Scenario Runner Demo
Pick a scenario, see the recorded result
An interactive view of the same precomputed, fixed-seed what-if library the project page cites. Nothing runs live here: every number below is traceable to a recorded CRN-paired run on the synthetic line. The first four tabs map to the agent's four what-if decision modules; the monitoring tab is a separate slow-drift detection demo, not a fifth what-if decision module.
選一個情境，看紀錄下來的分析結果
這裡展示的是專案首頁引用的同一批預先計算、固定 seed 的 what-if 情境庫。頁面不即時跑模擬：下面每個數字都能追溯到合成產線上一次 CRN 配對的紀錄 run。前四個分頁對應 agent 的四類 what-if 決策模組；劣化監控分頁是獨立的慢性劣化偵測示範，不是第五個 what-if 決策模組。
Which station should receive the next tool?先加哪一站的 tool？
Pick a station; the table compares whole-line impact and investment-stress net cost (illustrative rates, for ranking only) across all four options.選一個站點，表格同時比較四個選項的全線影響與投資壓力淨成本（示意費率，僅供排序）。| Station站點 | Added Lot-Processing Slots增加的 lot 處理槽數 | Avg Cycle-Time Reduction平均 cycle time 降幅 | Investment-Stress Net Cost Change投資壓力情境淨成本變化 |
|---|---|---|---|
| LITHO | |||
| FURNACE | |||
| DEPO | |||
| METRO |
Maintain now or wait?現在 PM 還是延後？
Pick a PM timing and compare its total cost against the alternatives.選一個 PM 時機，和其他選項比較總成本。The recorded library covers PM timing on LITHO, the engineered bottleneck. The underlying engine accepts a station parameter, but only the LITHO runs are published.情境庫目前收錄瓶頸站 LITHO 的 PM 時機 run。底層引擎支援指定站點，但公開的數字僅限 LITHO。
Which queue rule fits the objective?哪個 queue rule 符合目標？
Pick an operating objective; the winning policy is read from the same policy comparison runs.選一個營運目標，最佳 policy 讀自同一批 policy 比較 run。Does faster flow increase quality risk?加速 flow 是否增加 quality risk？
Flip between baseline and the LITHO +1 scenario; speed and yield risk share one view.在 baseline 和 LITHO +1 情境之間切換，速度和良率風險放在同一個畫面。The recorded library covers the LITHO +1 scenario: LITHO is the engineered bottleneck, so the speed-yield tension is sharpest there.情境庫目前收錄 LITHO +1 情境：LITHO 是設計上的瓶頸站，速度與良率的張力在這裡最明顯。
Which monitor catches slow drift in time?哪種監控來得及抓到慢性劣化？
Pick a monitoring rule and see when it alerts in the 160-day labeled backtest.選一種監控方式，看它在 160 天標註回測中何時發出警報。| Monitoring approach | Signal watched | Alert logic | Result |
|---|---|---|---|
| 監控方式 | 監控訊號 | alert 邏輯 | 結果 |
| Output-only monitor只看 output 的監控 | Daily output / throughput | Alert only when output drops below the clean baseline lower control limit.只有當 output 低於乾淨 baseline 的下控制線時才 alert。 | |
| EWMA drift monitorEWMA 漂移監控 | Daily median lot cycle time每日 lot cycle time 中位數 | Alert when small persistent cycle-time drift accumulates.小而持續的 cycle-time 偏移累積後 alert。 |
Observation: daily output remains noisy but ordinary, while daily median cycle time separates from the clean twin after degradation starts.
圖表觀察：daily output 看起來仍在正常波動中，但 daily median cycle time 在 degradation 開始後逐漸和 clean twin 分離。
Scope: this page is a bounded demo over a precomputed scenario library; it does not run new simulations or accept free-form parameters. The line is synthetic with fixed seeds; cost figures are illustrative assumptions used for ranking, not real-fab prices or savings forecasts. The memo preview uses a fixed template, not a live language model.
聲明：本頁是一個有界的 demo，展示預先計算好的情境庫，不會即時跑新模擬，也不接受自由輸入參數。產線為合成資料、固定 seed；成本數字為示意假設，僅用於相對排序，不是實廠價格或節省預測。memo 預覽使用固定模板，並非即時語言模型生成。