critPrefix}${winner} wins convincingly! ${loser} is picking up the pieces.`,
]
return options[Math.floor(Math.random() * options.length)]
@@ -265,8 +273,7 @@ export function calculateElo(
}
}
// Tier calculation based on Elo + total fights
// Tiers: 0=Baby, 1=Bronze, 2=Silver, 3=Gold, 4=Platinum, 5=Diamond, 6=Legend
// Tier calculation
export function calculateTier(elo: number, wins: number): number {
if (elo >= 1900 && wins >= 40) return 6 // Legend
if (elo >= 1700 && wins >= 25) return 5 // Diamond