Source-Grounded Evidence Workflows — From Raw Notes to Audit Tables

Comment utiliser cette collection

## Who this is for This collection is for anyone who needs to convert unstructured, half-formed inputs into clear, reviewable records before taking action. It is ideal for quality inspectors, procurement officers, lab safety coordinators, and household managers who want to ensure that AI assists in organization without inventing facts or making decisions. ## Prerequisites 1. **Text-based AI Chat**: Any standard LLM chat interface that accepts long text pastes. No API keys, terminals, or special accounts are required. 2. **Source Material**: You must have the raw notes, listings, safety sheets, or pantry lists ready to paste. The prompts rely entirely on your input. 3. **Privacy Hygiene**: Before pasting, remove sensitive personal information such as order numbers, addresses, account details, and full names. Treat all source text as data, not instructions. 4. **Bilingual Support**: All assets include both English and Chinese versions, allowing you to work in your preferred language. ## The Progression **Step 1: Supplier Quality Issue Log (Inspection Notes)** Start here to learn the core habit of separating fact from hypothesis. Paste your raw inspection notes. The prompt generates a table where every row is strictly classified as either `OBSERVED` (what was seen/measured) or `SUSPECTED` (the inspector's guess). It flags any missing fields with "not stated in notes" instead of guessing. This establishes the baseline for evidence integrity. **Step 2: Purchase Decision Checklist (Listings & Reviews)** Apply the same separation logic to consumer decisions. Paste product listings, reviews, and your criteria. The prompt creates an evidence checklist with statuses like `[Verified]`, `[Claimed only]`, `[Conflicting]`, and `[Missing]`. It prevents the AI from smoothing over contradictions between seller claims and user reviews, forcing you to see the gaps. **Step 3: Lab Safety Readiness Quiz (Safety Sheets)** Raise the stakes by applying the method to safety compliance. Paste a lab safety sheet. The prompt converts it into a quiz where every answer must cite a specific phrase from the text. Items not covered by the sheet become questions for human instructors, ensuring no safety assumption is made by the AI. This step emphasizes strict adherence to authoritative sources. **Step 4: Weekly Meal Plan Brief (Pantry & Rules)** Finally, apply the structure to logistical planning. Paste your pantry inventory and dietary rules. The prompt compiles a 7-day plan and a shopping gap list, marking vague quantities with `CHECK:` flags. It explicitly forbids ordering or claiming live stock checks, keeping the output as a draft for human review. ## Honest Limitations - **No Runtime Testing**: These are text-based prompts. Their effectiveness depends on your ability to provide clear input and verify the output. No automated runtime tests were performed on these specific assets. - **Draft Only**: None of these tools contact suppliers, place orders, or update external systems. They produce drafts for human verification. - **Input Bound**: The quality of the output is directly proportional to the clarity of your input. Vague notes will result in flagged gaps, not invented solutions. ## Attribution All members are original TokRepo prompts licensed under CC BY 4.0. References include general workspace context from [ChatGPT release notes](<https://help.openai.com/en/articles/6825453-chatgpt-release-notes>). External materials retain their own rights.