Case Proof Workshop
Status: Failed.
Your muse
Build a legal case agent like meta-models/legal-agent-workshop. It reads a fictional dispute file, drafts a checklist, proves each item with exact page quotes, asks me for evidence when it is stuck, and files a memo I sign off. Show every turn live.
From the owner: Stop the run.
Scouted the brief, inspo, and core skills. Now inspecting platform helpers and UI tokens before writing the plan.
The plan: Case Proof Workshop
Prove every claim with page quotes, then sign the memo.
- Upload a fictional dispute file in PDF or text
- Agent parses the file and lists parties, claims, and pages found
- Agent drafts a numbered proof checklist of what must be proved
- User approves or edits the checklist
See the full plan
Who it's for
Students and hobbyists practicing fictional dispute analysis.
How it works
- Upload a fictional dispute file in PDF or text.
- Agent parses the file and lists parties, claims, and pages found.
- Agent drafts a numbered proof checklist of what must be proved.
- User approves or edits the checklist.
- Agent works the checklist item by item, streaming every turn live.
- For each item the agent shows proved with exact page quotes or stuck with reason.
- For stuck items the agent asks a short evidence question.
- User types evidence or uploads a supporting note.
- Agent re tries the stuck item with the new evidence.
- Agent assembles a memo with proof table, quotes, gaps, and conclusion.
- User reviews, edits, and signs off the memo.
- Signed memo is saved and available after reload.
Screens
- Case Intake. Upload fictional dispute file, see parsed parties, claims, and page list.
- Checklist Board. Review, approve, edit, and reorder the proof checklist with status per item.
- Proof Workspace. Run proof item by item and see exact page quotes or stuck reasons.
- Evidence Q and A. Answer agent questions for stuck items and attach evidence to items.
- Memo Sign Off. Review memo draft with proof table, edit text, sign off, and download.
- Turn Log. Stream every agent turn live with steps, tool calls, and results.
What goes in
- PDF: Fictional dispute file, up to 20 pages and 4.5 MB. Text PDF preferred. Scanned PDF or photo also accepted.
- Text (optional): Short typed answers to agent evidence questions when an item is stuck, plus memo sign off note.
What comes out
- Text: Numbered proof checklist with proved or stuck status per item.
- Text: Proof card per item with exact quotes plus page numbers, or stuck reason plus user supplied evidence.
- Text: Filing memo draft with findings, quote table, gaps, and sign off record. Downloadable as text.
Models
Model Used for Why Muse Wonder Read dispute file, draft checklist, match exact page quotes, ask evidence questions, and draft memo. Only model that reads text, image, and PDF input with long context and returns structured JSON for proof and memo work. Limits
- For fictional practice disputes only. Not a lawyer and not legal advice for real cases.
- Page quotes come from the uploaded file text. Scanned pages need readable text or proof fails.
- When an item cannot be proved from the file, the agent marks it stuck and asks the user instead of guessing.
- Memo is a draft until the user signs off in the app. Nothing is filed anywhere else.
- All cases, checklists, and memos stay in this browser. They do not follow the user to another device.
- One run handles one dispute file up to about 20 pages. Large sets are handled by putting documents in Wonder context, not by embeddings.
From your inspo
Text only note with no references. Person wants a legal case agent that reads a fictional dispute, proves a checklist with page quotes, asks for help when stuck, and files a memo for sign off.
Takes
- Agent job that reads a dispute, makes a proof checklist, and proves each item
- Exact page quote proof pattern with stuck plus ask user flow
- Memo filing with sign off step
Cost per run: about $0.27 (est.). One run uses about 4 Wonder calls: parse 12k in plus 800 out, checklist 8k in plus 600 out, proof 20k in plus 1200 out, memo 10k in plus 1000 out. Input about 50k tokens at $4 per 1M is $0.20, output about 3.6k tokens plus reasoning at $20 per 1M is $0.072. Total about $0.27.
What you asked for 0 of 12 checked and working
- R1: Read an uploaded fictional dispute file and keep it after reload, with parsed parties, claims, and page list. Not checked yet
- R2: Draft a numbered proof checklist from the file that the user can approve and edit before proof starts. Not checked yet
- R3: Prove each checklist item with exact page quotes showing quote text plus page number, still there after reload. Not checked yet
- R4: When an item is stuck with no quote found, mark it stuck and ask the user for specific evidence, then retry with the answer. Not checked yet
- R5: File a memo built from the proved items that the user reviews, edits, and signs off, kept after reload. Not checked yet
- R6: Show every agent turn live in a streaming turn log with input, reasoning summary, tool call, and result as it happens. Not checked yet
- R7: Provide screens for Intake, Checklist Board, Proof Workspace, Evidence Q and A, Memo Sign Off, and Turn Log with clear navigation between them. Not checked yet
- R8: Show an empty state before any file exists explaining what fictional file to upload, and an error state when a file is too large or unreadable. Not checked yet
- R9: Allow editing and deleting checklist items, evidence answers, and memo drafts, with changes saved after reload. Not checked yet
- R10: Follow the meta-models/legal-agent-workshop agent pattern with explicit models steps for parse, checklist, prove, and memo visible in the turn log. Not checked yet
- R11: Agent reads the fictional dispute file page by page and keeps page numbers attached to every quote used for proof. Not checked yet
- R12: Keep multiple uploaded dispute files in a case list so each case files a memo separately and stays after reload. Not checked yet
This build didn't ship.
Stopped by you.
A broken product is never shown as working. Muse Code's notes are in its steps.