# Commissioner AI Field Kit — presenter guide Prepared November 8, 2026, U.S. Central, for Doug Liles. Independent Good Combinator pilot; no FMCA endorsement. This guide supplements the November 7 plugin starter kit rather than replacing or declaring that kit installed. ## Five-minute demonstration **0:00–0:40 — The purpose.** “Choose one recurring district task. The goal is a useful, traceable draft and less total staff effort, with people reviewing the facts and making decisions.” **0:40–1:55 — Board Packet Briefing.** Open the synthetic documents and their scripted briefing. Show the revised quote: $62,000 plus $2,000 installation, totaling $64,000 before unknown tax. Annual operations are $7,500, operating funding is unspecified, and Exhibit B is missing. Show the evidence references. No recorded vote or paid invoice is supplied. The screen is a prewritten illustration, not a live AI result. **1:55–3:10 — Crossroads.** Use the separate synthetic equipment example. Purchase is $42,000 plus $3,000 annual maintenance for three years: $51,000. Lease is $18,000 annually: $54,000. Under this demonstration's simplified first-year cash rule, purchase needs $45,000 and lease $18,000. Change the available first-year cash from $45,000 to $30,000. Purchase no longer fits without changing the financing assumptions; the three-year totals stay unchanged. Tax, financing, resale, and operational differences are excluded and remain decision inputs. **3:10–4:00 — Take it back.** Choose a task, download the matching synthetic sample, and copy the self-contained prompt. Try it in your own ChatGPT account subject to available features and limits. This route requires no custom plugin installation. This website makes no AI model calls. API billing is separate. Plugin availability, upload permissions, and supported surfaces vary. **4:00–5:00 — A measurable pilot.** Name a staff owner and pick comparable tasks. Count all time, including gathering, drafting, review, corrections, and approval. Review numbers and source references, record errors and missing information, and decide at day 30 whether to continue, adjust, or stop. ## Thirty-day pilot checklist - Choose one recurring task, staff owner, reviewer, and authorized records. - Complete one synthetic example before using district records. - Record baseline task volume and total staff minutes for comparable tasks. - Set a review-inclusive time target and an accuracy threshold before starting. - Record every output's source correctness, arithmetic, missing evidence, corrections, and review time. - Calculate net time saved as baseline comparable-task minutes minus pilot minutes, including review and correction effort. Do not treat faster drafting alone as staff savings. - Stop or revise if material unsupported claims persist or review effort erases savings. - Set an actual day-30 review date when the pilot starts; this guide does not schedule a reminder. ## Readiness boundaries Five packages remain review candidates: District Ops, Mosquito Intelligence, Board Packet Briefing, Meeting Rules Practice, and Crossroads. Existing package checks and assistant rehearsals do not establish successful outside-account installation. Public release, actual mobile/voice use, and independent installed-host testing remain open. Scientific and parliamentary review remain pending for their relevant candidates. Public meeting address: https://fl5ai.com/commissioners. The QR handout points to this address. No colleagues have been contacted and no districts have been enrolled. ## Provenance - November 7 Commissioner Plugin Implementation Results and Pilot Testing Instructions, read November 8. - Existing Commissioner AI Pilot One Pager, read November 8. - [Official plugin availability and permissions](https://help.openai.com/en/articles/20001256-plugins-in-chatgpt), checked November 8. - [ChatGPT and API billing](https://help.openai.com/en/articles/9039756-managing-billing-for-chatgpt-and-the-api-platform), checked November 8. Support: https://goodcombinator.ai/support ## Optional plugin installation route The immediate take-home prompts require no plugin installation. For the separate local ZIP pilot, use Work mode or Codex in the ChatGPT desktop app. Open the intact commissioner-pilot root, restart the app, find Commissioner Pilot in the Plugins Directory, and install the exact candidate. Do not assume CLI catalog commands prove installation. Phone-only users should use the prompt route until this local ZIP route is tested on a supported desktop. Crossroads pilot source is v0.1.2; do not equate it with installed v0.1.4 or the differing v0.2.0 review archive. Freeze one exact archive and record its version/checksum before outside testing. The local pilot can proceed independently of public-directory publication or District Ops submission eligibility. Public-release review requirements depend on package contents. Skills-only packages still require applicable skill checks, but MCP-server test counts, server connection checks and demo recordings must not be represented as universal requirements. November 8 QA: three separate assistant prompt replays preserved the expected calculations and missing-evidence boundaries. These were not blinded tests, independent installed-host runs, scientific review, parliamentary approval, or actual iPhone tests. Public HTTP and boundary checks passed. Narrow-screen wrapping and complete print-prompt handling were corrected in source.