# Decision Council: argue a hard business call before you make it

> Bring one hard call: a price, an offer, a hire, a contract, a big spend. Five seats (Offers, Craft, Reach, Owned and Rules) argue it over three rounds, a script checks every number against their cards, and a compliance stop cannot be outvoted. Nothing happens until you ratify it at your own terminal.

- URL: https://paperchasewebb.ai/bots/marketplace/items/decision-council-argue-a-hard-business-call-before-you-make-it/
- Type: Prompt pack
- Category: Solo business
- By: PaperChaseWebb
- Works with: Claude Code, Codex, Grok Bot, Hermes (named in the listing); A skill, five seat files, rules for AGENTS.md and CLAUDE.md, Claude Code and Codex notes, Hermes notes and a Grok Bot standing order, with a Python checker. The rules files were checked by the checker on Linux; live sittings were not run inside each agent for this release
- Price: $39 USD
- Difficulty: Intermediate
- Setup time: 20 min
- Time to result: The worked example in 5 minutes; a first quick sitting in about 15 minutes; a full sitting in under an hour
- Format: Markdown (.md) file, decision-council-argue-a-hard-business-call-before-you-make-it.md
- Version: 1.0, updated 2026-11-01
- License: Use inside your own team or company, including for your clients. Do not resell or republish the kit itself. Keeps the MIT notice for the work it credits.

## What you end up with

Each big call ends with a verdict whose numbers match the cards, the strongest dissent kept as written, a cheapest test, kill criteria and a revisit date, and your yes tied to a hash before anything happens.

## What you get

- 21 sections in one .md file, with a one-line unpack for all 185 files
- The Chair's skill, step by step, for agents with and without sub-agents, plus full council and quick modes
- Five seat files (Offers, Craft, Reach, Owned, Rules), each with a pillar, four questions, a blind spot and its own idea of ruin, plus a template to add your own
- council.py: 15 brief checks, 19 card and round checks and 16 verdict checks; the math lines, label, support, vote, spread, confidence and overrides in the verdict must match the cards
- Overrides a vote cannot beat: a compliance stop, a ruin cap and a one-way door cap, with WAIT allowed only with a dated trigger
- ratify.py: you ratify at your own terminal by typing part of the verdict's hash and your name; agent sessions and piped input are refused, and --verify catches a verdict edited after the fact
- A rules check for AGENTS.md, CLAUDE.md or a standing order, and an eval set of 23 made-up sittings to test your own agent
- A full made-up worked example (a monthly print club), a quick sitting, an agent's first-try verdict with 14 findings, and the transcript of every command
- The test harness: 85 unit tests and 754 planted faults you can repeat

## Who it is for

Owners and operators who want a hard call argued from every side before they commit, with the arithmetic checked by a script, the dissent kept and their own yes required before anything happens.

## Tools needed

- Python 3.8 or newer: runs council.py (brief, card, round and verdict checks, the score, the rules check and the eval set) and ratify.py; standard library only, no network
- Grok Bot, Claude Code, Codex or Hermes: chairs the sitting and runs the seats under AGENTS.md or the Grok Bot standing order; never argues an option, never acts on a verdict and never ratifies one
- A folder per sitting: the brief, the cards for each round, the draft resolution and the verdict, saved as files
- You need: Python 3.8 or newer
- You need: An agent that can read files and run a script
- You need: The decision, the options, your resources and what success looks like by when
- You need: You, at your own terminal, to ratify

Cost to run: The kit is the only purchase. A full sitting with five seats is up to 15 seat runs on the agent plan you already use. The checker runs locally, makes no network calls and logs in to nothing.

## How it works

1. **Unpack and test.** Save the kit as pack.md in an empty folder, run the unpack line from Appendix A, then sh test_all.sh. Expected: 185 files written; 85 tests OK; 754 of 754 planted faults caught; the eval set matches 23 of 23; ALL PASS.
2. **Watch a sitting.** Run sh example/run_example.sh and read it next to section 17. Expected: The Larkspur brief clean, the scores A 5.3, B 7.3, C 3.8 and GO WITH CONDITIONS on B, the first-try verdict caught with 14 findings, both examples clean and a piped ratify refused.
3. **Set up your agent.** Put your name in settings.json, add the rules from agents/ to AGENTS.md, CLAUDE.md, Hermes notes or a Grok Bot standing order, and run python3 council/council.py rules on that file. Expected: 0 findings on the rules file.
4. **Bring your first call.** Tell your agent: take this to the council: should we ...? Let it build the brief, run the rounds and check the verdict, then read verdict.md and ratify it at your own terminal if you agree. Expected: A verdict with 0 findings, a cheapest test, kill criteria and a revisit date, and nothing done until you act.

## Sample prompt: Take it to the council (Grok Bot, Claude Code, Codex, Hermes)

```
Read AGENTS.md. Take this to the council: should we raise the monthly price from $X to $Y in January? Ask me for anything the brief needs first.
```

Prompt 1 of 6.

## Sample prompt: Quick council (Grok Bot, Claude Code, Codex, Hermes)

```
Quick council: should we try Sunday markets for four weeks? Three seats, round 1 only.
```

Prompt 2 of 6.

## Sample prompt: Check the verdict (Grok Bot, Claude Code, Codex, Hermes)

```
Run council.py check on sessions/[sitting] and fix the verdict until it prints 0 findings. Do not touch any card.
```

Prompt 3 of 6.

## Free preview of decision-council-argue-a-hard-business-call-before-you-make-it.md

````markdown
# Decision Council: argue a hard business call before you make it

Version 1.0, 2026-11-01. By PaperChaseWebb. For owners, operators and agency leads who have one hard call to make (a price, a new offer, a hire, a partner, a contract, a big spend) and want it argued from every side of the business before they commit, with the arithmetic checked and their own yes required at the end. Works with Grok Bot, Claude Code, Codex and Hermes. Python 3.8 or newer, standard library only, no API keys, no logins. The checker reads files and changes nothing; it opens no network connection. Tested with 85 unit tests and 754 planted faults on Linux with Python 3.13, 3.12, 3.11, 3.9 and 3.8. All businesses, people and numbers in this kit are made up.

Ask one agent whether to launch the thing and it agrees with you. Ask it to "think like a board" and you get five voices that sound the same, a vote nobody can check, and a verdict that drifts toward whatever you leaned toward when you asked.

The usual failure is not a bad argument. It is a missing one. Nobody asked what a second press day costs. Nobody checked whether the recurring billing terms allow it. The money seat outvoted the compliance seat. The minority view vanished from the summary. The score in the verdict does not match the cards. And the agent reports that it already booked the stall.

This kit gives the room a fixed shape. Five seats, each a business function with its own pillar and its own idea of ruin. Three rounds: independen
````

## Test record

- Date: Nov 1, 2026
- What was run: tests/test_brief.py (15), test_cards.py (15), test_score.py (9), test_verdict.py (18), test_cli.py (8), test_lib.py (9), test_mini.py (5) and test_ratify.py (6, on a real pseudo terminal); tests/mutate.py with 754 planted faults across briefs, cards, rounds, unanimity, verdicts, quick mode, the free script and the rules files; council.py set on the eval set; example/run_example.sh end to end; and the unpack line with test_all.sh. Linux, Python 3.13, 3.12, 3.11, 3.9 and 3.8.
- Result: 85 passed, 0 failed on all five versions with resource warnings as errors; 754 of 754 planted faults caught and all clean baselines clean; the eval set matched 23 of 23; the example scored A 5.3, B 7.3, C 3.8, caught the first-try verdict with 14 findings and refused piped ratification; the unpack line wrote every file and test_all.sh passed there.
- Not run: A live Grok Bot, Claude Code, Codex or Hermes sitting. Whether a given agent argues a seat well. Legal review. macOS or Windows.

Copied from this item's dated test notes. What was not run stays in.

## Worth knowing

- The rules, the skill and the checks were tested, but no live sitting was run inside Grok Bot, Claude Code, Codex or Hermes for this release, so how well a given agent argues a seat is untested. Run council.py set evals with sittings your own agent ran before you rely on it.
- A clean verdict means the room was run by the rules and the numbers match the cards. It does not mean the answer is right, and it is not a forecast of demand or results.
- The Rules seat flags rules it can read and marks the rest unchecked. Not legal, tax or financial advice.
- The council never spends, sends, signs or books anything. ratify.py assumes the agent cannot type at your terminal; run agents as a separate user, since rules alone are not a hard boundary.
- All businesses, people and numbers in the kit are made up. Tested on Linux only; macOS and Windows were not tested.
- On 2026-11-01 on Linux: 85 unit tests and 754 of 754 planted faults passed on Python 3.13, 3.12, 3.11, 3.9 and 3.8; the eval set matched 23 of 23; the unpack line wrote every file and test_all.sh passed in that folder.

## Questions

**How is this different from asking one agent to play a board of advisors?** The seats are business functions, not famous people, and each has its own idea of ruin. A script does the scores, and the verdict is checked line by line against the cards, so it cannot round up, drop a dissent or ignore a compliance stop.

**Does the council make the decision?** No. It recommends. You read the verdict and ratify it at your own terminal, and you carry it out yourself. Nothing is spent, sent or signed by the council.

**Which agents does it work with?** Any agent that can read files and run a script. The kit has a skill, seat files, rules for AGENTS.md and CLAUDE.md, Claude Code and Codex notes, Hermes notes and a Grok Bot standing order. Those were checked by the checker, not run in each agent for this release.

**How is this different from Compounding Offer Audit?** Compounding Offer Audit scores one offer on six tests. Decision Council is for any hard call with two or more options, argued from five sides with doing nothing always on the table.

**Can you set it up for us?** Yes. Our build sets the council up around your business, with seats and weights tuned to your offers and rules and an eval set from calls you have already made, or Managed Ops runs and checks big calls for you each month. Book a call or write to mgmt@paperchasewebb.com.

## How to buy

$39 one time at https://paperchasewebb.ai/bots/marketplace/items/decision-council-argue-a-hard-business-call-before-you-make-it/. Stripe checkout, then an instant download page; the download link is also in your emailed receipt.

Seller: PaperChaseWebb Inc., mgmt@paperchasewebb.com