🐝

Put the hive to work on a real document

Each playbook gives you the exact text to enter into every WaxFrame field for a specific document type. Open WaxFrame on one screen and this page on another. Work through the fields top to bottom — replace anything in [brackets] with your own information.

Tip: Every playbook on this page is also available as a template inside the app. On the Project screen, click 📋 Use Template, pick whether you're starting from scratch or refining an existing draft, then click the template to populate all six Goal fields automatically — no copy/paste needed. Use this page when you want to read the full reasoning behind each playbook, and the in-app templates when you want to apply one fast.

New to WaxFrame? Read the User Manual first — it walks you through every screen step by step.

⭐ Quick Start

First time? Run this playbook. Universal, low stakes, converges fast, and walks you through every screen.

Quick Start — Chocolate Chip Cookies
New to WaxFrame? Run this playbook first. Universal, low stakes, converges fast, and walks you through every screen.

A small, low-stakes first project. Classic chocolate chip cookies — pantry ingredients, familiar territory for almost everyone — written from scratch by the hive. Same complete WaxFrame flow as any other document; the topic is just harmless enough that you can focus on learning the workflow without anything riding on the outcome. The goal fields below are written in plain human voice, the way you'd actually describe what you want to a friend. With a well-phrased goal, the hive typically converges in 2 rounds.

Setup — Step 3 of 5 — Your Project screen

Copy each value into the matching field on the Project screen. Fields marked * are required to continue.

FieldWhat to enter
Project name *Chocolate Chip Cookies
Version *v1
Document type *Recipe
Target audience *Myself and friends that enjoy chocolate chip cookies that are easy to make
Desired outcome *Create a recipe that is simple and easy but makes great cookies
Scope & constraintsLeave blank
Tone & voiceLeave blank
Additional instructionsNo extra ingredients like nuts
Length Constraint
(field below the goal fields)
Leave blank
Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)

Click Start from Scratch and leave the text box empty. That's it — the six goal fields above are all the hive needs. Head to the work screen and click Smoke the Hive.

Rounds
2 rounds typical — draft in round 1, majority convergence in round 2 (measured, not estimated)
What to Watch For

This is your first hands-on look at the workflow. While you run it, pay attention to:

  • How the reviewers propose numbered changes in the Live Console
  • How the Builder decides which suggestions to accept, reject, or flag
  • How conflicts appear when AIs disagree — and how you resolve them
  • How the working document evolves between rounds
  • How quickly the hive converges when the goal is well-defined
💡

Once the cookies run converges, export the final recipe — it's a real working recipe you can use. Then click 🏁 Finish on the work screen and choose Start New Project to wipe the session and begin work on your next document. The playbooks below have field-by-field templates for common document types — pick the one closest to what you're writing.

↑ Back to top

🔒 Running at work or on an internal AI server

Caveats for using these playbooks on a corporate or internal AI gateway — the substrate matters.

The round counts shown in the playbooks below were measured on public-AI hives — ChatGPT, Claude, Gemini, and similar consumer-facing models with internet access. If you're running WaxFrame on the AI platform your IT team set up at work (Open WebUI, LM Studio, on-prem deployments), expect different round counts and convergence patterns.

The substrate matters. An internal AI server hosts whatever models your IT team has provisioned, and those models can have knowledge cutoffs months or years apart. On factually-anchored documents this can extend convergence or cause non-convergence. On opinion-or-style-driven documents — which most of the playbooks below are — the substrate matters much less.

Two quick mitigations:

  • For factual content: use Setup 4 (Reference Material) to inject a shared baseline so all AIs work from the same source of truth.
  • For testing a new prompt: run the same project setup on a public-AI hive at home as a control before assuming WaxFrame is the problem.

Full guidance on running WaxFrame in enterprise environments is in the user manual: Appendix C — Using WaxFrame at Work ↗.

↑ Back to top

How to read the convergence numbers

What "rounds to convergence" actually means — and why it's not the same as "perfect."

Every playbook below shows a measured round count. Two stopping signals can produce that number, and they mean different things.

Unanimous convergence Every reviewer in the hive marked the round as "no changes needed." Full agreement. The strongest stopping signal — the document has been approved by all of them.
Majority convergence This occurs when a majority of the Hive have approved the document but there are some that feel the document needs more work. The document having been approved by the majority is ready to be shipped, but holdouts will offer up their last suggestions which you can review and decide if you'd like to incorporate them or just Finish.

Neither convergence type means the document is perfect — only YOU can decide that, it's your document after all. So, If you would like to make hand crafted amendments you can do so as always on the working document itself or if you want to have the builder "tighten the third paragraph" or "adjust the closing line it is too long" you can send those NOTES and run another round, you are in charge!

↑ Back to top

Career & Hiring

Résumés, cover letters, job descriptions, thank-you notes. High-stakes writing where audience and positioning matter and multiple reviewers catch drift a solo AI misses.

✉️
Cover Letter
Create or refine a cover letter that connects your background directly to a specific role.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Replace anything in [brackets] with your own information.

FieldWhat to enter
Document type *Cover letter
Target audience *Hiring manager at [company name] reviewing applications for a [job title] position
Desired outcome *Reader wants to interview me. Three short paragraphs: a strong opening hook, a direct connection between my experience and the role, and a confident close with a clear call to action.
Scope & constraintsThree paragraphs maximum. No generic openers like "I am writing to express my interest." No filler phrases. No sign-offs like "I look forward to hearing from you."
Tone & voice[Professional / conversational / enthusiastic] — pick one that fits the company
Additional instructionsDo not add claims about my experience that are not supported by what I provide. Do not fabricate anything.
Starting from scratch? Change these fields:
Desired outcomeGenerate a cover letter from scratch using only the talking points I provide in Notes. Three paragraphs: strong hook, direct connection to the role, confident close with a call to action.
Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)
Refining a draft
Go to Setup 5 — Starting Document. Paste your existing cover letter text using the Paste Text button.
Starting from scratch
Go to Setup 5 — Starting Document. Click Start from Scratch and leave the text box empty. Before Round 1, open the Notes drawer and paste your talking points: why you are a fit for this specific role, one concrete accomplishment to highlight, and what specifically excites you about the company or position.
Rounds
6–10 rounds typical — cover letters take more rounds than other short documents because each round tunes specificity of the hook and tightens the role connection (measured, not estimated)
Best AIs For
Claude — voice & authenticity ChatGPT — punchy rewrites Grok — tone calibration
💡

On Setup 4 — Reference Material, paste the full job description. The hive will see it every round and check that your cover letter language, emphasis, and examples actually match what the employer is asking for. The AIs cannot visit URLs, but a pasted JD gives them direct authoritative source to cite against — much more reliable than describing the role in Notes.

↑ Back to top
🔍
Job Description
Write a job posting that attracts the right candidates and accurately represents the role.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Replace anything in [brackets] with your own information.

FieldWhat to enter
Document type *Job description
Target audience *Candidates applying for a [job title] role — be specific: e.g. Mid-level software engineers with 3–5 years of experience, Entry-level candidates in defense or aerospace
Desired outcome *Qualified candidates immediately understand the role, the requirements, and why the job is worth applying for. Unqualified candidates self-select out. Responsibilities are listed in order of importance, most critical first.
Scope & constraintsList responsibilities from most important to least. Flag any must-have requirement that could unnecessarily narrow the candidate pool. Include a brief company culture statement at the end. Salary range: [amount or write OMIT].
Tone & voice[Professional / startup / enterprise] — match the culture of the company
Additional instructionsDo not add requirements that are not on my original list. Do not soften or remove must-have qualifications without flagging it as a suggestion first.
Starting from scratch? Change these fields:
Desired outcomeGenerate a complete job description for a [job title] role at a [company type] from the list of responsibilities and requirements I provide in Notes. List responsibilities by importance, most critical first.
Real-world example — JD that took 21 rounds
Run these exact values yourself to reproduce the convergence behavior documented above. Use the values literally on a first run to learn the rhythm; on later runs, swap the specifics for your own role and notes.
Project nameJD — Network Engineer Altura Systems
Versionv1
Document typeJob description
Target audienceNetwork engineers with 3 to 7 years of experience, currently at an MSP or small-to-mid enterprise, who want more hands-on work and less ticket-queue churn
Desired outcomeQualified candidates read it and apply. The non-qualified self-select out — we don't need to filter through 200 resumes. Responsibilities ranked by time spent, not vanity.
Scope & constraintsSalary range goes at the top, not buried. List must-haves separately from nice-to-haves. No jargon like "ninja" or "rockstar." Include one sentence about the company culture that is actually true — we are small, loud, and nobody micromanages.
Tone & voiceHonest and specific, not startup-cute. Write like a person, not a recruiter.
Additional instructionsIf any required qualification looks like it would narrow the pool unnecessarily, flag it and suggest an alternative. Do not list certifications as required — list them as preferred.
Length ConstraintLeave blank
Starting DocumentClick Start from Scratch and leave the text box empty
Notes payload (paste into the Notes drawer before clicking Smoke the Hive on Round 1):
Role: Network Engineer, full-time, in-office 3 days per week, hybrid otherwise
Location: Tampa, FL
Salary: $95k to $120k DOE + profit share
We are Altura Systems, 14 people, IT services for small healthcare and law firms across Florida.
Responsibilities, in order of time spent:
- Design and deploy small-to-mid office networks (40 to 200 users), mostly Cisco Meraki and UniFi
- Troubleshoot client network issues — remote and on-site
- Run site surveys and wireless heatmaps with Ekahau
- Document everything in our internal wiki
- Occasional after-hours cutovers, comp time given
Must have:
- 3+ years hands-on network engineering
- Comfortable with Meraki or equivalent cloud-managed platform
- Can read a switch config and know what's broken
- Clean driving record (client site travel)
Preferred:
- CCNA or equivalent
- Any wireless certs (CWNA, CWDP)
- MSP background
Culture: small team, we fix our own mistakes, no politics, no unnecessary meetings, you own your work end to end.
Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)
Refining a draft
Go to Setup 5 — Starting Document. Paste your existing job description using the Paste Text button.
Starting from scratch
Go to Setup 5 — Starting Document. Click Start from Scratch and leave the text box empty. Before Round 1, open the Notes drawer and paste your list: core responsibilities, required qualifications, nice-to-haves, and any company culture detail you want included.
Convergence
≈19 minutes · 21 rounds (measured, not estimated) — even from scratch with full reference materials, real convergence on a quality JD takes 20+ rounds, not 3–5. Round 21 reached majority convergence (3 of 4 AIs satisfied) with the holdout offering minor wording suggestions.
Best AIs For
ChatGPT — clear structure Claude — inclusive language Gemini — role benchmarking
💡

Before Round 1, open Notes and write: "Flag any must-have requirement that could reasonably be a nice-to-have — overly strict requirements reduce applicant quality by narrowing the pool too early."

↑ Back to top
📄
Résumé
Polish a draft résumé for a specific role, or generate one from your career notes.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Replace anything in [brackets] with your own information. Fields marked * are required to continue.

FieldWhat to enter
Document type *Résumé
Target audience *Hiring manager or recruiter reviewing candidates for a [job title] role — e.g. Hiring manager at a defense technology company
Desired outcome *Reader invites me to interview. The résumé shows clear impact and achievements, not just job duties. Every bullet demonstrates value.
Scope & constraintsDo not fabricate experience. Do not remove job titles, companies, or dates. Strengthen what is already there — do not invent.
Tone & voiceConfident, professional, action-oriented — strong verbs, no passive voice, no "responsible for"
Additional instructionsDo not remove or change any metrics, percentages, or dates — these are factual and verified by me.
Length Constraint
(field below the goal fields)
Enter 1 · select Pages for a one-page résumé, or 2 for two pages
Starting from scratch? Change these fields:
Desired outcomeGenerate a complete résumé for a [job title] role from the notes I provide. Use only what I give you — do not add experience I have not mentioned.
Additional instructionsBuild only from my notes. Do not fabricate any experience, skills, or achievements.
Real-world example — Résumé that took 11 rounds
Run these exact values yourself to reproduce the convergence behavior documented above. Use the values literally on a first run to learn the rhythm; on later runs, swap the specifics for your own resume and notes.
Project nameResume — Dana Reyes Wireless
Versionv1
Document typeRésumé
Target audienceHiring manager at a mid-size enterprise or defense-adjacent company reviewing candidates for Senior Wireless Engineer roles
Desired outcomeReader invites me to interview. The résumé leads with impact — not job duties. Every bullet shows a result, a scale, or a problem solved.
Scope & constraintsDo not invent experience. Do not remove job titles, companies, or dates. Strengthen what is there. Tighten the older entries so the more recent work gets more space. No "responsible for." No "utilized" when "used" works.
Tone & voiceConfident, direct, active verbs.
Additional instructionsDo not change any numbers, dates, percentages, or dollar amounts. Those are verified.
Length ConstraintEnter 1 · select Pages
Starting DocumentClick Paste Text and paste the text below
Paste this as the starting document:
Dana Reyes
Tampa, FL · dana.reyes@example.com · 813-555-0114

Summary
Wireless network engineer with 8 years of experience. CWNA, CWDP, CWAP. Aruba, Cisco Meraki, and Ruckus. Specializes in warehouse and industrial RF environments.

Experience

Senior Wireless Engineer, Vantage Logistics — Jan 2022 to Present
- Responsible for wireless network across 12 warehouse sites and 3 corporate offices
- Led migration from Cisco Meraki to Aruba ArubaOS 8
- Did site surveys with Ekahau
- Supported autonomous mobile robots operating on the floor

Wireless Engineer, Meridian IT Services — Jun 2018 to Dec 2021
- MSP role supporting 30+ small and mid-size clients
- Handled escalations for wireless issues
- Designed networks for new client buildouts

Network Technician, Gulfstream Communications — Aug 2016 to May 2018
- Installed and configured wireless access points
- Ran cable, tested drops, closed tickets

Certifications
CWNA, CWDP, CWAP, Aruba ACMA

Education
BS Information Technology, University of South Florida — 2016
Mid-stream Notes injection — paste between Round 2 and Round 3, not at the start:
This run deliberately held key facts back until Round 2 — letting the hive first read the existing draft on its own terms. After Round 2 completed, the Notes drawer (📝 Notes in the top bar) was opened and the payload below pasted in. The next launch ran as a Builder Only round, incorporating the new facts directly without another round of reviewer voting. Notes are one-shot — applied to the next round only, then cleared automatically. This is the right place to drop "I forgot to mention" details or new context mid-project without restarting from scratch.
Job posting I'm targeting: Senior Wireless Engineer at a defense technology company. They emphasize industrial RF, mobility for robotics, and large-scale deployments. Title on posting is "Senior Wireless Network Engineer."

Things I forgot to include in my resume:
- At Vantage, I cut wireless-related Tier-1 tickets by 42% year-over-year after the Aruba migration
- Designed the RF for a 46-foot ceiling warehouse in Memphis with 6-foot aisle widths — unusual deployment, delivered on time, zero post-deployment escalations
- At Meridian I was the only engineer with CWAP — became the go-to escalation point for wireless issues across all clients
Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)
Refining a draft
Go to Setup 5 — Starting Document. Upload your résumé file (Word or PDF) using the Upload File button, or paste the text using Paste Text. If uploading a PDF, use a text-based export — not a designed template file — for best results.
Starting from scratch
Go to Setup 5 — Starting Document. Click Start from Scratch and leave the text box empty. Before clicking Smoke the Hive on Round 1, open the Notes drawer (📝 Notes in the top bar) and paste your career details: job titles, companies, dates, key achievements, skills, and education.
Convergence
≈11 minutes · 11 rounds (measured, not estimated) — even refining a strong existing draft with reference materials and notes, real convergence on a quality résumé takes 10+ rounds. Round 11 reached majority convergence (4 of 6 AIs satisfied) with 3 of those 4 simply confirming no further changes needed.
Best AIs For
ChatGPT — clean rewrites Claude — tone & structure Gemini — keyword alignment
💡

On Setup 4 — Reference Material, paste the full job description from the posting. The hive will align language, emphasis, and skill phrasing to the actual role on every round — this is one of the highest-leverage moves you can make on a résumé. Notes are cleared after each round; reference material persists, so you only have to paste it once.

↑ Back to top
🔗
LinkedIn About
Write the About section of your LinkedIn profile — what you actually do, who you do it for, and the credentials that prove it. Recruiter-scannable in 30 seconds, peer-credible on a closer read.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Replace anything in [brackets] with your own information. Fields marked * are required to continue.

FieldWhat to enter
Document type *LinkedIn About Profile
Target audience *Recruiters scanning for [your key certs / platforms / tools] keywords in 30 seconds. Peers in the field reading it on a closer pass to decide if you actually do the work.
Desired outcome *Recruiter sees credentials and platform experience and flags it as a match. Peer reads it and thinks "this person actually does the work, not just talks about it."
Scope & constraintsLead with what you actually do, not where you work. Include credentials and platform experience without it reading like a resume bullet list. End with one line that gives a peer something to connect on.
Tone & voiceProfessional but human. Not stiff. Engineer that can speak customer.
Additional instructionsNo "passionate about," "results-driven," or "proven track record." If a sentence sounds like it could have been written about anyone in this field, rewrite it until it could only have been written about me. No "open to opportunities" or "looking for my next chapter."
Length Constraint
(field below the goal fields)
Enter 2000 · select Characters (LinkedIn's About section permits up to 2,600 — leaving headroom keeps the run from bumping the ceiling)
Real-world example — LinkedIn About converged in 2 rounds, 1 minute
Run these exact values yourself to reproduce the convergence behavior documented above. Use the values literally on a first run to learn the rhythm; on later runs, swap the specifics for your own role and credentials. The Reference Material payload below is the identity scaffold — it does most of the work.
Project nameLinkedIn Profile — About Section
Versionv1.0
Document typeLinkedIn About Profile
Target audienceRecruiters scanning for CWNA / CWAP / Aruba / Ekahau keywords in 30 seconds
Desired outcomeRecruiter sees credentials and platform experience and flags it as a match. Peer reads it and thinks "this person actually does the work, not just talks about it."
Scope & constraintsLead with what I actually do, not where I work. Include credentials (CWNA, CWDP, CWAP) and platform experience (Aruba, Ruckus, Cisco Meraki, Ekahau) without it reading like a resume bullet list. End with one line that gives a peer something to connect on.
Tone & voiceProfessional but human. Not stiff. I'm not "corporate" — I am an engineer that can speak customer.
Length ConstraintEnter 2000 · select Characters
Starting DocumentClick Start from Scratch — leave the text box empty
Paste this as Setup 4 — Reference Material (1,163 characters):
Role:  Senior Wireless Network Engineer (contractor) at Anduril Industries.

Credentials: CWNA, CWDP, CWAP. Working through CWNE.

Primary platform: Aruba on-prem ArubaOS 8.x. Past platforms
include Ruckus and Cisco Meraki. Tools:  Ekahau and iBwave for predictive and validation surveys. Strong on enterprise, warehouse, and industrial RF environments.

What I actually do day to day:  design wireless networks for sites where the floor plan and the predictive model disagree, validate them on-site, and own the deployment through Day 2 operations. The work is half RF physics and half "the customer told me there's no metal in the ceiling and the ceiling is actually a metal grid."

Identity stack at work:  Okta, Kolide, Cloudflare. Not Active Directory.

What I want a peer to know about me:  I'd rather walk a site than guess at one. Predictive surveys are a starting point. The answer is always on the floor with a laptop.

What I want a recruiter to know about me:  I have shipped production wireless deployments at scale, on-prem, in industrial environments, with constraints (air-gapped networks, restricted CDN access, no AD). I know what I don't know.
Round 1 Notes drawer injection — paste BEFORE clicking Smoke the Hive on Round 1:
Open the Notes drawer (📝 Notes in the top bar) and paste the buzzword guard below into the 🎯 This-round notes section. This is the directive that produced clean draft prose on Round 1. The Builder applied it during the Draft phase; the result was strong enough that Round 2 reached majority convergence with two of three reviewers saying NO CHANGES NEEDED.
No "passionate about." No "results-driven." No "proven track record." If a sentence sounds like it could have been written about anyone in my field, rewrite it until it could only have been written about me.
Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)
Refining your existing About
Go to Setup 5 — Starting Document. Click Paste Text and paste your current LinkedIn About section. The hive will refine in place — preserving your structure where it works, rewriting where it doesn't.
Starting from scratch
Go to Setup 5 — Starting Document. Click Start from Scratch and leave the text box empty. The Reference Material from Setup 4 is the identity scaffold the hive will draft from. Open the Notes drawer and paste the Round 1 buzzword guard before hitting Smoke.
Convergence
≈1 minute · 2 rounds (measured 2026-05-09) — but this fast convergence reflects three things working together: a tight 3-AI hive (ChatGPT + Claude + Gemini-as-Builder, where majority is 2 of 3), a richly-detailed Reference Material payload (1,163 characters of identity scaffold), and a one-shot Round 1 Builder note directing the buzzword guard. On a fresh run with the template alone (no Round 1 note, generic identity scaffold), expect 4–6 rounds. The Reference Material is doing most of the heavy lifting — invest there. Final document came in around 1,500 characters, well under the 2,000 ceiling.
Best AIs For
ChatGPT — clean prose Claude — tone & voice Gemini — keyword alignment & Builder
💡

The identity scaffold structure (Role / Credentials / Primary platform / Tools / What I actually do / Identity stack / What I want a peer to know / What I want a recruiter to know) is reusable across roles. Swap the field values for your own and the hive aligns to whatever profile you're writing — wireless engineer, accountant, ops manager, lawyer. The structure is the trick; the values are personal.

↑ Back to top
🙏
Thank-You Letter
Write a warm, genuine thank-you for any professional or personal occasion.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Replace anything in [brackets] with your own information.

FieldWhat to enter
Document type *Thank-you letter
Target audience *The person receiving it — be specific: e.g. My hiring manager John Smith, A client named Sarah at Acme Corp, My mentor Dr. Jones
Desired outcome *Reader feels genuinely appreciated and remembers the specific moment I am referencing.
Scope & constraintsNo generic openers like "I hope this finds you well." No corporate sign-offs like "Best regards." One specific moment or gesture must be named — no vague thank-yous.
Tone & voice[Warm and personal / professional but sincere / heartfelt] — pick one
Additional instructionsReference the specific thing I am thanking them for. Do not add any details I have not provided. Do not fabricate anything.
Length Constraint
(field below the goal fields)
Enter 200 · select Words
Starting from scratch? Change these fields:
Desired outcomeGenerate a thank-you letter from scratch using only the bullet points I provide in Notes. Make it feel specific and genuine — not generic.
Real-world example — Thank-You that took 3 rounds
Run these exact values yourself to reproduce the convergence behavior documented above. Use the values literally on a first run to learn the rhythm; on later runs, swap the specifics for your own letter.
Project nameThank-You — Marco Contractor
Versionv1
Document typeThank-you letter
Target audienceMarco Delgado, general contractor who led the remodel of our kitchen and back porch
Desired outcomeMarco feels genuinely appreciated and remembers the specific moment that made this job different. I want him to feel like he can use this as a reference in his own work.
Scope & constraintsNo "I hope this finds you well." No "Best regards." Name the one specific thing he did that went beyond. Handwritten tone — this is going in a card, not an email.
Tone & voiceWarm, personal, specific — from one person to another, not a form letter
Additional instructionsDo not invent details. Only use what is in Reference Material. Don't mention price — this isn't about money.
Length ConstraintEnter 200 · select Words
Starting DocumentClick Start from Scratch and leave the text box empty
Reference Material payload (paste into Setup 4 — Reference Material):
Marco Delgado, owner of Delgado Build, finished our kitchen and back porch remodel last Friday.
Things to thank him for:
- The big one: when the countertop supplier delivered the wrong slab on day 17 and the project was going to slip two weeks, Marco drove to Orlando on a Saturday, picked up the right slab himself, and installed it Sunday morning. We had our son's birthday party on the new porch the following Saturday as planned.
- He showed up when he said he would, every day, for six weeks.
- His crew cleaned up every night — my wife kept commenting on it.
- He caught a framing issue the inspector missed and fixed it before closing the wall.
What we want him to know:
- We'll absolutely use him again.
- Our neighbor Janet has already asked for his number.
- He's welcome to drop by and show the kitchen to a future client if he ever needs to.
Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)
Refining a draft
Go to Setup 5 — Starting Document. Paste your existing draft letter using the Paste Text button.
Starting from scratch
Go to Setup 4 — Reference Material and paste your bullet points: who you are thanking, what specifically happened or what they did, and any particular detail you want mentioned by name. Then on Setup 5 — Starting Document, click Start from Scratch and leave the text box empty. Optionally, before Round 1, open the Notes drawer (📝 Notes in the top bar on the Work screen) for any one-shot Round 1 emphasis.
Convergence
≈1 minute · 3 rounds (measured, not estimated) — short documents with bulleted Reference Material converge fast. The recent from-scratch run (Thank-You — Marco Contractor v1) reached majority convergence at Round 3.
Best AIs For
Claude — warmth & authenticity Grok — keeping it real ChatGPT — clean final polish
💡

If the result sounds like a corporate email instead of a personal letter, add a Note before the next round: "This should sound like a real person wrote it, not a template. Rewrite any sentence that sounds generic or corporate."

↑ Back to top

Business & Sales

Proposals, summaries, formal responses, cold outreach. Documents where clarity, persuasion, and stakeholder alignment matter more than any single AI's style.

💼
Business Proposal
Sharpen a proposal to win a client, secure a partnership, or pitch a new venture.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Replace anything in [brackets] with your own information. The Scope & constraints field is where you tell the hive who you are and what you sell — fill in the identity/offering/pricing scaffold below the section list.

FieldWhat to enter
Document type *Business proposal
Target audience *[Client name or type] — be specific: e.g. IT Director at a mid-size manufacturing firm, Procurement committee at [company name], Owner Operator at [client business]
Desired outcome *Reader approves the proposal or schedules a follow-up meeting. They clearly understand what we are offering, what it costs, and what happens next.
Scope & constraintsRequired sections: executive summary, problem statement, proposed solution, pricing or next steps. Do not add claims not supported by the existing content.

Identity & offering — fill these in:
• Who you are: [your company name + what kind of business]
• Services or products you offer: [be specific]
• What you do NOT offer: [explicit guardrail — e.g. "Do not invent services I don't offer"]
• Pricing structure: [hourly rate + minimum, or fixed-fee, or tiered packages]
Tone & voiceConfident, credible, professional — not salesy
Additional instructionsDo not change any pricing figures, timelines, or deliverable commitments. These are factual. Use [PRICE] or [TIMELINE] as placeholders where you've left them blank.
Length Constraint
(field below the goal fields)
Range mode, 1–2 pages (template default — 1 page = ~600 words, 2 pages = ~1200 words). Adjust if your proposal needs different length bounds.
Starting from scratch? The identity scaffold in Scope & constraints does the work the Notes drawer used to:
Setup 5 — Starting DocumentClick Start from Scratch and leave the text box empty. The hive will build the proposal from your Project-screen Scope & constraints field — that's where your company name, services, pricing, and "what I don't do" guardrails live now. No Notes-drawer setup needed before Round 1; the identity content is already in the prompt envelope via Scope.
Real-world example — Wireless survey service for a small recreation business, 3 rounds, 3-AI budget hive
A from-scratch business proposal for a fictional recreational-business client (Brightwater Canoe and Kayak) from a real wireless-engineering consultancy (Eye Productions). Demonstrates the identity scaffold in action — the Scope & constraints field carries the full company-and-offering payload, and the hive shapes a complete proposal around it without a starting draft. Run with the same 3-AI budget hive used for the LinkedIn DFS post and Bay Area Contractor letter (paid Claude + paid ChatGPT + free Gemini-as-Builder). All Project-screen prompts below are the literal values used in the run.
Project nameBP - Wireless Survey Service for Brightwater Canoe and Kayak
Versionv1.0
Document typeBusiness proposal
Target audienceBrightwater Canoe and Kayak — Owner Operator
Desired outcomeReader approves the proposal or schedules a follow-up meeting. They clearly understand what we are offering, what it costs, and what happens next.
Scope & constraintsRequired sections: executive summary, problem statement, proposed solution, pricing or next steps. Do not add claims not supported by the existing content.

I am Eye Productions a full service Wireless Engineer. I can do surveys (predictive, active and passive, validation) and troubleshooting as well as turnkey operations and deployment my basic service rate is $100 an hour min 4 hours. Do not make up stuff I don't do
Tone & voiceConfident, credible, professional — not salesy
Additional instructionsDo not change any pricing figures, timelines, or deliverable commitments. These are factual. Use $100/Hour as placeholders where I have left them blank.
Length LimitRange mode, 1–2 pages
Hive compositionBuilder: Gemini (free tier). Reviewers: ChatGPT (paid), Claude (paid), Gemini-as-reviewer (free). All other AIs toggled off via the per-session bee toggles on the work screen.
Starting DocumentFrom Scratch (no draft pasted) — the proposal was generated entirely from Setup 3 fields. Identity, services, and pricing all flowed in through Scope & constraints; no Notes-drawer pre-injection needed.
Why the identity scaffold matters:
Earlier versions of this template's Scope field only addressed structure — required sections, "do not invent claims" — but left identity, offering, and pricing for the user to figure out. A non-expert user picking the Business Proposal template would type their target audience and outcome correctly, then either skip Scope entirely or write something generic. The hive would respond with placeholder language ("our team," "competitive pricing," "industry-leading service") because it had nothing concrete to anchor on. The improved Scope field prompts the user to provide the specifics that make a proposal real: "I am Eye Productions a full service Wireless Engineer. My basic service rate is $100 an hour min 4 hours. Do not make up stuff I don't do." That language flows directly into the Builder's prompt envelope and the resulting proposal cites those exact services, that exact rate, that exact minimum — no invented pricing, no padded credentials, no "passionate team of experts." Same result the JD playbook documents for upfront-notes injection (v3.21.11), achieved here through Scope rather than Notes.
Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)
Refining a draft
Go to Setup 5 — Starting Document. Upload your proposal file or paste the text. You can use [PRICE] or [TIMELINE] as placeholders for sensitive figures — the AIs will work around them and leave them unchanged.
Starting from scratch
Go to Setup 5 — Starting Document. Click Start from Scratch and leave the text box empty. All your identity content is already in the Scope & constraints field on Setup 3 — the hive will build the proposal directly from there.
Convergence
≈8 minutes wall-clock across 3 rounds (measured, not estimated — per the document footer's "Produced by WaxFrame in 3 rounds and 8 minutes" byline). Final output: 1175 words / 8369 characters — sits at the upper end of the 1–2 page range under v3.36.12's corrected 600-words-per-page math (2 pages ≈ 1200 words). SCRATCH-mode runs with a fully-populated identity scaffold converge faster than draft-refinement runs because there's no rough-prose to wrestle into shape — the hive builds clean from Round 1. Plan for 3–6 rounds depending on how complete your Scope & constraints scaffold is and whether you have specific client objections to address.

3-AI vs 2-AI reference data (v3.36.14, May 2026): A 2-AI Auto-Mode run on the same Brightwater scaffold took 11 rounds / ≈16 minutes because Round 5 produced a tied USER DECISION block with no majority, halting Auto until the user manually picked. A 3-AI run (Brightwater BP v3.0) on the same scaffold converged in 5 rounds / ≈10 minutes — the third reviewer broke decision ties automatically and Auto chained without interrupts. Recommendation: run Auto Mode with at least 3 AIs. The cost is one more reviewer per round; the win is fewer manual decision interrupts and faster wall-clock.

Second-scenario data point (Wellness for Law Firm v1.0, May 2026): A different BP scaffold (employee wellness program proposal aimed at a managing partner) ran with a 5-AI hive (ChatGPT, Gemini, Grok, Perplexity, Mistral, Gemini-as-Builder). 11 rounds to majority convergence, final 857 words. Word count trended down 897 → 854 → 857 across rounds — a steady tightening pattern, not the structural rewrites Brightwater needed. The 11-round figure here is coincidentally identical to the Brightwater 2-AI run's 11 rounds, but the cause is different: this run had more reviewers per round, each with their own polish suggestions, so iteration count is high but each round produced a small tweak rather than a fundamental restructure. Larger hives can take more rounds even on simpler scaffolds because there are more voices to satisfy.

Length-overshoot caveat: The hive consistently over-produces 25–50% on the first draft regardless of length target. Reviewers focus on prose quality and do not address length — only the Builder responds to length directives. If the document holds steady above the ceiling for multiple rounds despite an armed length guard, plan to inject the ✂ Trim template via a Builder-Only round once the bloat-guard ceiling crosses. The v3.0 reference run did exactly this: a single Trim injection at Round 4 compressed 1499 → 1083 words (a 27.6% reduction) in one Builder-Only round, and the next reviewer round confirmed convergence at the new size.
Best AIs For
ChatGPT — structure & flow Gemini — research backing Claude — persuasive language Perplexity — market context
💡

The "Do not make up stuff I don't do" line in the Brightwater example is the most important sentence in that Scope field. AIs default to filling gaps with plausible-sounding boilerplate; an explicit anti-invention guardrail keeps the proposal honest. State exactly what you don't offer alongside what you do — "I do not provide ongoing managed services. Do not invent ongoing service tiers." — and the hive will respect those boundaries through every round.

↑ Back to top
📬
Email & Outreach
Sharpen cold outreach, sales emails, follow-ups, or any important one-off message that needs to land.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Replace anything in [brackets] with your own information.

FieldWhat to enter
Document type *Be specific about the email type: e.g. Cold outreach email, Follow-up email after a meeting, Sales introduction email
Target audience *Who is receiving this — be specific: e.g. VP of Engineering at a mid-size SaaS company, A former colleague I have not spoken to in two years, Hiring manager who interviewed me last week
Desired outcome *Reader opens it, reads it in full, and takes one specific action — e.g. replies to schedule a 20-minute call, clicks the link, responds with a yes or no.
Scope & constraintsLead with value to the recipient — not background on the sender. One clear ask only. No fluff, no jargon, no "I hope this email finds you well." Include a subject line as the first line of the document.
Tone & voice[Professional / direct / warm] — pick one. For cold outreach, direct tends to work better than warm.
Additional instructionsDo not add background about me unless I specifically provide it. The email should be about what the reader gets, not who I am.
Length Constraint
(field below the goal fields)
Enter 150 · select Words
Starting from scratch? Change these fields:
Desired outcomeGenerate a complete email from scratch using only what I provide in Notes. Lead with value to the recipient. One clear ask. Include a subject line as the first line.
Real-world example — Cold outreach that took 3 rounds
Run these exact values yourself to reproduce the convergence behavior documented above. Use the values literally on a first run to learn the rhythm; on later runs, swap the specifics for your own recipient and message.
Project nameOutreach — Ferris at Lightkeeper
Versionv1.0
Document typeCold outreach email to a potential channel partner
Target audienceFerris Okafor, VP of Partnerships at Lightkeeper Security, a managed security services firm. We've never spoken. He's active on LinkedIn and posts about SMB cybersecurity.
Desired outcomeFerris replies to schedule a 20-minute intro call. One clear ask, no bait-and-switch.
Scope & constraintsLead with value to him, not background on us. One ask. No "I hope this email finds you well." Include a subject line as the first line. Reference one thing from his recent LinkedIn activity so it doesn't read as scraped.
Tone & voiceDirect, warm, not a vendor
Additional instructionsDo not add claims about our company unless they appear in Reference Material. Do not oversell the first call — the ask is 20 minutes, nothing more.
Length ConstraintEnter 0.5 · select Pages
Starting DocumentClick Start from Scratch and leave the text box empty
Reference Material payload (paste into Setup 4 — Reference Material):
Ferris's recent LinkedIn post (last week): he wrote that most SMB security incidents start with unmanaged wireless — not ransomware, not phishing. Direct quote: "your MSP partner should be treating your Wi-Fi like a security surface, not a convenience."

Our company: Altura Systems, a 14-person MSP in Tampa. We run wireless for about 40 SMB clients across healthcare and legal. We've seen the same pattern Ferris is describing — clients keep getting breached through guest SSIDs that were stood up in 2019 and forgotten.

The ask: 20-minute intro call to see if the pattern he's describing matches what we're seeing.

Sender:
- Name: Dana Reyes
- Role: Director of Wireless Services, Altura Systems
Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)
Refining a draft
Go to Setup 5 — Starting Document. Paste your existing draft email using the Paste Text button. Include the subject line at the top if you have one.
Starting from scratch
Go to Setup 4 — Reference Material and paste your context: who the recipient is, what you know about them (recent LinkedIn activity, prior interactions, any quote you want referenced verbatim), what you want them to do, and why it benefits them specifically. The hive cannot browse the web — anything you want it to reference must be pasted into Reference Material first. Then on Setup 5 — Starting Document, click Start from Scratch and leave the text box empty. Optionally, before Round 1, open the Notes drawer (📝 Notes in the top bar on the Work screen) for any one-shot Round 1 emphasis.
Convergence
≈1 minute · 3 rounds (measured, not estimated) — short documents with bulleted Reference Material converge fast. The recent from-scratch run (Outreach — Ferris at Lightkeeper v1.0) reached majority convergence at Round 3 — same pattern as the Thank-You playbook test.
Best AIs For
Grok — punchy subject lines Claude — concise body copy ChatGPT — CTA clarity
💡

Short emails sometimes produce conflicts over single word choices. If a conflict feels trivial, use the Builder Decision path and let the Builder decide — or bypass it if you have already made the edit directly in the document.

↑ Back to top
📊
Executive Summary
Distil a long report, plan, or proposal into a tight, decision-ready summary for leadership.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Replace anything in [brackets] with your own information.

FieldWhat to enter
Document type *Executive summary
Target audience *Be specific about who reads this — e.g. VP of Operations and CFO, Board of Directors, Program managers and senior leadership
Desired outcome *Reader understands the situation, the recommendation, and why it matters — in that order. They can make a decision or take action without reading the full source document.
Scope & constraintsLead with the conclusion. Strip jargon. Do not expand — only tighten and clarify what is already here.
Tone & voiceDirect, authoritative, jargon-free — written for someone who has 90 seconds to read it
Additional instructionsDo not add detail not present in the source material. Do not change any figures, recommendations, or conclusions.
Length Constraint
(field below the goal fields)
Enter 400 · select Words — or enter 1 · select Pages for a one-pager
Starting from scratch? Change these fields:
Desired outcomeGenerate an executive summary for [project, report, or initiative name] from the bullet points I provide. Answer three questions in order: what is the situation, what is the recommendation, why does it matter. Lead with the conclusion.
Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)
Refining a draft
Go to Setup 5 — Starting Document. Paste your existing executive summary using Paste Text, or upload the file. Do not paste the full source document — paste only the summary itself.
Starting from scratch
Go to Setup 5 — Starting Document. Click Start from Scratch and leave the text box empty. Before Round 1, open the Notes drawer and paste your key bullet points: the situation, the recommendation, why it matters, and any decisions or actions required. For a long source document, paste the key conclusions only — not the whole thing.
Rounds
7 rounds, ≈41 minutes wall-clock (measured, not estimated). Reference run: 2025 Sustainability Report, 2026-05-10. 5-AI hive (ChatGPT, Claude, Gemini, Grok, Perplexity) with DeepSeek as Builder, REFINE mode on a hedged corporate draft. Final 142 words / 998 characters. Majority convergence at R8 (5 of 6 satisfied — R1 was the setup round, so the WaxFrame round counter shows 7).

USER DECISION-heavy run. 7 USER DECISION blocks surfaced across the run (1 in R2, 3 in R4, 3 in R6) — corporate sustainability prose has a lot of choices about what to commit to vs. what to hedge, and reviewers don't agree on which side to land on. Two Builder-Only rounds applied locked decisions via mid-stream notes. If you're refining a heavily-edited corporate draft full of soft commitments, expect 5–9 rounds with multiple decision blocks rather than a quick 3-round polish.

Length-cap insight. The 400-word Hard cap was set at the recommended value but the hive landed at 142 words and stayed there from Round 3 onward. Executive summaries written tight will self-bound well under a 400-word ceiling — the Hard cap acts as a safety rail rather than an active constraint. If your source has a hard one-page or 250-word requirement, set the cap accordingly; otherwise 400 is generous enough that you'll never hit it.
Best AIs For
Claude — distillation & structure ChatGPT — executive language Gemini — fact checking
💡

If the AIs keep adding detail back in, open Notes before the next round and write: "Do not expand this document. Only tighten and clarify what is already here. Remove anything that does not directly support the recommendation."

↑ Back to top
📋
RFP Response (Request for Proposal)
Craft a disciplined, structured response to a formal RFP that addresses every stated requirement.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Replace anything in [brackets] with your own information.

FieldWhat to enter
Document type *RFP response
Target audience *Evaluation committee at [issuing organization] reviewing responses to [RFP name or number]
Desired outcome *Every stated requirement in the RFP is addressed directly and clearly. Evaluators can find our response to each item without hunting for it. Our differentiators are clear and credible.
Scope & constraintsFormal, precise, no marketing language. Do not add capability claims not supported by the existing content. Address requirements in the order they appear in the RFP.
Tone & voiceFormal, authoritative, direct — this is a compliance document, not a sales pitch
Additional instructionsDo not change any figures, dates, or technical specifications. Do not omit or skip any RFP requirement, even if the answer is brief.
Length Constraint
(field below the goal fields)
Check the RFP for a page or word limit and enter it here — e.g. 20 · Pages. Leave blank if the RFP has no stated limit.
Starting from scratch? Change these fields:
Desired outcomeGenerate a draft RFP response for [project name] issued by [organization]. We are [brief description of your team or company]. Our relevant experience: [list]. Address each requirement in the order it appears in the RFP. Do not add capability claims not supported by what I provide.
Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)
Refining a draft
On Setup 4 — Reference Material, paste the full RFP requirements text. Then go to Setup 5 — Starting Document and upload your draft response (or paste the text). Every reviewer will check your draft against the RFP every round and flag any requirement that is not addressed.
Starting from scratch
On Setup 4 — Reference Material, paste the full RFP requirements text. Then go to Setup 5 — Starting Document, click Start from Scratch, and leave the text box empty. The hive will generate a response skeleton that addresses each requirement in the order the RFP states them.
Rounds
20–60+ rounds — RFPs are long, complex, and contentious
Best AIs For
ChatGPT — compliance checking DeepSeek — technical depth Claude — clarity & concision
💡

If the RFP has a page or word limit, enter it in the Length Constraint field on the Project screen — that is where it belongs, not in the goal fields. WaxFrame will enforce it as a hard constraint every round.

↑ Back to top

Content & Marketing

Published writing — blog posts, presentations, content your audience sees. Voice, structure, and narrative flow benefit from multiple reviewers with different strengths.

📝
Blog Post / Article
Publish-ready content with a strong voice, clear structure, and solid flow.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Replace anything in [brackets] with your own information.

FieldWhat to enter
Document type *Blog post — or be specific about format: e.g. Opinion piece, How-to article, Listicle
Target audience *Who reads this and where — e.g. Small business owners new to AI, reading on LinkedIn, Senior engineers evaluating wireless platforms
Desired outcome *Reader finishes the post and [takes a specific action or understands a specific thing]. End with a clear call to action. e.g. Reader understands the three main options and clicks through to learn more.
Scope & constraintsMust cover: [list 3–5 key points]. Do not add statistics or facts that are not already in the draft — flag anything uncertain instead of inventing it.
Tone & voiceBe specific: e.g. Direct, short sentences, slightly sarcastic, no jargon or Authoritative and data-driven, formal but approachable. Without this the AIs will default to a bland, generic style.
Additional instructionsDo not change my unique angle or perspective. Strengthen the voice — do not sand it down into something that sounds like everyone else.
Length Constraint
(field below the goal fields)
Enter your target word count — e.g. 800 · Words
Starting from scratch? Change these fields:
Desired outcomeGenerate a complete [X-word] blog post on [topic] from the notes I provide. My unique angle is [describe what makes your take different]. Build the full post from my key points — do not add facts or statistics I have not provided.
Real-world example — Blog Post that took 16 rounds
First measured refining-a-draft example in the playbook (Thank-You, Email & Outreach, JD, and Résumé examples are from-scratch or refining shorter drafts). The user uploaded an existing draft rather than starting from scratch, then provided a thesis-and-CTA payload through Reference Material. Run these exact values yourself to reproduce the convergence behavior; on later runs, swap the specifics for your own draft and notes.
Project nameBlog — Why I Stopped Trusting One-Shot AI
Versionv1.0
Document typeBlog Post
Target audienceSenior engineers and technical leads who already use ChatGPT or Claude for work, reading on LinkedIn or a company-internal blog
Desired outcomeReader finishes the post and understands why running a document through multiple AI models in sequence produces meaningfully better output than a single high-quality model. They click through to the linked demo at the end.
Scope & constraintsCover: the "one-shot trap," why different models catch different issues, a concrete example with rough numbers, and what the reader can try today. Do not invent statistics.
Tone & voiceFirst-person, slightly opinionated, conversational — written the way you'd say it out loud, not how you'd write a whitepaper. Voice is someone who's been frustrated with single-model output for a year and finally solved the problem. Not promotional. Not AI-generated-sounding.
Additional instructionsNo buzzwords like "synergy," "leverage," "revolutionary." If a section feels padded, cut it. No marketing fluff just real world shit.
Length ConstraintLeave blank
Starting DocumentClick Upload File and select your existing draft post (a .docx, .txt, or .md file)
Reference Material payload (paste into Setup 4 — Reference Material):
Thesis: any single LLM, no matter how good, has blind spots. One model will confidently write fluff where another catches it. One model will tighten structure while another improves tone. Run the same document through 4 or 5 different models in sequence, each refining what the previous one did, and the output is better than any single model can produce on its own.

Concrete example to reference (real, from last month):
- Took a 900-word draft of a proposal
- Ran it through Claude alone: got a polished version, maybe 15% better
- Ran the same draft through Claude, then GPT-4, then Gemini, then back to Claude: got a version where the structural argument was noticeably stronger, the jargon was stripped, and two factual hedges had been flagged that the author hadn't noticed

What the reader should try today: pick a document they care about, hand it to 3 different models in sequence with the same prompt, compare the result to running it through any one model once.

Call to action at the end: link to a tool that automates this — multiple models, single document, convergence loop. (This is WaxFrame, but the post should not name it overtly — the link speaks for itself.)
Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)
Refining a draft
Go to Setup 4 — Reference Material if you have thesis statements, source material, key data points, or anything else the hive should reference but not edit. Then on Setup 5 — Starting Document, paste or upload your existing draft. The hive cannot browse the web — anything you want it to reference must be pasted into Reference Material first.
Starting from scratch
Go to Setup 4 — Reference Material and paste your key points, your angle, source material, and anything you specifically want included or avoided. Then on Setup 5 — Starting Document, click Start from Scratch and leave the text box empty. Optionally, before Round 1, open the Notes drawer (📝 Notes in the top bar on the Work screen) for any one-shot Round 1 emphasis.
Convergence
4 rounds, ≈77 minutes wall-clock (measured, not estimated). Reference run: Blog — Why I Stopped Trusting One-Shot AI v2.0, 2026-05-10. 5-AI hive (ChatGPT, Claude, Gemini, Grok, Perplexity) with DeepSeek as Builder, SCRATCH mode with Reference Material. 1,180-char thesis scaffold pasted into Setup 4 — Reference Material; empty Starting Document. Final 460 words / 2,795 characters. Majority convergence at R4 (4 of 6 satisfied).

4× faster than the v1.0 measurement on the same topic. The first measured run on this exact thesis (v1.0, earlier) took 16 rounds. The v2.0 run shipped at 4 rounds — a 4× improvement driven by two changes: (1) the source thesis went into Reference Material instead of being attempted as a Starting Document, so the hive built fresh from scaffold rather than fighting an existing draft; and (2) the Additional Instructions field carried an explicit buzzword guard ("No 'synergy,' 'leverage,' 'revolutionary.' If a section feels padded, cut it.") that suppressed the AI-flavor language reviewers tend to argue about. Both are general-purpose levers — they apply to most Blog Post runs, not just this topic.

USER DECISION moderate-load run. 2 USER DECISION blocks surfaced in R2, both resolved via a Builder-Only round in R3 with locked-line notes ("Lock this line exactly as written — do not change it: [URL]"). Blog posts in the wild will surface 2–4 decisions per run depending on how strong the voice is in the draft.

Length-cap insight. Hard cap was set in characters (not words) for this run. The hive landed at 2,795 chars and stayed there from R3 onward. For voice-driven posts, the hive self-bounds at the length the prose wants to be — set the cap at your platform's actual limit (LinkedIn, Medium, etc.) as a safety rail rather than as an active constraint.

For comparison — the v1.0 run (without Reference Material approach): ≈13 minutes · 16 rounds (measured). Refining a strong existing draft with the source IN the Starting Document. Voice-driven posts take more rounds when reviewers tune phrasing and rhythm sentence by sentence on a draft they're committed to. Use this approach when you want to preserve specific phrasing from an existing draft; use the v2.0 approach (Reference Material + scratch) when you want to converge fast.
Best AIs For
Claude — voice & readability Grok — edge & personality Perplexity — factual grounding ChatGPT — structure
💡

If the voice starts sounding generic after a few rounds, open Notes and describe the author's style in a sentence — even a few adjectives like "direct, slightly sarcastic, uses short sentences" will steer the hive back toward a distinct voice.

↑ Back to top
💼
LinkedIn Post
Write a short-form LinkedIn post — a lesson, a hot take, or a war story — that reads as a peer talking to peers, not a thought leader trying to grow an audience.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Replace anything in [brackets] with your own information.

FieldWhat to enter
Document type *LinkedIn Post
Target audience *My LinkedIn network — be specific: e.g. Wireless engineers, IT managers, recruiters in my field, Former colleagues from my last industry, current peers in this one. They scroll fast. If the first two lines don't land, they keep scrolling.
Desired outcome *Reader stops scrolling, reads the whole post, and either comments or sends a connection request. Post should feel earned, not braggy.
Scope & constraintsCover one specific topic with a clear opinion or lesson. Hook in the first line. Use real numbers and specifics where you have them. End with a question that invites comments. Do not name a current employer if there are confidentiality or OPSEC concerns.
Tone & voiceConversational, direct, no LinkedIn-influencer voice. Sounds like a peer talking to peers, not a thought leader trying to grow an audience.
Additional instructionsNo "I'm proud to announce," "thrilled," or "humbled" — the LinkedIn opener virus. No emojis except maybe one at the end. No call to "DM me to learn more." No buzzwords (synergy, leverage, revolutionize). If a paragraph reads like a brochure, cut it.
Length Constraint2000 characters (LinkedIn's hard cap), hardcap mode
Starting from scratch? Drop your topic, real numbers, and the lesson you want to land into Setup 4 Reference Material:
Setup 4 — Reference MaterialPaste the topic, the specific facts and numbers you want to anchor the post around, the lesson or opinion you want it to land on, and the question you'd like to end with. The hive uses Reference Material as source-of-truth — anything you put here will be reflected directly in the post; anything you don't put here, the AIs will not invent.
Real-world example — LinkedIn Post on DFS in defense, 2 rounds, 3-AI budget hive
A wireless-engineering hot take on abandoning DFS in defense environments: the 16-DFS-channel math problem (only 9 clean channels left in 5 GHz), the inevitability of co-channel contention in 952k sq ft mixed-use buildings, and a credit to Peter MacKenzie's "CCC vs CCI" reframe. Real voice, real opinion, no employer-naming. Run with the same 3-AI budget hive used for the Bay Area test (paid Claude + paid ChatGPT + free Gemini-as-Builder). Reached majority convergence at Round 2 (ChatGPT and Gemini both NO CHANGES NEEDED, Claude offered three minor concision edits not adopted). All Project-screen prompts below are the literal values used in the run.
Project nameLinkedIn Post — 6 Months in and what I have learned about DFS
Versionv1.0
Document typeLinkedIn Post
Target audienceMy LinkedIn network (wireless engineers, former colleagues from the cable industry I was in for 30 years, some assorted high school classmates, IT people, and a few recruiters)
Desired outcomeReader stops scrolling, reads the whole post, and either comments or sends a connection request
Scope & constraintsCover: What is DFS, who uses it and why. How many channels does 5 GHz have and how many are DFS. Do not cover any other frequency. Explain CCI and how those of us in the industry are now calling it CCC because it isn't "really interference" you are mitigating. Explain how it is now inevitable in a mixed-use (office, production, warehouse) large building (some as big as 952k sq ft). My company is in defense and I abandoned DFS immediately after being hired.
Tone & voiceTechnical, direct, a bit sarcastic but willing to teach
Additional instructionsWe now call CCI "CCC" thanks to my friend Peter MacKenzie (@mackenziewifi)
Length Limit2000 characters, hardcap
Hive compositionBuilder: Gemini (free tier). Reviewers: ChatGPT (paid), Claude (paid), Gemini-as-reviewer (free). All other AIs toggled off via the per-session bee toggles on the work screen.
Starting DocumentFrom Scratch (no draft pasted) — the post was generated entirely from Setup 3 fields and the Tone direction
Why this converged in 2 rounds:
The previously-measured LinkedIn Post run on an internal AI server took 5 rounds. The same 3-AI public-API hive on a more sharply-specified prompt converged in 2. Two factors compounded: a tightly-scoped Project field set (the Scope listed exactly what to cover and what not to cover — no other frequencies, no other topics), and a real opinion in Tone ("a bit sarcastic but willing to teach") that gave reviewers a thesis to build around rather than a vague topic to explore. The lesson worth taking forward: short-form social posts converge fastest when the writer brings a specific opinion, names what they will and won't cover, and trusts the hive to shape the prose around that frame.
Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)
Refining a draft
Go to Setup 5 — Starting Document. Paste your existing post draft using the Paste Text button. Even a rough first attempt with bullet points works — the hive will shape it.
Starting from scratch
Go to Setup 5 — Starting Document. Click Start from Scratch and leave the text box empty. All your content lives in Setup 4 Reference Material — the topic, the facts, the lesson, the question. The hive builds the post from there.
Convergence
≈1.2 minutes of hive processing across 2 rounds (measured, not estimated — sum of per-round max-reviewer + Builder elapsed). Round 2 reached majority convergence (2 of 3 AIs satisfied, Claude as lone holdout with minor concision suggestions). Final output: 304 words / 1857 characters — 92% of the 2000-character hardcap, length-guard never triggered. SCRATCH-mode short-form converges fast when Project-screen scope is sharply specified; a vaguer prompt on the same category previously took 5 rounds. Plan for 2–6 rounds depending on how tightly your Scope and Tone bound the post.
Best AIs For
ChatGPT — hook crafting Claude — voice authenticity Gemini — concision & structure
💡

If you work in a sensitive industry (defense, healthcare, finance, legal), keep employer specifics out of the post. "My company is in defense" is fine; naming the company, the building square footage, the specific platforms, or active vendor evaluations is not. The DFS post above worked because it discussed a universal RF problem with attribution to a public industry figure (Peter MacKenzie) — none of the content was company-confidential. State your OPSEC constraints in Scope & constraints explicitly: "Do not name my employer. Do not reference any active project."

↑ Back to top
🖥️
Presentation Outline
Build a slide-by-slide speaker outline ready to drop into PowerPoint, Keynote, or Slides.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Replace anything in [brackets] with your own information.

FieldWhat to enter
Document type *Presentation outline — speaker notes format
Target audience *The audience for the talk — e.g. C-suite leadership team, New hire onboarding group, External conference attendees
Desired outcome *A slide-by-slide outline ready to copy into a presentation tool. Each slide must have: a slide title, 3–5 speaker note bullets, and one suggested visual or data point. Open with a strong hook, close with a clear call to action.
Scope & constraints[X]-minute talk. [Number] slides maximum. Bullet points only — no prose paragraphs. This is a speaker outline, not a script. WaxFrame outputs text — you will paste this into your presentation tool separately.
Tone & voice[Informative / persuasive / conversational] — match the formality level of the audience
Additional instructionsDo not write full sentences in the speaker bullets. Keep each bullet to one line. Do not add slides beyond the count I specify.
Starting from scratch? Change these fields:
Desired outcomeGenerate a complete slide-by-slide presentation outline for a [X]-minute talk on [topic]. Each slide: title, 3–5 speaker note bullets, one suggested visual. Open with a hook, close with a call to action. Build it from the topic list I provide in Notes.
Real-world example — Slide deck outline that converged in 8 rounds
From-scratch run with a strong Reference Material brief. The brief contains both the recommendation and the supporting reasoning, so the hive isn't being asked to invent the position — it's being asked to structure that position into 12 executive-ready slides under a 45-minute meeting constraint. Word count never moved more than 15 words across the 8 rounds (749–763 word range): the structure converged in Round 1 and rounds 2–8 were language refinement. An earlier v1.0 run with looser scope language ran 40 rounds without the system stopping on its own — tightening Scope to "Maximum 12 slides" and giving Tone a specific direction ("Executive-ready, direct, no hedging") is what made v3.0 converge cleanly.
Project nameDeck Outline — Wi-Fi 7 Readiness
Versionv3.0
Document typePowerpoint Presentation
Target audienceIT leadership at a 1,200-person professional services firm — CIO, Director of Infrastructure, and two senior architects, sitting in a 45-minute meeting
Desired outcomeAudience leaves with a clear answer to one question: should we include a Wi-Fi 7 pilot in our FY27 budget? The outline should end with a concrete recommendation, not a list of considerations.
Scope & constraintsMaximum 12 slides. Opening slide is the question, closing slide is the recommendation with a dollar figure. No "agenda" slide. No "thank you" slide.
Tone & voiceExecutive-ready, direct, no hedging. If the answer is "not yet," say "not yet" and explain why.
Additional instructionsStructure the information to fit on slides. Don't write speaker notes I'll do that later
Length ConstraintLeave blank — the "Maximum 12 slides" line in Scope is the cap that mattered
Starting DocumentClick Start from Scratch and leave the text box empty
Reference Material payload (paste into Setup 4 — Reference Material):
Driving question: "Should we pilot Wi-Fi 7 in FY27, or wait?"

Recommendation: pilot in FY27 at a single floor of HQ, budget $95k, defer enterprise-wide deployment to FY28 or FY29 depending on client device penetration.

Reasoning:
- Wi-Fi 7 standard is ratified. Enterprise APs are shipping in volume from Cisco, Aruba, Juniper, Extreme.
- Client device penetration is still low — under 12% of the user base has a Wi-Fi 7 capable endpoint as of Q1 2026. No reason to refresh the fleet now.
- HQ 14th floor is being refreshed in FY27 anyway due to an expiring lease remodel. That gives a "free" pilot footprint.
- Pilot goals: validate real-world throughput gains, test roaming behavior with mixed Wi-Fi 6E and Wi-Fi 7 clients, train the team on the new management platform.
- If pilot is successful, shovel-ready plan for FY28 enterprise rollout. If not, $95k spent to learn instead of $2.1M to fail.

Context the deck must establish:
- Where we are today: Wi-Fi 6E, Aruba Central managed, last refreshed FY24
- Where the industry is going: Wi-Fi 7 ratified, deployment guidance from Gartner suggests "selective pilots in FY26/FY27, mainstream in FY28"
- Client device mix: 70% corporate-managed laptops, 25% BYOD phones, 5% IoT/sensors

Hard constraints:
- FY27 budget cycle closes in July 2026. Decision must happen before then.
- Any dollar figure in the deck is directional — detailed pricing comes from the RFP, not this deck.
Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)
Refining a draft
Go to Setup 4 — Reference Material if you have source data (statistics, quotes, prior decisions, supporting analysis) the deck must reference but not edit. Then on Setup 5 — Starting Document, paste your existing outline using the Paste Text button. The hive cannot browse the web — anything the deck must reference must be pasted into Reference Material first.
Starting from scratch
Go to Setup 4 — Reference Material and paste your driving question, recommendation, supporting reasoning, hard constraints (dates, dollar figures), and any context the deck must establish. Then on Setup 5 — Starting Document, click Start from Scratch and leave the text box empty. Optionally, before Round 1, open the Notes drawer (📝 Notes in the top bar on the Work screen) for any one-shot Round 1 emphasis.
Convergence
≈31 minutes · 8 rounds (measured, not estimated) — slide-deck outlines converge fast when the Reference Material brief contains both the recommendation and the supporting reasoning. The recent run (Deck Outline — Wi-Fi 7 Readiness v3.0) reached majority convergence at Round 8 with word count holding steady in the 749–763 range across all 8 rounds. An earlier v1.0 run with the same brief but looser scope language ran 40 rounds without converging — Scope and Tone specificity are the levers that distinguish the two outcomes.
Best AIs For
ChatGPT — slide structure Claude — narrative arc Grok — audience hook Perplexity — current stats & data
💡

WaxFrame outputs plain text. Once the outline is where you want it, export it and use it as your script layer when building the actual slides in PowerPoint or Keynote. Each slide's bullet points become your speaker notes.

↑ Back to top

Personal & Everyday

Recipes, personal letters, and everyday writing where polish and tone matter but the stakes are friendly.

🍳
Recipe
Turn rough notes or a draft recipe into a polished write-up with clear instructions and consistent formatting.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Replace anything in [brackets] with your own information.

FieldWhat to enter
Document type *Recipe
Target audience *Who will cook from this — e.g. Home cook, beginner level, Experienced baker, Meal prep audience, intermediate skill
Desired outcome *Anyone following this recipe can cook the dish successfully on the first try. Instructions are clear, ingredient quantities are precise, and steps are in a logical order with nothing assumed.
Scope & constraintsDo not change the core recipe — only clarify and improve what is there. Ingredient quantities and cooking times must stay as written. Include: ingredient list with quantities, numbered steps, and at least one tip on substitutions or storage.
Tone & voice[Warm and conversational / precise and technical / beginner-friendly] — pick one
Additional instructionsDo not substitute ingredients without flagging it as an optional variation. Do not add unverified cooking times or techniques — flag anything uncertain instead of inventing it.
Starting from scratch? Change these fields:
Desired outcomeGenerate a complete recipe for [dish name] from the notes I provide. Include an ingredient list with quantities, numbered steps, and at least one tip on substitutions or storage. Do not add unverified techniques or cooking times.
Scope & constraintsCuisine or tradition: [e.g. Southern Italian, Korean BBQ, or leave blank]. Servings: [X]. Build only from the ingredients and details I provide.
Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)
Refining a draft
Go to Setup 5 — Starting Document. Paste your recipe using the Paste Text button. Include the ingredient list, quantities, and steps — even rough notes work. The AIs will shape it into a proper recipe format.
Starting from scratch
Go to Setup 5 — Starting Document. Click Start from Scratch and leave the text box empty. Before Round 1, open the Notes drawer and paste: the dish name, cuisine or tradition, number of servings, the key ingredients you want used, and any dietary constraints.
Rounds
Quick Start (cookies): 2 rounds (measured) — the bundled Quick Start template is tightly scaffolded and converges fast.

Recipe with Reference Material approach: 4 rounds (measured) — Publix-style Southern potato salad, May 2026. 5-AI hive (ChatGPT, Claude, Gemini, Grok, Perplexity) with DeepSeek as Builder. The source recipe (2,636 characters from a Publix copycat write-up) was pasted into Setup 4 — Reference Material instead of being used as a Starting Document. The Project setup carried the full identity scaffold: audience ("People who love Southern Style Potato Salad"), outcome (clear, precise, first-try success), scope (preserve core recipe — clarify only), tone ("warm and conversational with a southern drawl"), and Notes ("flag substitutions as optional variations; don't invent cooking times"). Word trajectory was 1243 → 1238 → 1226 → 1226 across the four rounds — minimal trimming, all the convergence work happened on prose quality and one mid-stream USER DECISION about bell peppers (resolved via a Builder-Only round between R2 and R3). Final 1226 words, 47 minutes wall-clock.

Why Reference Material over Starting Document for recipes: when the source recipe goes into the Starting Document, the hive treats it as a draft to refine — and recipes are unusually subjective, so reviewers debate technique, quantities, and wording extensively. Putting the source into Reference Material instead, with an empty Starting Document, lets the hive write fresh from the scaffold while pulling facts from the reference. The hive isn't fighting an existing draft; it's building one. Convergence is faster because there's less to argue about — the prose is whatever the Builder wrote in Round 1, not a reviewer's prior commitment.

General recipe refinement (no Reference Material): 5–20+ rounds (estimated) — when refining a non-templated recipe with detailed Notes but the source recipe sits in the Starting Document, AIs debate technique, quantities, and wording extensively. The estimate range is wide because recipes are unusually subjective: ingredient ratios, technique opinions, and wording preferences all surface as conflicts. Tighten Scope & constraints to constrain the conversation (e.g., "no nuts, no spices beyond salt and vanilla, do not change the butter-to-flour ratio").
Best AIs For
Claude — clarity & flow ChatGPT — culinary knowledge Gemini — technique tips
💡

If you want the recipe to stay true to a specific cuisine or tradition, state it explicitly in Scope & constraints: "Keep this authentic to Southern Italian technique — do not substitute ingredients or modernise methods." Without this the AIs will suggest variations freely.

↑ Back to top
🧾
Contractor / Vendor Letter
Refine a letter to a contractor, vendor, or service provider — invoice disputes, scope concerns, punch-list items, or anything where you need to be clear, professional, and firm without escalating.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Replace anything in [brackets] with your own information.

FieldWhat to enter
Document type *Letter to my [contractor / vendor / service provider] regarding [the first invoice / scope dispute / punch-list items]
Target audience *The [contractor / vendor / project manager] who did the work — be specific about the relationship: The general contractor we hired, The vendor that installed our HVAC
Desired outcome *A clear, professional record of my concerns and the items I need addressed before [I pay / we close out the project / we move forward]. The recipient understands what is wrong, what I expect, and what comes next.
Scope & constraintsInclude all of the items I have listed including pricing but don't make anything else up and don't change my dollar figures — they are all actual. Keep my factual claims (dates, amounts, names) intact.
Tone & voiceProfessional and courteous but firm. Not adversarial, not apologetic.
Additional instructionsDo not soften my concerns. Do not invent context, dates, or amounts I haven't provided. If a section is unclear, suggest a rewording rather than guessing what I meant.
Real-world example — Bay Area First Invoice Concerns, 12 rounds, 3-AI budget hive
A homeowner-to-contractor letter raising concerns about a first invoice on a multi-room remodel: receipts for marked-up materials, a 75% cost overrun on a flat-rate quote, and a punch list of installation issues (missing shelves, leaky shut-off valves, misaligned shower head, unilateral shower-door selection). Run with a deliberately budget-tier hive: paid Claude + paid ChatGPT + free Gemini-as-Builder. Reached majority convergence at Round 12 (Claude holdout, ChatGPT and Gemini satisfied). All Project-screen prompts below are the literal values used in the run — no AI-generated phrasing. Document tightened from 735 words to 624 (-15%) over 11 substantive rounds, monotonic compression with no bouncing.
Project nameBay Area First Invoice Concerns
Versionv3.0
Document typeLetter to my contractor regarding the first invoice
Target audienceThe contractor that did the work
Desired outcomea better understanding of my concerns about the costs and the work performed and an explanation of items that need to be addressed before I pay him
Scope & constraintsInclude all of the items I have listed including pricing but don't make anything else up and don't change my dollar figures they are all actual
Tone & voiceProfessional and courteous but firm
Additional instructions(blank — Notes left empty for this run)
Length ConstraintLeave blank
Hive compositionBuilder: Gemini (free tier). Reviewers: ChatGPT (paid), Claude (paid), Gemini-as-reviewer (free). All other AIs toggled off via the per-session bee toggles on the work screen.
Starting DocumentExisting draft pasted in (refining, not from scratch)
Why this configuration matters:
A 3-AI hive with one paid OpenAI seat, one paid Anthropic seat, and a free Gemini tier is the realistic floor for a budget-conscious WaxFrame user. This run validates that the configuration produces real convergence on a personal letter in a reasonable time window — about 7 minutes of hive processing across 12 rounds, sub-$1 in API spend across ChatGPT and Claude (Gemini contributed at zero cost as both reviewer and Builder). Each AI brought a different strength to the work: ChatGPT consistently flagged concision opportunities (line-3 phrasing was its signature concern across multiple rounds), Claude focused on semantic precision and tense consistency (catching a transcription error and recovering from a tense regression it had introduced two rounds earlier), and Gemini stayed quiet most rounds but surfaced sharp catches when it spoke up — the difference between "hand towel and a towel rack" and "hand towel holder and a towel rack" was Gemini-only.
Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)
Refining a draft
Go to Setup 5 — Starting Document. Paste your existing letter using the Paste Text button. Don't worry about polish — even a rough first pass with bullet-point concerns and dollar figures works. The hive will shape it into a coherent letter while preserving your factual content.
Starting from scratch
Go to Setup 5 — Starting Document. Click Start from Scratch and leave the text box empty. Before Round 1, open the Notes drawer and paste: who you're writing to, the project / job in question, the specific concerns (with any dates, amounts, and item names), and what you want to happen next.
Convergence
≈7 minutes of hive processing across 12 rounds (sum of per-round max-reviewer + Builder elapsed times — reviewers run in parallel within a round, so the round wall-time is bounded by the slowest reviewer plus the Builder, not the sum of all reviewer calls). Round 12 reached majority convergence (2 of 3 AIs satisfied) with a single holdout offering minor wording suggestions. Document compressed from 735 to 624 words (-15%). Most aggressive tightening happened in Rounds 6–7 (-60 words combined); late rounds 10–11 trimmed under 1% per round, signaling genuine end of useful iteration. Total wall-clock varies based on how much time you spend reviewing each round's output between Smoke clicks — the hive processing time is a floor, not a ceiling. Plan for 7–10 minutes of attentive engagement per multi-round refine session, more if you're reading every reviewer suggestion in detail.
Best AIs For
ChatGPT — concision Claude — semantic precision Gemini — sharp catches when active
💡

State your factual constraints explicitly in Scope & constraints: "Don't change my dollar figures — they are all actual." Without this, AIs will round, rephrase, or "improve" specific numbers and dates in ways that quietly invalidate your record. The literal language above kept all dollar amounts and the 75% overrun calculation intact across 12 rounds.

↑ Back to top

Reviews & Recommendations

Restaurant, hotel, business, and service reviews. Useful when the goal is to turn a real experience into a fair, specific, practical review that helps the next person decide what to do.

🍽️
Restaurant Review
Create or refine a useful restaurant review covering food, service, atmosphere, value, logistics, and whether you'd return.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Replace anything in [brackets] with your own information. Fields marked * are required to continue.

FieldWhat to enter
Document type *Restaurant review
Target audience *People deciding whether this restaurant is worth visiting. They want practical details about food, service, atmosphere, value, and logistics, not vague praise or complaints.
Desired outcome *Create a useful, honest review that explains what the experience was actually like, what was ordered, what was good, what was disappointing, whether the price made sense, and whether I would return.
Scope & constraintsInclude visit context, food and drinks ordered, pricing if known, service, atmosphere, cleanliness, parking or location notes, standout items, disappointments, and final recommendation. Do not invent dishes, prices, staff names, dates, or facts that were not provided.
Tone & voiceConversational, detailed, fair, practical, and direct. Preserve the reviewer's natural voice. Honest criticism is fine, but avoid making it sound like a rant unless the source material genuinely supports that tone.
Additional instructionsEnd with a clear bottom line: whether I would return, who this restaurant is best for, and any specific warning, recommendation, or timing advice for future visitors.
Length Constraint
(field below the goal fields)
Default 500 · select Words. For Google Maps use 750–1,200 · Characters. For TripAdvisor leave blank for full long-form.
Starting from scratch? Change these fields:
Desired outcomeGenerate a complete restaurant review from the details I provide in Reference Material (Setup 4). Organize the review around the actual visit, what I ordered, food quality, service, atmosphere, value, and final recommendation.
Additional instructionsIf important details are missing, write around them without inventing. Do not add fake menu items, fake prices, fake staff names, or fake dates.
Setup — Step 4 of 5 — Reference Material
Refining a draft
Skip this screen — click Continue. Your existing review goes in Setup 5 below.
Starting from scratch
Click Paste Text. Paste the detail checklist below, fill in the lines with what you remember from the visit, then continue to Setup 5.

Reference Material persists across every round and is never altered by the hive — exactly the right home for raw visit facts. Notes is for one-shot mid-round Builder tweaks (like "tighten the third paragraph"), not setup facts.

Detail checklist — copy and paste into Reference Material:
Restaurant:
Location:
Date/time:
Who was there (solo / couple / family / group / business):
Dine-in / takeout / delivery / patio / bar:
Reservation or walk-in:
Parking notes:
First impression of exterior/interior:
Noise / lighting / seating / cleanliness:
What was ordered (drinks, apps, entrées, sides, dessert):
Prices remembered:
Food quality (flavor, temperature, portion, freshness, presentation):
Best item:
Worst item or disappointment:
Service quality:
Problems and how staff handled them:
Would you return:
Who is this restaurant best for:
Setup — Step 5 of 5 — Starting Document screen
Refining a draft
Go to Setup 5 — Starting Document. Click Paste Text and paste your existing review.
Starting from scratch
Go to Setup 5 — Starting Document. Click Start from Scratch and leave the text box empty. The hive will use your filled-in Reference Material from Step 4 to write the review from scratch.
A good restaurant review usually covers
Visit context · arrival / first impression · atmosphere and logistics · food and drink ordered · food quality and portion/value · service quality · problem handling, if any · standout positives · standout negatives · bottom-line recommendation.
Convergence
≈14 minutes · 5 rounds (measured 2026-05-10) — refining an existing v1.0 draft of a restaurant review (Pearl's Saltwater Grille v2.0, Savannah, GA) with no Reference Material attached. 6-AI hive (ChatGPT, Claude, Gemini, Grok, Perplexity as reviewers; DeepSeek as Builder). Final 392 words / 2,179 characters — down from 459 words in the original draft, 14.6% tighter. Majority convergence at Round 5 with 3 USER DECISIONS surfaced and resolved across two Builder-Only rounds (1 conflict from R1 resolved in R2; 2 conflicts from R3 resolved in R4).
Real-world example — Restaurant review that took 5 rounds (REFINE on existing draft)
Run these exact values yourself to reproduce the convergence behavior documented above. Use the values literally on a first run to learn the rhythm; on later runs, swap the specifics for your own restaurant.
Project nameReview - Pearl's Saltwater Grille
Versionv2.0
Document typeRestaurant review
Target audiencePeople deciding whether this restaurant is worth visiting. They want practical details about food, service, atmosphere, value, and logistics, not vague praise or complaints.
Desired outcomeCreate a useful, honest review that explains what the experience was actually like, what was ordered, what was good, what was disappointing, whether the price made sense, and whether I would return.
Scope & constraintsInclude visit context, food and drinks ordered, pricing if known, service, atmosphere, cleanliness, parking or location notes, standout items, disappointments, and final recommendation. Do not invent dishes, prices, staff names, dates, or facts that were not provided.
Tone & voiceConversational, detailed, fair, practical, and direct. Preserve the reviewer's natural voice. Honest criticism is fine, but avoid making it sound like a rant unless the source material genuinely supports that tone.
Additional instructionsEnd with a clear bottom line: whether I would return, who this restaurant is best for, and any specific warning, recommendation, or timing advice for future visitors.
Length ConstraintRange — 200 to 1,500 words
Starting DocumentUpload (or paste) the v1.0 draft below into Setup 5 — Starting Document
Starting from scratch? Change these fields:
Desired outcomeGenerate a complete restaurant review from the details I provide in Reference Material (Setup 4). Organize the review around visit context, food and drinks, service, atmosphere, value, and final recommendation.
Additional instructionsIf important details are missing, do not invent them. Focus on the details provided and flag missing items only if needed.
Paste this as the starting document (the v1.0 draft refined in the run above):
Savannah, GA

" A romantic dining establishment directly on the water"

I made reservations at this location online for Saturday 8:30 PM through the OpenTable app. In the notes I asked for a booth. When I arrived I was told that a table was not quite yet ready and I could head to the bar. We did so and just as we were ordering a drink we were told that the table was ready. It would seem that this location does not have any booths but we were seated right by the window on a four top.

For entrées we ordered the shrimp and grits and stuffed fish with crab. I had glanced over a few reviews so I ordered a cup of she crab soup to try it out because all of the reviews rave about it. It did not disappoint! Additionally we were brought out a basket of hush puppies that were perfectly golden brown and served up with an herb butter that was delicious. Truthfully, these were the best hush puppies I've had in years. The flavor was perfect and they were not dry at all. For a little variety I even dunked them in to the soup. I really couldn't stop eating them and there were quite a few in the basket.

Our entrées came really quickly. My fish was perfectly prepared and not only was it stuffed with crab but there was additional crab next to the Rice that easily amounted to quite a few ounces of crab I was pretty impressed. The only change I made to the dish was scrapping the asparagus in lieu of the mixed vegetables, including zucchini, squash, and carrots.

The shrimp and grits contained decent sized shrimp wrapped with bacon and an occasional bit of andouille sausage. The grits were creamy and delicious as well.

Although I was relatively stuffed I ordered the peach cobbler. I must preface by saying that I truly love desserts and this did not disappoint! It was served warm in a special ramekin with crumbly cobbler and whole peach slices. Ice cream was served on the side which was perfect topped with whipped cream. Needless to say I made room!

As for the ambience, the restaurant is of medium noise level the tables are not really isolated all that much although there is some separation and each table does have a candle on it which is a nice touch of course. Just before we received our entrée, we were able to go out onto the deck and take some really nice pictures even though the sun had already gone down.

If am ever in the area again I will definitely revisit this fine restaurant.
What this run produced: 392 words / 2,179 characters · majority convergence at Round 5 · 3 USER DECISIONS surfaced and resolved via two Builder-Only rounds (R2, R4) · 6-AI hive (DeepSeek as Builder; ChatGPT, Claude, Gemini, Grok, Perplexity as reviewers) on WaxFrame v3.36.30 Pro · 14:07 wall-clock.
Best AIs For
Claude — natural voice and fairness ChatGPT — structure and polished rewrites Gemini — practical completeness and missing-detail detection Grok — conversational tone and punchier phrasing
💡

The best restaurant reviews aren't just about whether the food was good. They tell the reader what to expect: parking, seating, noise, service pacing, portion size, prices, and whether the place is worth the money. Specifics outperform adjectives.

↑ Back to top
🏨
Hotel Review
Create or refine a hotel review covering room, sleep quality, location, amenities, service, value, and dealbreakers.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Fields marked * are required to continue.

FieldWhat to enter
Document type *Hotel review
Target audience *Travelers deciding whether to book this hotel — especially business travelers, families, road-trippers, or people comparing nearby properties.
Desired outcome *Create a detailed, practical hotel review that helps readers understand the room, sleep quality, location, amenities, service, value, and any problems that affected the stay.
Scope & constraintsCover trip context, room type or room number if provided, rate/value, check-in, room layout, cleanliness, bed, bathroom, HVAC, noise, darkness, internet, breakfast, gym, pool, bar, parking, location, staff, and final recommendation. Do not invent amenities, prices, loyalty benefits, room numbers, or facts.
Tone & voicePractical, detailed, fair, and conversational. Preserve useful personal observations and specific traveler-focused details.
Additional instructionsInclude whether I would stay again and what type of traveler this hotel is best suited for. Mention dealbreakers clearly.
Length Constraint
(field below the goal fields)
Default 800 · select Words. For Google Maps use 750–1,200 · Characters. For TripAdvisor leave blank for full long-form.
Starting from scratch? Change these fields:
Desired outcomeGenerate a complete hotel review from the details I provide in Reference Material (Setup 4). Organize the review around the stay context, check-in, room, sleep quality, bathroom, amenities, location, value, and final recommendation.
Additional instructionsIf important details are missing, do not invent them. Focus on the details provided and flag missing items only if needed.
Setup — Step 4 of 5 — Reference Material
Refining a draft
Skip this screen — click Continue. Your existing review goes in Setup 5 below.
Starting from scratch
Click Paste Text. Paste the detail checklist below, fill in the lines with what you remember from the stay, then continue to Setup 5.

Reference Material persists across every round and is never altered by the hive — exactly the right home for raw stay facts. Notes is for one-shot mid-round Builder tweaks (like "tighten the third paragraph"), not setup facts. Writing a review weeks after the trip? The checklist scaffolds your memory by prompting for the specifics that matter — bed, noise, HVAC, blackout curtains, hot water, Wi-Fi, parking, fees — the stuff that fades fast.

Detail checklist — copy and paste into Reference Material:
Hotel:
Location:
Dates of stay:
Trip type (business / vacation / family / event / road trip):
Who was there (solo / couple / family / group):
Room type and number if relevant:
Rate / points / resort fee / parking fee / other fees:
Check-in experience:
Staff interactions:
Room size, layout, furniture, outlets, desk:
Cleanliness and maintenance:
Bed and pillows:
Noise (hallway, street, airport, neighbors, elevators, kids):
Room darkness / blackout curtains:
HVAC / temperature control:
Bathroom layout, shower pressure, hot water, towels, toiletries:
Wi-Fi speed and reliability:
Breakfast quality and variety:
Gym / pool / bar / lounge / laundry / shuttle / parking:
Nearby restaurants / walkability / transit / airport / attractions / work site:
Problems and how staff handled them:
Would you stay again:
Who is this hotel best for:
Real-world example — Hotel review that took 19 rounds
Run these exact values yourself to reproduce the convergence behavior documented above. Use the values literally on a first run to learn the rhythm; on later runs, swap the specifics for your own stay.
Project nameReview — Melbourne Marriott Hotel
Versionv1.0
Document typeHotel review
Target audienceTravelers deciding whether to book this hotel — especially business travelers, families, road-trippers, or people comparing nearby properties.
Desired outcomeCreate a detailed, practical hotel review that helps readers understand the room, sleep quality, location, amenities, service, value, and any problems that affected the stay.
Scope & constraintsCover trip context, room type or room number if provided, rate/value, check-in, room layout, cleanliness, bed, bathroom, HVAC, noise, darkness, internet, breakfast, gym, pool, bar, parking, location, staff, and final recommendation. Build only from the stay details in my Reference Material — do not invent amenities, prices, loyalty benefits, room numbers, or facts.
Tone & voicePractical, detailed, fair, and conversational. Preserve useful personal observations and specific traveler-focused details.
Additional instructionsInclude whether I would stay again and what type of traveler this hotel is best suited for. Mention dealbreakers clearly.
Length ConstraintNo limit (template default — set per-platform at publish time)
Starting DocumentClick Start from Scratch and leave the text box empty
Reference Material payload (paste into Setup 4 — Reference Material as "Paste Text" before Round 1):
Hotel: Melbourne Marriott Hotel
Location: Corner Exhibition and Lonsdale Streets, Melbourne
Dates of stay: 04/01/2026 - 04/03/2026
Trip type (business / vacation): I was traveling to Australia for the first time ever and it was a last minute work trip that I pulled together in 4 days.
Who was there: Solo
Room type and number if relevant: 1 Bedroom King Suite, Room 403
Rate / points / resort fee / parking fee / other fees: AUD $320-$350 a night credit card fee of $13.07
Check-in experience: I arrived early and was prepared to drop my bags and just take off exploring but they provided me a room instead!
Staff interactions: Excellent and Friendly staff who were expedient.
Room size, layout, furniture, outlets, desk: Very Spacious Suite view of the city corner and nightlife.
Cleanliness and maintenance: quite clean and well maintained property
Bed and pillows: Comfortable bed and I always ask for extra pillows and had no trouble getting them here.
Noise (hallway, street, airport, neighbors, elevators, kids): Very quiet
Room darkness / blackout curtains: It was probably almost a 9 with the bedroom door shut. With it open it was more like a 7 out of 10
HVAC / temperature control: excellent I sleep cold and this was great. I lowered it as low as it would go in celcius and it felt great
Bathroom layout, shower pressure, hot water, towels, toiletries: Water pressure on the shower was AMAZING definitely an incredible feature
Wi-Fi speed and reliability: wi-fi at my level is free for "gold" and I never had any issues connecting or using the wifi
Breakfast quality and variety: This breakfast was the highlight! It was the most amazing breakfast I have ever had in any Marriott property ANYWHERE. It was filled with traditional, and non-traditional items, 5 juices, pastries and danishes, fresh fruit, savory and sweet options I mean it was amazing the wait staff are super friendly and accommodating.
Nearby restaurants / walkability / transit / airport / attractions / work site: This is in the heart of downtown and you can walk to comedy clubs, shows, restaurants and parks. It's really in the thick of it. There is even a 7-11 directly across the street for last minute I forgot items.
Problems and how staff handled them:
Would you stay again: YES!
Who is this hotel best for: Business and pleasure travelers alike
What this run produced: 400 words / 2,488 characters · majority convergence at Round 18 · zero USER DECISIONS surfaced · 6-AI hive (DeepSeek as Builder; ChatGPT, Claude, Gemini, Grok, Perplexity as reviewers).
Setup — Step 5 of 5 — Starting Document screen
Refining a draft
Go to Setup 5 — Starting Document. Click Paste Text and paste your existing review.
Starting from scratch
Go to Setup 5 — Starting Document. Click Start from Scratch and leave the text box empty. The hive will use your filled-in Reference Material from Step 4 to write the review from scratch.
A good hotel review usually covers
Trip context · location and arrival · check-in and staff · room layout and cleanliness · sleep quality (bed, noise, darkness, HVAC) · bathroom and shower · internet and work-travel usefulness · breakfast and amenities · value for the rate · bottom-line recommendation.
Convergence
19 rounds (measured, not estimated) — Round 18 reached majority convergence; Round 19 was a builder-only finalize. Notable: zero USER DECISIONS surfaced across the entire run. The questionnaire-scaffold approach pre-answers most ambiguities reviewers would normally escalate, so the round count is driven by prose-polish rather than decision pressure points. Hotel reviews can therefore run long without any user input — expect to hit majority convergence between rounds 15 and 20, not the 3–5 rounds typical of shorter documents.
Best AIs For
Claude — voice and natural travel-review flow ChatGPT — structure and polished platform-ready rewrite Gemini — completeness and practical traveler details Perplexity — useful as reviewer for travel/logistics gaps, but do not let it invent facts
💡

For hotel reviews, sleep quality and hidden friction matter more than lobby marketing. Noise, HVAC, blackout curtains, hot water, Wi-Fi, parking, and fees are often what help the next traveler most.

↑ Back to top
🧾
Business / Service Review
Create or refine a review of a business, contractor, service provider, parking company, repair service, delivery experience, or other non-restaurant/non-hotel business.
Setup — Step 3 of 5 — Your Project screen

Fill in each field on the Project screen. Fields marked * are required to continue.

FieldWhat to enter
Document type *Business or service review
Target audience *People deciding whether to hire, visit, book, or use this business. They care about reliability, value, professionalism, communication, and how problems are handled.
Desired outcome *Create a fair but useful review that explains why I used the business, what happened, what went well, what went wrong, how the business handled it, and whether I would recommend them.
Scope & constraintsInclude reason for using the business, booking or arrival process, staff behavior, service quality, pricing/value, problems, resolution attempts, observed business practices, and final recommendation. Do not exaggerate, speculate beyond the facts, or invent details.
Tone & voiceClear, direct, specific, and fair. Honest criticism is allowed, but avoid sounding like a rant unless the source material genuinely supports it.
Additional instructionsEnd with practical advice: who should use this business, who should avoid it, and what to watch out for. Do not add legal conclusions, accusations, or claims beyond the facts provided.
Length Constraint
(field below the goal fields)
Default 500 · select Words. For Google Maps use 750–1,200 · Characters. For Yelp/TripAdvisor leave blank for full long-form.
Starting from scratch? Change these fields:
Desired outcomeGenerate a complete business or service review from the details I provide in Reference Material (Setup 4). Organize the review around why I used the business, what happened, staff behavior, service quality, value, problem handling, and final recommendation.
Additional instructionsDo not add legal conclusions, accusations, or claims beyond the facts provided. If the experience was negative, keep the review specific and evidence-based.
Setup — Step 4 of 5 — Reference Material
Refining a draft
Skip this screen — click Continue. Your existing review goes in Setup 5 below.
Starting from scratch
Click Paste Text. Paste the detail checklist below, fill in the lines specifically (quoted prices, dates, what was said, what happened), then continue to Setup 5.

Reference Material persists across every round and is never altered by the hive — exactly the right home for raw service-experience facts. Notes is for one-shot mid-round Builder tweaks (like "tighten the third paragraph"), not setup facts. "They quoted X, charged Y, did Z, and responded this way" is useful; "they ripped me off" is not — the more specific your details, the stronger and more defensible the review.

Detail checklist — copy and paste into Reference Material:
Business:
Location:
Type of service or product:
Why you used them:
Date/time:
Booking process (app / website / phone / walk-in / reservation / estimate / quote):
Price quoted vs. price paid:
Arrival/check-in process:
Staff behavior and communication:
What went well:
What went wrong:
Delays / confusion / unexpected charges / damage / poor workmanship / service failures:
How the business responded when a problem came up:
Whether they fixed the issue:
Observed practices future customers should know about:
Would you use them again:
Who should use them and who should avoid them:
Setup — Step 5 of 5 — Starting Document screen
Refining a draft
Go to Setup 5 — Starting Document. Click Paste Text and paste your existing review.
Starting from scratch
Go to Setup 5 — Starting Document. Click Start from Scratch and leave the text box empty. The hive will use your filled-in Reference Material from Step 4 to write the review from scratch.
A good business / service review usually covers
Why the reviewer used the business · booking / arrival / first interaction · what service was expected · what actually happened · staff professionalism and communication · price and value · problems and resolution · trust factor · practical advice for future customers · bottom-line recommendation.
Convergence
2–4 rounds typical (estimated, not measured). Round 1 builds a coherent timeline. Round 2 improves fairness, clarity, and usefulness. Additional rounds are useful for reducing emotional heat or shortening for Google Maps.
Best AIs For
Claude — fairness and human tone ChatGPT — clean structure and concise rewrites Gemini — missing-detail detection and practical completeness Grok — blunt but readable wording, useful when the source is too sterile
💡

Bad service reviews are strongest when they are specific. "They ripped me off" is weak. "They quoted X, charged Y, did Z, and responded this way" is useful. Keep the review grounded in what actually happened.

↑ Back to top
✈️
Trim to TripAdvisor
Convert a long-form review into a TripAdvisor-ready version: detailed narrative travel-context tone, 500–900 words. The source stays visible to reviewers every round so factual cuts can be verified against the original.

This is a refine-only template. It works on a review you already have — typically the output of the Restaurant, Hotel, or Business / Service playbook. You paste that source review in, and the hive shapes it into a TripAdvisor-format version. There is no "from scratch" path: write the review first with one of the review playbooks, then bring it here.

Setup — Step 3 of 5 — Your Project screen

These fields are pre-filled by the template. Listed here so you know what the hive is working toward.

FieldWhat the template sets
Document typeTripAdvisor review
Target audienceTripAdvisor readers deciding whether to visit or book based on detailed travel reviews — they want narrative arc, practical context, and an honest recommendation.
Desired outcomeProduce a TripAdvisor-ready review from the source: detailed narrative travel-context tone, 500–900 words, full experience arc from arrival to departure to recommendation.
Scope & constraintsUse only the facts from the source review. Preserve the chronological arc: arrival/booking context, the experience, the recommendation. Cut redundancy where helpful, but TripAdvisor readers expect detail — do not over-compress.
Tone & voiceDetailed narrative. Travel-context. Slightly more formal than Yelp, more thorough than Google Maps. Honest. Helpful. Specific.
Length Constraint
(field below the goal fields)
Range · 500–900 · Words (pre-set). TripAdvisor rewards detail — this is the longest of the three platform formats.
Setup — Step 4 of 5 — Reference Material

If your source review is significantly longer than 900 words, also paste it into Reference Material. That keeps the original visible to reviewers every round so they can verify facts and judge what got cut. If your source is already close to 500–900 words, Reference Material isn't necessary — the working document holds the same content. Reference Material persists across every round and is never altered by the hive; it's the right home for the source of truth.

Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)

Go to Setup 5 — Starting Document. Click Paste Text and paste your existing long-form review. The hive trims and reshapes it into the TripAdvisor format.

What a TripAdvisor version keeps
Location and booking context up front · the experience told chronologically (arrival → during → departure) · practical travel logistics (parking, location, room/table, view, reservation, wait time) · what worked and what didn't · a clear closing recommendation and who the place is best for. More detail than Google Maps, more structure than Yelp.
Convergence
2–4 rounds typical (estimated, not yet measured as a standalone run). Round 1 reshapes the source into the narrative arc and trims to range. Round 2 tightens travel-context detail and pacing. Additional rounds help if the version needs to be warmer, more specific on logistics, or closer to the original voice. A standalone Trim to TripAdvisor measurement is pending — the only TripAdvisor data on file came from the retired combined Multi-Platform template, so it isn't quoted here.
Best AIs For
Claude — preserving narrative voice while trimming ChatGPT — chronological structure and clean pacing Gemini — completeness check against the source, logistics detail Grok — punchier recommendation close
💡

Write your full, messy review once with the Restaurant / Hotel / Business-Service playbook, then run it through Trim to TripAdvisor. You get a platform-ready version without re-writing from scratch or losing the facts — and the source stays in Reference Material so nothing drifts.

↑ Back to top
📍
Trim to Google Maps
Convert a long-form review into a Google Maps-ready version: skim-friendly, bottom-line-first, 750–1,200 characters. The source stays visible to reviewers every round so the brutal cuts can be verified against the original.

This is a refine-only template. It works on a review you already have. Google Maps cuts are aggressive — typical source-to-target ratio is 5–10×, so the source belongs in Reference Material where reviewers can check that the cuts kept the right facts. There is no "from scratch" path here.

Setup — Step 3 of 5 — Your Project screen

These fields are pre-filled by the template. Listed here so you know what the hive is working toward.

FieldWhat the template sets
Document typeGoogle Maps review
Target audienceGoogle Maps readers scanning quickly for the bottom line — they want the recommendation up front, the key facts that support it, and to leave.
Desired outcomeProduce a Google Maps-ready review: skim-friendly, bottom-line-first, 750–1,200 characters. Lead with the recommendation. Keep only the few facts that actually support the verdict.
Scope & constraintsBrutal cut. Drop the chronological narrative arc. Lead with the bottom line. Keep only specific details that justify the recommendation. Strip everything else.
Tone & voiceDirect. Practical. Skimmable. First-person fine. No fluff, no marketing-speak, no corporate polish.
Length Constraint
(field below the goal fields)
Range · 750–1,200 · Characters (pre-set). Note that's characters, not words — Google Maps reviews are short.
Setup — Step 4 of 5 — Reference Material

Strongly recommended: paste the source review into Reference Material as well. Google Maps cuts are brutal — a typical 800+ word source loses 70%+ of its content. Without the source visible every round, reviewers can't evaluate whether the cuts kept the right facts. Reference Material persists across every round and is never altered by the hive.

Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)

Go to Setup 5 — Starting Document. Click Paste Text and paste your existing review. If it's well over 1,200 characters, the Source Size Check will fire an oversized recommend card on paste — that's expected. Move the long source into Reference Material and let the working document hold the trimmed version as it forms.

What a Google Maps version keeps
The recommendation in the first sentence · two or three specific facts that justify it · one practical tip a future visitor needs · a clean verdict close. Everything else gets cut. If a reader can't get the bottom line in five seconds, the version isn't done.
Convergence
≈23 minutes · 5 rounds (measured 2026-05-17, Manly Bikes Google v1.0). REFINE mode on an 888-word / 5,389-character source review, with the source moved into Reference Material after the oversized card fired. 7-AI hive (ChatGPT, Claude, Mistral, Grok, DeepSeek, Perplexity as reviewers; Gemini as Builder). DeepSeek dropped at R2 via the slow-responder card. R3 was the real debate round; R4 was Builder-Only — all six reviewers voted no changes, the textbook convergence signal — and R5 confirmed unanimous no-changes and locked. Final document 155 words / 1,003 characters (mid-range), a 5.37× trim. Zero USER DECISIONS.
Best AIs For
ChatGPT — concise, skimmable structure Claude — keeping the right facts through a brutal cut Gemini — verifying cuts against the source in Reference Material Grok — punchy bottom-line-first opener
💡

The measured run trimmed an 888-word review to 1,003 characters in 5 rounds — that's the job. Don't fight the aggressiveness; Google Maps readers genuinely won't read more than ~1,200 characters. Keep the source in Reference Material so the hive cuts hard and safely.

↑ Back to top
💬
Rewrite as Yelp
Rewrite a long-form review as a Yelp review: conversational, personality-forward, first-person, 300–700 words. Voice transformation is the main job — same facts, different prose register.

This is a refine-only template. It works on a review you already have. Unlike the trim templates, the main job here isn't cutting length — it's voice transformation: same facts, rewritten into a real Yelp register. There is no "from scratch" path here.

Setup — Step 3 of 5 — Your Project screen

These fields are pre-filled by the template. Listed here so you know what the hive is working toward.

FieldWhat the template sets
Document typeYelp review
Target audienceYelp readers — they value real voice, personality, specific lived details. They smell fake reviews and corporate polish from a mile away.
Desired outcomeRewrite the source review as a Yelp review: conversational, personality-forward, first-person, 300–700 words. Same facts, different prose register. Real voice.
Scope & constraintsVoice transformation is the main job. Same facts, different rendering. Casual phrasing OK. First-person fine. No corporate polish, no marketing-speak, no AI-generated blandness.
Tone & voiceConversational, first-person, personality-forward. Casual phrasing ("honestly," "tbh," "look —") is fine when it fits. Specific lived details over generic praise. Yelp readers reward authenticity.
Length Constraint
(field below the goal fields)
Range · 300–700 · Words (pre-set). Mid-length — enough room for voice to breathe without rambling.
Setup — Step 4 of 5 — Reference Material

Recommended: paste the source review into Reference Material. Voice rewrites preserve facts, but the new prose register can mask drift — "he gave us instructions for wet conditions" quietly becoming "he gave us a free poncho" is invisible without the source for comparison. Reference Material keeps the original visible to reviewers every round so they can catch silent fact-slip while the voice changes.

Setup — Step 5 of 5 — Starting Document screen (the next screen after Reference Material)

Go to Setup 5 — Starting Document. Click Paste Text and paste your existing review. If it's over the 700-word ceiling, the Source Size Check will fire an oversized recommend card on paste — move the long source into Reference Material and let the working document hold the Yelp rewrite as it forms.

What a Yelp version keeps
A conversational first-person opener · the reviewer's actual personality and quirks · specific lived details (the small ones that prove it's real) · honest highs and lows · a recommendation that sounds like a person, not a brand. The facts don't change — the register does.
Convergence
≈21 minutes · 8 rounds (measured 2026-05-17, Manly Bikes Yelp v1.0). REFINE mode on the same 888-word source (1.27× over the 700-word ceiling, so the Source Size Check fired on paste). 7-AI hive (ChatGPT, Claude, Mistral, Grok, DeepSeek, Perplexity as reviewers; Gemini as Builder), DeepSeek dropped at R2 via the slow-responder card. Majority convergence triggered at R8 with holdouts present. Final document 542 words / 3,142 characters — 60% of the 300–700 word range, the sweet spot. Round times descended steadily (7.6 → 2.4 → 4.9 → 1.2 → 3.3 → 0.9 → 0.7 → 0.5 min) as no-changes votes accumulated — healthy convergence. Zero USER DECISIONS. Voice rewrites take more rounds than trims because reviewers iterate on register, not just length.
Best AIs For
Grok — natural Yelp voice and personality Claude — authentic first-person register without losing facts Gemini — fact-drift detection against the source ChatGPT — keeping the rewrite specific instead of generically chatty
💡

The measured run landed at 60% of the Yelp range versus the retired combined template's 10% — same source, dramatically better positioning. The single-platform template gives voice room to breathe instead of squeezing it out under length pressure. If the rewrite reads corporate, smoke another round — the register usually loosens up by R3–R4.

↑ Back to top

WaxFrame

About

Version
LicenseAGPL-3.0 — open source, free to use and modify with attribution. Read license →
AuthorR David Paine III — weirdave.com
TestingCandy
StackVanilla HTML, CSS, JavaScript — no frameworks, no server, no install required.

Built with ❤️ by WeirDave and Claude.

WaxFrame Pro

Enter your license key to continue.

Don't have a key? Buy WaxFrame Pro →

Manage License

Your WaxFrame Pro license is active.

••••••••-••••••••-••••••••-••••••••