The escalation reason is computed, then it stops in the log

The seam

Your own post states it:

For our final approach, we built a single multi-task model that predicts three things at once: Escalation decision (3-way classification) Escalation reason (8 categories) Guideline citations (which guidelines to cite)

C-2 · SOURCED · fetched 2026-08-12

The decision reaches the customer. The reason and the citation stop in the log. At 97% reason accuracy and 98.7% citation AUC, those outputs are good enough to hand to a colleague. This memo is about handing them over.

The contract

Five fields, rendered at the moment of escalation instead of a scrollback: what Fin tried, why it stopped, confidence, what the customer has been told, what not to repeat. One of them is yours already.

When escalating due to a guideline match, we cite the specific guideline.

C-9 · SOURCED · fetched 2026-08-12

That field arrives as fact, with an id and a fetch date behind it. The other four are designed, and the card says so.

When nothing flags it

A contract that only works when the model raises its hand is a courtesy. In the second prototype Fin does not escalate, answers wrongly, and nothing flags it. The field that earns its place there reports an absence: nothing tested the diagnosis, and no transcript turn can say so.

What it costs

Filling in why it stopped, beyond the citation you already compute, means writing escalation guidance. Your documentation prices that.

The more escalation guidance you add, the more often Fin escalates.

C-10 · SOURCED · fetched 2026-08-12

Broader guidance leads to more escalations, fewer resolutions, and higher human support volume.

C-11 · SOURCED · fetched 2026-08-12

The corpus is also bounded.

You can have up to 100 active guidance rules at a time. If you try to activate more, you'll see an 'invalid parameters' error.

C-12 · SOURCED · fetched 2026-08-12

So this spends a capped resource and costs resolutions. Measure first: time from take-over to first reply, against a transcript-only baseline on the same queue. If the number does not move, withdraw the contract.

What this does not propose

Preview, Batch Test, Procedure Simulations, guidance authoring assistance and post-publish analytics already exist. They test the rule being written. This is about the conversation after one has fired.

Why a contract rather than an assistant

pass^k means success every time across k repeats, a hard bar for anything generative at a handover. So this surface generates nothing: every field is computed or declared designed, and the card says which.

Allowing an agent to take long horizon decisions bounded only by the availability of tools is a recipe for unreliable behaviour.

C-15 · SOURCED · fetched 2026-08-12

A briefing that summarises is a generation step. Five named fields is a rendering step. The contract page specifies all five, including what happens when each is missing.