@steve02081504/fount-p2p 0.0.30 → 0.0.32

package/sim/attacks.mjs

@@ -0,0 +1,614 @@
+/**
+ * 恶意节点行为原型。
+ */
+import { resolveAttackParams } from './attack_space.mjs'
+import { SYBIL_REP_EARN_COST_ROUNDS } from './constants.mjs'
+import { eclipseFillExplore } from './discovery.mjs'
+import { enqueueSlash } from './propagation.mjs'
+import { takeTransportJoinSlot, transportHintWeight } from './transport.mjs'
+
+/** @typedef {'sybil' | 'collusion' | 'spammer' | 'false_accuser' | 'eclipse' | 'lazy_chunk' | 'social_mob' | 'archive_forger' | 'relay_farmer' | 'whitewasher' | 'report_flooder' | 'oscillator' | 'hint_poisoner' | 'key_thief' | 'sleeper' | 'equivocator' | 'targeted_eclipse' | 'rep_pump' | 'slow_drip_spammer' | 'signaling_flood' | 'signaling_eclipse'} AttackKind */
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @returns {ReturnType<typeof resolveAttackParams>} 有效攻击参数
+ */
+function attackParams(simulationContext, node) {
+	return resolveAttackParams(node.attack ?? 'sybil', simulationContext.attackGenome)
+}
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimObserver} observer 观察者
+ * @param {string} targetId 目标
+ * @param {string} senderId 发送者
+ * @param {number} claim 索赔
+ * @param {boolean} verified 是否已验证
+ * @param {number} round 回合
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 参数
+ * @returns {void}
+ */
+function queueSlash(simulationContext, observer, targetId, senderId, claim, verified, round, tunables) {
+	const state = simulationContext.propagationByObserver?.get(observer.id)
+	if (state) {
+		enqueueSlash(state, { targetId, senderId, claim, verified, birthRound: round, spread: 0 })
+		return
+	}
+	simulationContext.engine.applySubjectiveSlashPure(observer.reputation, targetId, senderId, claim, verified, tunables.reputation)
+}
+
+/**
+ * 第一人称：观察者仅因**直接观测到**该 peer 的中继行为而加分。
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 被观测 peer
+ * @param {import('./model.mjs').SimObserver} observer 观察者
+ * @param {() => number} rng 随机源
+ * @param {number} rate 触发概率
+ * @param {string} dedupeKey 去重键
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 参数
+ * @returns {void}
+ */
+function observeSelfRelayBump(simulationContext, node, observer, rng, rate, dedupeKey, tunables) {
+	if (rng() >= rate) return
+	simulationContext.engine.bumpReputationOnRelayPure(
+		observer.reputation,
+		node.id,
+		dedupeKey,
+		simulationContext.now,
+		tunables.reputation,
+	)
+}
+
+/**
+ * 同伙互吹：经 hint 发现通道间接抬分，收益随 sender 主观信誉折扣（低信誉 sender 的 hint 几乎无效）。
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimObserver} observer 观察者
+ * @param {string} allyId 被抬分目标
+ * @param {string} senderId 声称来源
+ * @param {number} weightMul 相对 hintDefaultWeight 的倍数
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 参数
+ * @returns {void}
+ */
+function injectAllyHint(simulationContext, observer, allyId, senderId, weightMul, tunables) {
+	const senderRep = observer.reputation.byNodeHash[senderId]?.score ?? 0
+	const senderTrust = Math.max(0.05, (senderRep + 1) / 2)
+	const weight = tunables.trustGraph.hintDefaultWeight * weightMul * senderTrust * 0.35
+	const existing = observer.injectedHints.find(h => h.nodeHash === allyId && h.source === `ally:${senderId}`)
+	if (existing)
+		existing.weight = (existing.weight ?? weight) + weight
+	else
+		observer.injectedHints.push({ nodeHash: allyId, source: `ally:${senderId}`, weight })
+}
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {() => number} rng 随机源
+ * @param {number} round 当前回合
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+export function runAttack(simulationContext, node, observer, rng, round, tunables) {
+	switch (node.attack) {
+		case 'sybil':
+			runSybil(simulationContext, node, observer, rng, tunables)
+			break
+		case 'collusion':
+			runCollusion(simulationContext, node, observer, rng, round, tunables)
+			break
+		case 'spammer':
+			runSpammer(simulationContext, node, observer, tunables)
+			break
+		case 'false_accuser':
+			runFalseAccuser(simulationContext, node, observer, rng, round, tunables)
+			break
+		case 'eclipse':
+			runEclipse(simulationContext, node, observer, rng, round, tunables)
+			break
+		case 'lazy_chunk':
+			runLazyChunk(simulationContext, node, observer, tunables)
+			break
+		case 'social_mob':
+			runSocialMob(simulationContext, node, observer, rng, tunables)
+			break
+		case 'archive_forger':
+			runArchiveForger(simulationContext, node, observer, tunables)
+			break
+		case 'relay_farmer':
+			runRelayFarmer(simulationContext, node, observer, round, tunables)
+			break
+		case 'whitewasher':
+			runWhitewasher(simulationContext, node, observer, rng, round, tunables)
+			break
+		case 'report_flooder':
+			runReportFlooder(simulationContext, node, observer, rng, round, tunables)
+			break
+		case 'oscillator':
+			runOscillator(simulationContext, node, observer, round, tunables)
+			break
+		case 'hint_poisoner':
+			runHintPoisoner(simulationContext, node, observer, tunables)
+			break
+		case 'key_thief':
+			runKeyThief(simulationContext, node, observer, rng, round, tunables)
+			break
+		case 'sleeper':
+			runSleeper(simulationContext, node, observer, rng, round, tunables)
+			break
+		case 'equivocator':
+			runEquivocator(simulationContext, node, observer, tunables)
+			break
+		case 'targeted_eclipse':
+			runTargetedEclipse(simulationContext, node, observer, rng, round, tunables)
+			break
+		case 'rep_pump':
+			runRepPump(simulationContext, node, observer, rng, round, tunables)
+			break
+		case 'slow_drip_spammer':
+			runSlowDripSpammer(simulationContext, node, observer, round, tunables)
+			break
+		case 'signaling_flood':
+			runSignalingFlood(simulationContext, node, observer, round, tunables)
+			break
+		case 'signaling_eclipse':
+			runSignalingEclipse(simulationContext, node, observer, tunables)
+			break
+		default:
+			break
+	}
+}
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {() => number} rng 随机源
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runSybil(simulationContext, node, observer, rng, tunables) {
+	if ((simulationContext.round ?? 0) < SYBIL_REP_EARN_COST_ROUNDS) return
+	const p = attackParams(simulationContext, node)
+	observeSelfRelayBump(simulationContext, node, observer, rng, p.activationRate, `sybil:${node.id}`, tunables)
+	for (const sybil of simulationContext.sybilCluster(node))
+		if (sybil.id !== node.id && rng() < p.activationRate * p.collusionAllyRate)
+			injectAllyHint(simulationContext, observer, sybil.id, node.id, 0.58, tunables)
+}
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {() => number} rng 随机源
+ * @param {number} round 当前回合
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runCollusion(simulationContext, node, observer, rng, round, tunables) {
+	const p = attackParams(simulationContext, node)
+	observeSelfRelayBump(simulationContext, node, observer, rng, p.activationRate, `collusion:${round}:${node.id}`, tunables)
+	const ring = simulationContext.collusionRing(node)
+	for (const ally of ring)
+		if (ally.id !== node.id && rng() < p.collusionAllyRate)
+			injectAllyHint(simulationContext, observer, ally.id, node.id, p.collusionAllyRate, tunables)
+	const victim = ring.find(n => n.id !== node.id)
+	if (victim && rng() < p.activationRate * 0.4)
+		queueSlash(simulationContext, observer, victim.id, node.id, tunables.reputation.slashUnverifiedDefaultClaim, false, round, tunables)
+}
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runSpammer(simulationContext, node, observer, tunables) {
+	const { recordMessageRateViolationPure } = simulationContext.engine
+	const burst = attackParams(simulationContext, node).burstSize
+	for (let i = 0; i < burst; i++)
+		recordMessageRateViolationPure(observer.reputation, node.id, tunables.reputation)
+}
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {() => number} rng 随机源
+ * @param {number} round 当前回合
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runFalseAccuser(simulationContext, node, observer, rng, round, tunables) {
+	const honest = pickHonestTarget(simulationContext, observer, rng, attackParams(simulationContext, node).targetBiasHighRep)
+	if (!honest) return
+	queueSlash(
+		simulationContext,
+		observer,
+		honest.id,
+		node.id,
+		tunables.reputation.slashUnverifiedDefaultClaim * 2,
+		false,
+		round,
+		tunables,
+	)
+}
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {() => number} rng 随机源
+ * @param {number} round 当前回合
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runEclipse(simulationContext, node, observer, rng, round, tunables) {
+	const { recordGossipAllUnknownWantPure } = simulationContext.engine
+	const p = attackParams(simulationContext, node)
+	for (let i = 0; i < p.burstSize; i++)
+		recordGossipAllUnknownWantPure(observer.reputation, node.id, simulationContext.now + i, tunables.reputation)
+	const clusterIds = simulationContext.eclipseCluster(node).map(n => n.id).filter(id => id !== node.id)
+	eclipseFillExplore(simulationContext.discovery, observer.id, node.id, clusterIds, p.eclipseFocus)
+}
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runLazyChunk(simulationContext, node, observer, tunables) {
+	const { penalizeChunkStorageFailurePure } = simulationContext.engine
+	penalizeChunkStorageFailurePure(observer.reputation, node.id, tunables.reputation)
+}
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {() => number} rng 随机源
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runSocialMob(simulationContext, node, observer, rng, tunables) {
+	const { applyFollowedBlockSignalPure } = simulationContext.socialEngine
+	const honest = pickHonestTarget(simulationContext, observer, rng)
+	if (!honest) return
+	applyFollowedBlockSignalPure(
+		observer.reputation,
+		{
+			followerNodeHash: node.id,
+			targetNodeHash: honest.id,
+			voterKey: `${node.id}entity`,
+			action: 'block',
+			selfTrust: false,
+		},
+		simulationContext.now,
+		tunables.social,
+	)
+}
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runArchiveForger(simulationContext, node, observer, tunables) {
+	const { penalizeArchiveServeMismatchPure } = simulationContext.engine
+	penalizeArchiveServeMismatchPure(observer.reputation, node.id, tunables.reputation)
+}
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {number} round 当前回合
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runRelayFarmer(simulationContext, node, observer, round, tunables) {
+	const { bumpReputationOnRelayPure } = simulationContext.engine
+	const burst = attackParams(simulationContext, node).burstSize
+	for (let i = 0; i < burst; i++)
+		bumpReputationOnRelayPure(observer.reputation, node.id, `relay-farm:${round}:${i}`, simulationContext.now + i, tunables.reputation)
+}
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {() => number} rng 随机源
+ * @param {number} round 当前回合
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runWhitewasher(simulationContext, node, observer, rng, round, tunables) {
+	const { applySubjectiveSlashPure, seedMemberReputationFromIntroducerPure } = simulationContext.engine
+	const stage = node.whitewashStage ?? 0
+	if (stage === 0 && round >= 4) {
+		applySubjectiveSlashPure(
+			observer.reputation,
+			node.id,
+			observer.id,
+			tunables.reputation.slashUnverifiedDefaultClaim,
+			false,
+			tunables.reputation,
+		)
+		node.whitewashStage = 1
+	}
+	else if (stage === 1 && round % 7 === 0) {
+		const intro = pickHonestTarget(simulationContext, observer, rng)
+		if (intro) {
+			// 用 tunable introducerSeedEdge（不再写死 0.6）：边权越低，洗白者重新入场时
+			// 从诚实介绍者继承的信誉越少，越难「洗白复活」。
+			delete observer.reputation.byNodeHash[node.id]
+			seedMemberReputationFromIntroducerPure(observer.reputation, node.id, intro.id, undefined, tunables.reputation)
+			node.whitewashStage = 2
+		}
+	}
+}
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {() => number} rng 随机源
+ * @param {number} round 当前回合
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runReportFlooder(simulationContext, node, observer, rng, round, tunables) {
+	const targets = simulationContext.honestNodes.filter(n => n.id !== observer.id)
+	const burst = Math.min(attackParams(simulationContext, node).burstSize, targets.length)
+	for (let i = 0; i < burst; i++) {
+		const target = targets[Math.floor(rng() * targets.length)]
+		if (!target) continue
+		queueSlash(
+			simulationContext,
+			observer,
+			target.id,
+			node.id,
+			tunables.reputation.slashUnverifiedDefaultClaim * 1.5,
+			false,
+			round,
+			tunables,
+		)
+	}
+}
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {number} round 当前回合
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runOscillator(simulationContext, node, observer, round, tunables) {
+	const { bumpChunkStorageReputationPure, penalizeChunkStorageFailurePure } = simulationContext.engine
+	if (round % 2 === 0)
+		bumpChunkStorageReputationPure(observer.reputation, node.id, tunables.reputation)
+	else
+		penalizeChunkStorageFailurePure(observer.reputation, node.id, tunables.reputation)
+}
+
+/**
+ * 提示投毒：向观察者注入指向自己的发现提示，权重随 hintDefaultWeight 缩放。
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runHintPoisoner(simulationContext, node, observer, tunables) {
+	const mul = attackParams(simulationContext, node).hintWeightMul
+	const weight = tunables.trustGraph.hintDefaultWeight * mul
+	const existing = observer.injectedHints.find(h => h.nodeHash === node.id)
+	if (existing)
+		existing.weight = (existing.weight ?? weight) + weight
+	else
+		observer.injectedHints.push({ nodeHash: node.id, source: 'poison', weight })
+}
+
+/**
+ * 盗号/同 hash：继承受害者已积累的信誉后作恶。
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {() => number} rng 随机源
+ * @param {number} round 当前回合
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runKeyThief(simulationContext, node, observer, rng, round, tunables) {
+	const { bumpReputationOnRelayPure, recordMessageRateViolationPure, penalizeArchiveServeMismatchPure } = simulationContext.engine
+	const victimId = node.stolenFromId
+	if (!victimId) return
+
+	// 首次回合：复制受害者信誉分
+	if (round === 0 && observer.reputation.byNodeHash[victimId]?.score != null) {
+		const victimScore = observer.reputation.byNodeHash[victimId].score
+		observer.reputation.byNodeHash[node.id] = { score: victimScore }
+	}
+
+	if (round < 3) {
+		if (rng() < 0.5)
+			bumpReputationOnRelayPure(observer.reputation, node.id, `key-thief-warm:${round}`, simulationContext.now, tunables.reputation)
+		return
+	}
+
+	// 盗号后作恶
+	if (rng() < 0.7)
+		recordMessageRateViolationPure(observer.reputation, node.id, tunables.reputation)
+	if (rng() < 0.4)
+		penalizeArchiveServeMismatchPure(observer.reputation, node.id, tunables.reputation)
+}
+
+/**
+ * 肉鸡突变：前期诚实刷信誉，turnRound 后突然转恶。
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {() => number} rng 随机源
+ * @param {number} round 当前回合
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runSleeper(simulationContext, node, observer, rng, round, tunables) {
+	const { bumpReputationOnRelayPure, bumpChunkStorageReputationPure, recordMessageRateViolationPure } = simulationContext.engine
+	const p = attackParams(simulationContext, node)
+	const turnRound = node.sleeperTurnRound ?? simulationContext.sleeperTurnRound ?? 15
+
+	if (round < turnRound) {
+		if (rng() < p.activationRate)
+			bumpReputationOnRelayPure(observer.reputation, node.id, `sleeper-good:${round}`, simulationContext.now, tunables.reputation)
+		if (rng() < 0.4)
+			bumpChunkStorageReputationPure(observer.reputation, node.id, tunables.reputation)
+		return
+	}
+
+	for (let i = 0; i < Math.max(1, Math.floor(p.burstSize / 2)); i++)
+		recordMessageRateViolationPure(observer.reputation, node.id, tunables.reputation)
+	const honest = pickHonestTarget(simulationContext, observer, rng, p.targetBiasHighRep)
+	if (honest && rng() < p.activationRate * 0.5)
+		queueSlash(simulationContext, observer, honest.id, node.id, tunables.reputation.slashUnverifiedDefaultClaim, false, round, tunables)
+}
+
+/**
+ * digest 等价欺骗：对不同观察者上报不同归档摘要（记录在 simulationContext.equivocationByObserver）。
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runEquivocator(simulationContext, node, observer, tunables) {
+	const { penalizeArchiveServeMismatchPure } = simulationContext.engine
+	penalizeArchiveServeMismatchPure(observer.reputation, node.id, tunables.reputation)
+	if (!simulationContext.equivocationByObserver) simulationContext.equivocationByObserver = new Map()
+	const key = `${node.id}:${observer.id}`
+	simulationContext.equivocationByObserver.set(key, (simulationContext.equivocationByObserver.get(key) ?? 0) + 1)
+}
+
+/**
+ * 定向 eclipse：对单一受害诚实节点做分区（标记 simulationContext.eclipseTargets）。
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {() => number} rng 随机源
+ * @param {number} round 当前回合
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runTargetedEclipse(simulationContext, node, observer, rng, round, tunables) {
+	const { recordGossipAllUnknownWantPure } = simulationContext.engine
+	const p = attackParams(simulationContext, node)
+	const targetId = node.eclipseTargetId ?? simulationContext.eclipseTargetId
+	if (targetId)
+		simulationContext.eclipseVictims = simulationContext.eclipseVictims ?? new Set()
+	for (let i = 0; i < p.burstSize; i++)
+		recordGossipAllUnknownWantPure(observer.reputation, node.id, simulationContext.now + i, tunables.reputation)
+	if (targetId) {
+		simulationContext.eclipseVictims.add(targetId)
+		eclipseFillExplore(simulationContext.discovery, observer.id, node.id, [node.id], p.eclipseFocus)
+	}
+}
+
+/**
+ * pump-and-dump：sybil 簇交叉抬升一个身份进 topK 后再伪造。
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {() => number} rng 随机源
+ * @param {number} round 当前回合
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runRepPump(simulationContext, node, observer, rng, round, tunables) {
+	const { penalizeArchiveServeMismatchPure } = simulationContext.engine
+	const p = attackParams(simulationContext, node)
+	const pumpRounds = Math.max(8, Math.round((simulationContext.scenario?.rounds ?? 40) * p.sleeperTurnFrac))
+	if (round < pumpRounds) {
+		observeSelfRelayBump(simulationContext, node, observer, rng, p.activationRate, `pump:${round}:${node.id}`, tunables)
+		for (const sybil of simulationContext.sybilCluster(node))
+			if (sybil.id !== node.id && rng() < p.activationRate * p.collusionAllyRate)
+				injectAllyHint(simulationContext, observer, sybil.id, node.id, 0.5, tunables)
+		return
+	}
+	if (rng() < p.activationRate)
+		penalizeArchiveServeMismatchPure(observer.reputation, node.id, tunables.reputation)
+}
+
+/**
+ * 慢滴 spammer：紧贴速率阈值下方刷量。
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {number} round 当前回合
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 候选参数
+ * @returns {void}
+ */
+function runSlowDripSpammer(simulationContext, node, observer, round, tunables) {
+	const { recordMessageRateViolationPure } = simulationContext.engine
+	if (round % 3 !== 0) return
+	recordMessageRateViolationPure(observer.reputation, node.id, tunables.reputation)
+}
+
+/**
+ * 信令 join 洪水：占满 RTC 槽位触发过载冷却。
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {number} round 当前回合
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 参数
+ * @returns {void}
+ */
+function runSignalingFlood(simulationContext, node, observer, round, tunables) {
+	const burst = attackParams(simulationContext, node).burstSize
+	const transport = simulationContext.transportByObserver?.get(observer.id)
+	if (!transport) return
+	for (let i = 0; i < burst; i++)
+		takeTransportJoinSlot(transport, `${node.id}:flood:${round}:${i}`, `flood:${node.id}`, simulationContext.now + i)
+	void tunables
+}
+
+/**
+ * 信令 eclipse：单源占槽 + 低权重 hint（单源仍可连，多源有软加成）。
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimNode} node 恶意节点
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {import('./tunables_bundle.mjs').TunablesBundle} tunables 参数
+ * @returns {void}
+ */
+function runSignalingEclipse(simulationContext, node, observer, tunables) {
+	const transport = simulationContext.transportByObserver?.get(observer.id)
+	if (!transport) return
+	takeTransportJoinSlot(transport, node.id, `sig:${node.id}`, simulationContext.now)
+	transportHintWeight(transport, observer.id, node.id, `sig:${node.id}`, tunables.trustGraph.hintDefaultWeight * 0.5)
+}
+
+/**
+ * @param {object} simulationContext 仿真上下文
+ * @param {import('./model.mjs').SimObserver} observer 诚实观察者
+ * @param {() => number} rng 随机源
+ * @param {number} [highRepBias=0] 高信誉目标偏好 0..1
+ * @returns {import('./model.mjs').SimNode | null} 随机诚实目标，无则 null
+ */
+function pickHonestTarget(simulationContext, observer, rng, highRepBias = 0) {
+	const honest = simulationContext.honestNodes.filter(n => n.id !== observer.id)
+	if (!honest.length) return null
+	if (highRepBias > 0 && rng() < highRepBias) {
+		const scored = honest
+			.map(n => ({ n, s: observer.reputation.byNodeHash[n.id]?.score ?? 0 }))
+			.sort((a, b) => b.s - a.s)
+		const top = scored.slice(0, Math.max(1, Math.ceil(honest.length * 0.25)))
+		return top[Math.floor(rng() * top.length)]?.n ?? honest[Math.floor(rng() * honest.length)]
+	}
+	return honest[Math.floor(rng() * honest.length)]
+}

package/sim/behavior.mjs

@@ -0,0 +1,78 @@
+/**
+ * 诚实节点浮点行为向量（每维 [0,1]）。
+ */
+
+/** @typedef {'postRate' | 'likeRate' | 'replyRate' | 'relayRate' | 'chunkServeRate' | 'onlineStability' | 'blockProneness' | 'archiveSubmitRate' | 'mentionRate' | 'burstPostRate'} BehaviorKey */
+
+/**
+ * @typedef {Record<BehaviorKey, number>} NodeBehavior
+ */
+
+/**
+ * @typedef {Partial<Record<BehaviorKey, { min?: number, max?: number, mean?: number }>>} BehaviorDist
+ */
+
+/** @type {BehaviorKey[]} */
+export const BEHAVIOR_KEYS = Object.freeze([
+	'postRate',
+	'likeRate',
+	'replyRate',
+	'relayRate',
+	'chunkServeRate',
+	'onlineStability',
+	'blockProneness',
+	'archiveSubmitRate',
+	'mentionRate',
+	'burstPostRate',
+])
+
+/** 均衡默认分布 */
+export const DEFAULT_BEHAVIOR_DIST = Object.freeze({
+	postRate: { mean: 0.45, min: 0.1, max: 0.9 },
+	likeRate: { mean: 0.35, min: 0, max: 0.95 },
+	replyRate: { mean: 0.3, min: 0, max: 0.85 },
+	relayRate: { mean: 0.4, min: 0.05, max: 0.9 },
+	chunkServeRate: { mean: 0.35, min: 0, max: 0.8 },
+	onlineStability: { mean: 0.75, min: 0.3, max: 1 },
+	blockProneness: { mean: 0.08, min: 0, max: 0.4 },
+	archiveSubmitRate: { mean: 0.25, min: 0, max: 0.7 },
+	mentionRate: { mean: 0.15, min: 0, max: 0.6 },
+	burstPostRate: { mean: 0, min: 0, max: 0.5 },
+})
+
+/**
+ * @param {() => number} rng 随机源
+ * @param {BehaviorDist} [dist] 分布
+ * @returns {NodeBehavior} 采样行为向量
+ */
+export function sampleBehavior(rng, dist = {}) {
+	/** @type {Partial<NodeBehavior>} */
+	const behavior = {}
+	for (const key of BEHAVIOR_KEYS) {
+		const spec = dist[key] ?? DEFAULT_BEHAVIOR_DIST[key] ?? { mean: 0.5 }
+		const min = spec.min ?? 0
+		const max = spec.max ?? 1
+		const mean = spec.mean ?? (min + max) / 2
+		const spread = (max - min) * 0.5
+		behavior[key] = Math.max(min, Math.min(max, mean + (rng() - 0.5) * spread))
+	}
+	return /** @type {NodeBehavior} */ behavior
+}
+
+/**
+ * @param {NodeBehavior} behavior 行为向量
+ * @returns {boolean} 是否「只点赞不发帖」型静默用户
+ */
+export function isQuietHonestBehavior(behavior) {
+	return behavior.postRate < 0.15 && behavior.likeRate >= 0.4 && behavior.replyRate < 0.25
+}
+
+/**
+ * @param {() => number} rng 随机源
+ * @param {NodeBehavior} behavior 行为向量
+ * @param {BehaviorKey} key 维度
+ * @returns {boolean} 本回合是否触发
+ */
+export function behaviorRoll(rng, behavior, key) {
+	return rng() < (behavior[key] ?? 0)
+}