Decisions with Claude: make criteria and uncertainty visible

Compare options with Decision Toolkit →

You have to choose between two vendors, two possible hires, or a price increase that could drive away part of your client base. You already have, in your head, the list of pros and cons for each option. It is not enough: it puts a verified figure, a rumor overheard in a meeting, and a personal preference on the same line, never distinguishing between them, and it says nothing about the cost of waiting one more month before deciding.

Comparing options with Claude is not about asking "which one should I pick?" and getting a ready-made answer. The Decision-Toolkit skill structures the comparison — criteria, assumptions, scenarios — so the decision stays yours, made with every element on the table rather than only the ones that came to mind spontaneously.

This article details that nine-angle method, what you actually get depending on whether you work in Claude Code or a plain conversation, and where it deliberately stops short of the decision itself.

Summary

1. Why a list of pros and cons is not enough
2. The nine angles that structure a comparison with Claude
3. Interactive HTML or text framework: what you get depending on your tool
4. Building your comparison step by step
5. What Decision-Toolkit does not do
6. Training to structure decisions with AI
7. Frequently asked questions

Why a list of pros and cons is not enough

A list of pros and cons treats a verified figure, a rumor overheard in a meeting, and a personal preference exactly the same way: one line, one plus or minus sign. That flat ranking explains why two reasonable people, given the same information, sometimes reach opposite conclusions — they simply weighed the same lines differently without ever noticing it during the discussion.

The bias that costs the most: deciding before framing the question

The most common temptation when facing an uncomfortable decision is to decide fast just to stop thinking about it. The problem is not speed itself: it is that a decision made before defining what actually matters gets justified afterward, by hunting for arguments that confirm a choice already made rather than looking for them beforehand. The skill reverses that order: it first asks what must be decided now, what can wait, and what a choice truly commits you to, before comparing anything at all.

Separating facts, preferences and assumptions

A listed price is a fact. "This supplier seems more reliable" is a preference — legitimate, but unproven. "The market will probably tighten next year" is an assumption, which could turn out to be wrong. The pack asks you to separate these three categories before scoring anything, precisely because a table that blends them gives a false impression of objectivity: a score out of ten given to an assumption looks just as solid as one given to a verified price, when nothing makes them comparable without this explicit distinction.

The nine angles that structure a comparison with Claude

Running a decision through nine complementary angles — reversibility, risk asymmetry, cost of waiting, weak signals and five others — serves one precise purpose: stopping a single angle, usually price or immediate comfort, from dominating the whole comparison without you consciously choosing that. It is not a checklist to tick mechanically; it is a filter that surfaces the angles you would have missed alone, without imposing an order of importance.

Reversibility and risk asymmetry

A reversible decision — trialing a new tool for a month — does not deserve the same caution as a commitment that is hard to undo, such as a permanent hire or a yearly contract with an exit penalty. Risk asymmetry completes this angle: if option A loses little on failure and gains a lot on success, while option B loses a lot and gains little, the two deserve different treatment even when their stated probability of success looks identical on paper.

Cost of waiting and weak signals

Waiting has a cost, even when it never shows up on an invoice: a competitor moving forward, an opportunity closing, a team stuck in uncertainty. Weak signals — an isolated remark from a client, a figure that shifts slightly before the others do — prove nothing on their own, but the pack asks you to name them rather than dismiss them out of excessive caution, as long as they are never treated as proof while they remain isolated.

Interactive HTML or text framework: what you get depending on your tool

Decision-Toolkit's output format depends directly on where you use it. The pack defaults to an interactive HTML guide — a file you open and click through from one scenario to another — but that format assumes an environment that writes and opens files. A Claude, ChatGPT or Gemini conversation without that capability cannot produce that file as-is: the pack then explicitly falls back to a structured text framework, not as an improvised workaround.

Why the format changes depending on the environment

This is not an arbitrary limitation of the skill: Anthropic's official documentation on Agent Skills states that network access and file execution vary by surface — full in Claude Code, variable on claude.ai depending on account settings, absent by default via the API without a code execution tool. A skill that produces an HTML file in Claude Code can therefore fall back, in a chat without that capability, to a text version that is equivalent in substance but different in form. To place these differences within the wider Claude offering, our complete guide to Claude's products covers the relevant plans and features. Confusing the two formats — announcing an interactive file that does not exist — is exactly the kind of unkept promise to avoid.

What to ask for if you are working in a plain chat

In an ordinary conversation, ask directly for the text framework: the list of nine angles applied to your decision, a comparison table and the scenarios, without demanding a file the environment cannot produce. If Claude nonetheless offers to "generate an HTML file," verify that a real, openable file is actually created before assuming you have the interactive version — otherwise, it is most likely a description of what the file would contain, not the file itself.

Building your comparison step by step

Five steps, in this order, avoid having to rebuild the comparison later because an important criterion was overlooked along the way or an assumption was treated as settled fact from the start.

Step 1: Define what must be decided now. Some decisions do not need settling this week; confusing them with urgent ones dilutes attention on both. Start by writing, in one sentence, the exact question you must answer and the real deadline constraining it — a deadline imposed by a third party, not one you set for yourself out of impatience.

Step 2: Gather real options and verified constraints. List the options that are actually available, not theoretical variants nobody will ever offer. Attach the constraints already known — budget, timeline, contractual obligations — so the pack does not start from a blank slate.

Step 3: Choose understandable criteria and make their weight explicit. A criterion you cannot explain to a colleague in one sentence will not be useful in the final table. Assign a weight to each criterion and state why, rather than letting an implicit weight hide behind a plain score.

Step 4: Ask for scenarios and a sensitivity test. An option that only wins because an uncertain figure was set at its minimum deserves testing with that figure at its maximum too. Explicitly ask what changes if an assumption turns out wrong.

Step 5: Decide as a human and record the reason retained. The table prepares the decision; it does not make it. Once decided, log the decision with its main reason: this record serves as a reference if the decision is questioned later.

What Decision-Toolkit does not do

The pack never decides for you, and that is a deliberate constraint rather than a regrettable limitation: a high-stakes decision carries a responsibility no tool can bear. It also does not replace legal or accounting advice when a decision has contractual or tax consequences, and it does not turn a score out of ten, assigned without any underlying measurement, into numeric proof: the score remains a judgement, whatever formatting the table gives it.

A well-built comparison can also reveal that none of the listed options is the right one, and that the real decision is to look for a third one or postpone the choice by a month to gather missing information. The pack leaves that possibility open rather than forcing a ranking between two mediocre options.

In our exchanges with small-business leaders and independents about their hard decisions, the blocker is almost never a lack of options: it is the lack of shared criteria to compare them, especially when several people must decide together. Once criteria and their weight are made explicit, the discussion changes nature — it becomes about the weight of a criterion, not about who is right in the abstract.

Training to structure decisions with AI

Using Decision-Toolkit effectively requires knowing how to turn a vague decision into concrete criteria, and recognizing when a text framework is enough versus an interactive guide. That is exactly what Educasium's Master Claude training works on, devoting several modules to using Claude Code and skills for profiles who had never opened a terminal before. For a business owner or a self-employed professional under the relevant scheme, the FIFPL covers part of the cost of a Qualiopi-certified training under the 2026 criteria set at €300/day and €900/year, with e-learning capped at 50% of the daily rate — funding that must be requested before training starts, not after.

Frequently asked questions

Can Decision-Toolkit tell me which option to choose?

No, and that is a deliberate design choice by the pack. It compares, crosses the angles and proposes scenarios, but it never decides for you or recommends a default option. The decision stays entirely your responsibility, with every criterion laid out on the table rather than only the ones that would have come to mind spontaneously without this structuring.

What happens if I use this skill in ChatGPT or Gemini instead of Claude?

The nine-angle framework remains applicable in any conversation, but the experience changes: the interactive version the pack expects is designed for an environment that runs code and writes files, which is not guaranteed outside Claude. In another tool, ask directly for the structured text framework and supply the nine angles yourself if the assistant does not already know them, rather than hoping for an interactive file that will not be generated. Also state your criteria and their weight explicitly before asking for the comparison, rather than assuming the assistant will independently apply the same nine angles as a dedicated pack.

Do I need Claude Code to get the interactive version?

In practice, yes: that is the environment the interactive HTML guide is designed for, with filesystem access to write it and open it in a browser. On claude.ai, file generation depends on the settings enabled on the account; without that capability, or via the API without a code execution tool, you get the text framework, which contains the same angles and scenarios in a non-clickable form. Always check that a genuinely openable file was created before assuming you have the interactive version, rather than trusting a description of what that file would contain.

How does the skill avoid turning a preference into a verified fact?

By requiring every row of the table to state its nature before it is scored: verified fact, stated preference, or assumption to be tested. A score given to an assumption stays visually identifiable as such, rather than blending into a single column that would suggest every row carries the same level of certainty. That discipline, more than the number of angles used, is what separates a useful comparison from a table that merely dresses up a hunch with numbers.

A useful comparison never replaces your judgement: it feeds it with explicit criteria, tested scenarios and a clear distinction between what is verified and what remains an assumption. That is exactly what the Decision-Toolkit comparison generator automates, in whichever version matches the environment where you use it.

To frame a decision that first requires documenting an existing process, our method for framing a process before automating it complements this approach upstream of any comparison.

Training 100% fundable via OPCO/FIFPL. Qualiopi-certified programme. To structure your decisions with AI as part of our Master Claude training, contact Educasium and specify your status (employee, self-employed, business owner) and your goal.