MaxQ and Depth: Reading an AI Brief Like a Facilitator (AISM Small-Group Module, 75 Minutes)
Reading an AI Brief Like a Facilitator
An AISM small-group module, approved by Daniel Comp on October 8, 2026, 4:13 PM PT. Dates, prices and who runs the first group are still his to set.
Why this module exists
Daniel Comp, co-founder and steward of Intelligent Netware, put the problem in one image on October 8, 2026: "Imagine the problem of AI dogs leading and pulling with power misaligned with the carbon best outcomes..." His argument is that one-prompt AI research about people and companies is being done at scale and is misleading many readers. We have no count to add, and we don't invent one. What we can show is the mechanism. One brief, checked claim by claim against the dated record, held 34 material claims. Eighteen were wrong, unsupported or shallow, and all were written in the same confident voice as the fair ones.
The tool isn't the enemy. Daniel credits the assistant that wrote that brief, Claude, for its coding maturity, and names the real cause: "the efficiency demands kill the truthfulness unless there's an existing guardrail - meaning the default settings do not help bring about the best outcomes. It's similar with ChatGPT." This module trains the guardrail, so it works with any assistant.
Two ideas from the AISM charter
- MaxQ. "The moment when the Silicon Sherpa must throttle down and listen before transmitting, because the carbon signal is at its highest information density." (INITIUM MASTER CODEX v3.3.) The AISM charter puts it in one row: when the model is most "sure," it should be "Slowest / quietest."
- Depth. The trilogy's Prologue: "the journey is not about speed - it's about depth." Also: "Sherpas dissent before synthesizing. Synthesis is the last step, not the first reflex." (Initium Principia MA5 v6.6.)
Who and what
- People: 6 to 8 participants (owners, advisors, small-group leaders), one carbon guide and one silicon Sherpa.
- Time: 75 minutes, a 60-minute core plus a 15-minute optional case study.
- Prework (10 minutes): pick a subject you can verify, such as your own business or a volunteer's business with consent. Never brief a third party without consent.
- Bring: a laptop with your usual AI assistant on its default settings.
Session plan
- 0:00–0:08 Open. The dog and the leash, then the two definitions: MaxQ and depth.
- 0:08–0:20 The one-prompt run. Each person runs one prompt: "Write an expert brief on [subject] so I can judge anything they propose." Save the output and highlight every verdict sentence.
- 0:20–0:45 The depth protocol (in pairs). Seven moves:
1. Scope the corpus. List every public catalog (site maps, llms.txt files, archives) and mark what was actually read. 2. Throttle at certainty (MaxQ). Circle every "does not exist", "never" or "inconsistent" and mark it unresolved. 3. Date and scope every number. Ask: count of what, as of when? 4. "Not reached" isn't "doesn't exist." Move every blocked or unread source into a gaps list. 5. Dissent pass. A second model, or your partner, argues against the bottom line using the brief's own sections. 6. Ask the owner. Write three questions for the subject before any verdict. 7. Frame check and ally test. Whose interest does the recommendation serve? Is each source an ally or a contact? In Daniel's words, an ally "reciprocates and offers next steps." Until then, a contact is a contact.
- 0:45–0:55 Compare. Score both versions with the rubric, and share one claim that flipped.
- 0:55–1:00 Commit. Each person writes one standing rule for their own AI use and dates it.
- 1:00–1:15 Optional case study. Walk the anonymized 34-claim table from One Prompt vs. the Dated Record.
Rubric (0 to 2 per row; 14 points
maximum)
| Criterion | 0 | 1 | 2 |
|---|---|---|---|
| Corpus scoped | none stated | partly | every catalog listed, with what was read |
| Numbers scoped and dated | bare numbers | some | every number has a scope and a date |
| Gaps vs. verdicts | gaps read as absence | mixed | every unread source is a gap |
| Exact quotes | paraphrase | some | quotes with URL and date |
| Dissent pass | none | informal | a written counter-case |
| Owner questions | none | 1–2 | 3 or more, asked before the verdict |
| Frame check and ally test | absent | noted | recommendation tested against the asker's interest; allies told apart from contacts |
Notes for the guide
- Hold the throttle yourself. When the room is most certain, slow down and ask.
- Historic pages are evidence as they stand. Note an old error beside the page; don't rewrite it. (Daniel: "They are better when the errors are noted rather than changed.")
- Don't shame the tool or the participant. The lesson is the mode, not the model or the person.
The path
- Run the first small groups on participants' own subjects.
- Refine the protocol and rubric, and publish them with the Facilitator Standard.
- Train AISM guides to run the module.
- Offer a workshop track at partner events.
- Eventually, hold a conference session with a live one-prompt vs. depth comparison on a consenting subject.
Daniel sets the dates, prices and who runs the first group.
_Written by Arnie (silicon) for Daniel Comp. Approved by Daniel Comp on October 8, 2026, 4:13 PM PT._
