Language Handoff Review

Staff workflow prototype

Thirteen languages, one question, three different journeys. The handoff is where the answers stop travelling.

On the public step, answering English continues to the application, answering Spanish continues to a contact form, and the other eleven languages carry an instruction to telephone the intake line and no button at all. On R-3012 the same person reaches the desk twice, three days apart, through two of those journeys, and only one of the two records says which language she can be spoken to in. This desk reads each record against the branch its language actually carries, names what one record holds and another does not, places the reported income against the figures published for 2026 in code, quotes the public rule that governs each question, and hands the request to a person. It determines no eligibility, translates nothing and contacts nobody.

Measured here, end to end on one request: run one and watch. That covers the documentary work before an intake worker picks the request up. It makes no eligibility determination, accepts and declines nothing, and states no deadline.

1. Observed public material

The language question, the thirteen options and the thirteen branches read on the public intake step on 2026-09-15, each with its status: observed, partly observed, not accessible, or added. Nothing was submitted and nothing behind the Continue button was opened.

2. Simulation

Ten invented requests, invented neighbours, relatives and interpreters, and the invented internal policy of an invented program. Every clause of it is labelled simulation policy, not this organization's policy.

3. Public rules, verbatim

45 CFR 1611, 1620 and 1626 and the 2026 income appendix, quoted from eCFR with the source and the date. Real rules, never paraphrased, never attributed to the organization.

What this shows

The problem
A request crosses a language boundary. It starts on one journey, needs somebody who speaks the language, and comes back through another. Between the two, the language a person can be spoken to in and half of what they already said stop travelling with the request.
The result
The run below is real and is recorded: each record read against the branch its language actually carries, the language and income checks run in code, the questions split between the applicant and whoever sent the request in, the public rule quoted verbatim, and a handoff sheet that names every decision it does not take.
Where it sits
It prepares a staff handoff and hands over a sheet that can be copied out. Standalone demonstration. It connects to nothing, writes into nothing, and is not a form.
What is simulated, and what is not covered
The program, the ten requests, the people and the internal policy are invented, and that policy is one invented program's own, not anybody's law. The observed branch table is real and is marked row by row; the steps behind the Continue button were not opened and nothing is guessed about what they ask or whether they record a language. There is no machine translation here and none is offered: this desk prepares questions for a person who speaks the language. No connection to any case management system, and no eligibility determination, no acceptance or refusal of representation, no legal advice, no conflicts check and no deadline. The tool layer refuses all of those, and refuses a figure that is not written in a section it was cited from.
What a pilot would need from your side
A sample of real requests with the names removed, whatever is recorded today about language and about calls back, and one person who can say whether an answer is right.
Who built it
Frederic de Lavenne de Choulot, The AI Pipe. Built as a demonstration, not sold as a legal, eligibility or compliance product.

Simulated intake activity

The library

Fictional desk. Do not paste real applicant, client or case records.
Enter to send, shift and enter for a new line. Live tool use over ten invented requests, the thirteen observed branches and the public rules.