v0.3.0

🎉 重大更新：单线转生重构

「深算 DeepCompute」核心玩法重构：从「按转生切换时代」改为「按模型推进时代」，回归经典挂机游戏的转生范式。

✨ 核心变化

• 单条连续线：模型 12 个顺序解锁（写诗机器人 → … → Transformer → … → 神级模型）；硬件 / 电力 / 员工各 9 档合并进一个商店
• 时代 = 进度标签：练出多模态进入「大模型时代」、练出通用智能体进入「AGI 时代」……只改名字与氛围，不再换内容
• 转生回到起点：练出 Transformer 即可转生，全部清零重开，但保留科技树 / 范式商店 / 范式点
• 练得越深 PP 越多：转生后继续冲多模态 / AGI / ASI，累计 FLOPs 越多，转生收益越高
• 软墙曲线：不积累永久加成越往后越慢；每轮转生的永久加成让软墙逐轮后移

🆕 本版本其它系统

• 模型推理持续消耗算力（算力不足时收入软降级）
• 训练消耗研究点 + 算力（双资源门槛）
• 员工工资与罢工机制
• 出售算力库存、可调烧库存倍率
• 设备研发（develop 解锁 + upgrade 升级）
• 状态栏 3×2 网格布局 + 供电不足 / 算力不足醒目警告

⚠️ 破坏性变更

• 存档升级为 v2，旧存档作废重开

📦 安装

```bash
dsh plugin --profile web add github:cipher2026/dsh-idle-deepcompute
```

🎉 Major update: single-line prestige redesign

DeepCompute is rebuilt around a classic idle-prestige loop: the "era" is now a progress label driven by which model you've trained, not a separate content set.

✨ Highlights

• One continuous line: 12 models unlocked in sequence, plus 9 hardware / 9 power / 9 staff tiers in a single shop
• Era = progress label: train Multimodal → "LLM era", train AGI → "AGI era"
• Prestige resets you back to the start (keeps tech tree, paradigm shop, PP)
• Deeper runs → more PP: keep pushing past Transformer for bigger prestige rewards
• Soft-wall curve: without permanent bonuses progress slows sharply; each prestige pushes the wall further

⚠️ Breaking

• Save version bumped to v2; old saves reset

📦 Install

```bash
dsh plugin --profile web add github:cipher2026/dsh-idle-deepcompute
```

v0.2.0

[0.2.0] - 2026-08-17

Added

• Sell compute stock: sell <amount|all> [money|rp]
• Adjustable stock-burn multiplier: burn <1-20>
• Model version upgrades: trained models can be repeatedly trained to higher versions (v1 → v2 → v3…), cost ×2.5/level and income ×2/level
• Staff wages and a strike mechanic: fire <staff-id> to dismiss staff and reduce wages
• New research-point (RP) uses: research rush rush, inspiration injection inspire, publish paper paper
• Hardware and power-equipment R&D: develop <id> to unlock, upgrade <id> to level up (+50% output/supply per level)
• Shop now shows each hardware's power draw and each staff member's wage
• UI switched to a right-side "深算" tab with a slide-out panel (no longer relies on sidebar DOM injection)

Changed

• Training now consumes compute at the production rate over time (no longer instant), with training slots limiting parallel jobs
• Compute stock can be burned as fuel to speed up training, or sold for money / research points