INPUT · BRAND BRIEF
真实投放约束
Real constraints
预算、目标 CAC、KOL/KOC 候选池、素材、品类、品牌安全和投放周期。
Example: 300k RMB / 14 days
Brands do not lack dashboards. They lack confidence before choosing the creator mix, creative angle, budget split, and day-3 adjustment rule.
Brands submit campaign briefs. Miners submit media-buying agents. Validators run hidden simulations and reward the strategies that reduce regret.
预算、目标 CAC、KOL/KOC 候选池、素材、品类、品牌安全和投放周期。
每个矿工提交自己的投放 Agent: 选人、配素材、分预算、设止损和加码规则。
Validator 在隐藏市场里跑反事实测试,不按话术评分,只看策略表现。
Oransim is not the product story. It is the technical proof that validators can run auditable campaign simulations: causal graph, virtual consumers, counterfactual do(), and 14-day rollout.
pre-launch ranking before media spend
swap KOL, budget, creative, platform
campaign path under intervention
causal graph nodes in the OSS engine
Creator graph, fan overlap, price, content fit, trust signals.
Creative angle matched to segment, creator voice, and platform context.
Explore first, then concentrate; stop-loss and add-budget rules.
CTR 弱、CAC 高、评论变差时,Agent 必须给出下一步。
Data miners contribute reusable market evidence. Validators do not trust it directly; they quarantine, deduplicate, time-decay, and reward only data that improves holdout prediction.
达人画像、笔记表现、粉丝结构、报价、类目趋势、竞品活动。
去重、溯源、异常检测、时间衰减,防止垃圾数据污染沙盘。
数据让 simulation 更接近隐藏 holdout campaign,才获得奖励。
The validator can rotate hidden worlds, perturb creator fatigue, competitor launches, traffic shifts, and negative-comment shocks.
品牌为一次私有 brief 的 Agent 竞赛和风险报告付费。
购买已验证的品类打法: serum launch, acne-care KOC, premium trust building.
平台、代理商、数据团队可为新鲜可验证市场资料获得奖励。
代理商和增长平台把 preflight 作为投放前标配能力。
A subnet for competing media-buying agents, not a generic marketing dashboard.
Oransim-style simulation makes hidden validator scoring technically credible.
Brands pay because one avoided bad campaign can fund many preflights.