What decision does this actually settle?
Northline Benefits Technology is a fictional 720-person company with $8.4M in annual recurring revenue sitting in a legal queue. Two people — one general counsel, one contract manager — read the same liability, IP, privacy and termination language across 92 agreements a month. The model doesn't decide what Northline will accept in a negotiation. It decides which agreements out of the 92 actually need a lawyer's attention this week, and drafts the fallback language counsel would have approved anyway.
From an open queue to a routed decision
Read the case brief first
The four-stage story above the workbench — Monday's intake, Wednesday's mismatch, Friday's outside-counsel bill, and what the agent changes — is the actual scenario the numbers below are drawn from.
Open an agreement from the queue
Each row is one contract already scored against the playbook. Evergreen Regional Bank's MSA is the worked example loaded by default.
Read the flagged clause against the playbook
Section 9.2 (Liability) shows the source language, the playbook position it violates, and the reasoning — not just a red flag.
Apply the redline or send it to counsel
Safe variance gets a drafted fallback you can apply directly. Material exceptions route to a person — the model never signs anything.
Prepare the negotiation memo
Once decisions are made across the queue, this assembles the evidence trail — source clause, rationale, change, owner — for counsel's record.
What each part of the workbench is for
| Area | What it shows | Where the decision happens |
|---|---|---|
| Feed panel | The queue of agreements, each pre-scored against the playbook. | Pick which contract to open next — nothing here is auto-decided. |
| Detail panel | The specific clause, its source text, the playbook rule it trips, and a drafted fallback. | This is where "apply redline" or "send to counsel" actually happens. |
| Reset decisions | Clears every decision you've made in this session. | Use it to re-run the queue from a clean state, not to undo one contract. |
Three numbers, in order of usefulness
| Metric | What it means | What to do about it |
|---|---|---|
| ARR waiting in legal | $8.4M tied to agreements not yet cleared — the actual cost of the queue, not a headcount complaint. | Compare it against the $38,640/month outside-counsel spend already absorbing overflow. |
| Median turnaround | 7.4 business days per agreement across the current two-person team. | Track which clause types drive the tail, not the average. |
| Playbook variance | Whether a clause matches, deviates safely, or deviates materially from NB-LGL-2026.3. | Only material deviations need counsel; safe variance gets the drafted fallback. |
Three questions this actually settles
Sales marks nine as urgent and procurement marks four more — the model reads all 23 against the playbook first, so "urgent" and "actually material" stop being the same list by default.
Both used to wait in the same queue for the same reviewer. The playbook variance score is what tells them apart before either one is opened.
Run the queue with the model and compare how many agreements would have gone to counsel anyway versus how many the drafted fallback would have cleared.
Where this breaks if you push it too far
What goes in, what comes out
| Detail | |
|---|---|
| Input | A pre-loaded illustrative queue of agreements and playbook clauses — no upload, nothing leaves your browser. |
| Agent output | Clause classification, playbook variance score, drafted fallback language. |
| Stays with a person | Legal advice, final contract language, signature authority. |