GutCheck Matrix — Weighted Micro-Decisions

Set the Table

Name what you're torn between, then pick the yardsticks you'll judge them by.

Your options 2–4

Stuck? Quick Chaos fills everything with nonsense you can actually decide between.

What matters 2–5

↑ more is better · ↓ less is better · weight sets how much it counts

Deciding something specific?

The Gut Sprint

One question at a time. Hit the first number you feel — the clock is watching, and how long you stall is part of the read.

Awaiting your first questionable judgement…

6snap calls · about 20 seconds

Don't deliberate. The whole point is to catch what you think before you talk yourself out of it.

GutCheck Matrix · How it works · About & the maths · Source · postage paid by your subconscious

Field Guide

How it works

GutCheck Matrix is for the small-to-medium decisions you keep circling. It asks a few fast questions, times how long you take to answer, and shows you which option sits closest to what you actually want — plus how far you should trust that answer.

The fast path

Name your options. Two to four. Press Enter in a name to add another.
Check what matters. Every decision starts with three criteria that fit anything: Want it, Costs me and Regret. Keep them, rename them, or pick a category — Travel, Buying, Weekend, Game night, Work — from the line under the criteria.
Run the Gut Sprint. One question at a time: rate each option 1–5 on each criterion. Go with your first instinct; the clock is running.
Reveal. Three seconds of drama, then the verdict.
Gut check. If it feels right, accept it. If it stings, say so, and tell it which option you were secretly hoping for.

For a coffee order that is six taps. Nothing else is required — every other control on the page is optional.

The arrows: which way is good

Each criterion has an arrow. ↑ more means a higher rating is better, like Want it. ↓ less means a higher rating is worse, like Costs me, where a 5 means very expensive. Always rate how much of the thing an option has, and let the arrow decide whether that is good or bad.

The app guesses the direction from the name; tap the arrow if it guesses wrong. Some criteria genuinely depend on your mood — Pace in Game night treats fast as good by default, so flip it on a night you want something slow.

Weights: what counts more

Low ×1, Med ×1.5 and High ×2 set how much each criterion counts. Or choose Weigh what matters first before the sprint: the app shows two criteria at a time and asks which matters more, and by how much. It works the weights out from your answers and checks whether they contradict each other. It is optional — skip it and your Low/Med/High settings stand.

Reading the verdict

Close to ideal: How near the winner is to a perfect option — one with the best score on every criterion. 100% means it beats or ties the others on everything.
Holds up: The app re-runs the decision 2,000 times with every rating and weight nudged slightly, and counts how often the same option still wins. Above 75% is solid. Below 60% is close to a coin flip, and the verdict will tell you so.
Inconsistency: Shown only if you ran the weight duel. Under 10% means your priorities hang together. Over 10% means that somewhere you said A beats B, B beats C, and C beats A.

The clock

Every snap call is timed. A stall only counts when it is longer than 2.5 seconds and well beyond your own usual pace — 1.8× your median — so careful, deliberate people are not penalised for being careful. If you stalled longest on an option that did not win, the verdict points it out. That pause is often the real answer.

Everything else

Surrender to Chaos — the same show with the maths inverted: it picks the worst option. Useful for finding out how you would feel about losing.
Quick Chaos — fills in random options and scores, for when you cannot even name the choices.
Adjust the grid by hand — change any rating directly after the sprint.
Receipt of Destiny — a copyable record of the decision, or a downloadable stamp image.
Retro Mode, Sound, Voice of Fate — a dark theme, synthesised sound effects, and a spoken countdown.

Keyboard

Where	Key	Does
Anywhere	`?`	Open this guide
Option name	`Enter`	Add another option
Weight duel	`←` `→`	Pick the side that matters more
Weight duel	`1`–`5`	Equal, A bit more, Clearly more, Far more, No contest
Weight duel	`Esc`	Stop duelling — answers so far are kept, the rest count as equal
Gut Sprint	`1`–`5`	Answer the current question
Gut Sprint	`←`	Back one
Gut Sprint	`Esc`	Stop early; unanswered ratings keep their current value
This guide	`Esc`	Close it

Opening this guide in the middle of a sprint pauses the clock, so reading it never counts as hesitating.

Your data

Everything you enter is kept in this browser’s local storage, on this device. There are no accounts and no analytics, and nothing you type is sent anywhere — the only time it leaves the page is if you press Share on the receipt. Reset Everything wipes it.

Why it exists

Most decision tools make you fill in a spreadsheet, and filling in a spreadsheet is the opposite of listening to your gut. Pure gut feel has the reverse problem: it is fast but unexamined, and it is easy to confuse what I want with what I am anxious about.

GutCheck Matrix sits between the two. It keeps the input fast and instinctive, then uses established decision-analysis maths to show what those instincts add up to — and it treats your reaction to the answer as evidence too. That is why the override exists: if the verdict disappoints you, the disappointment is telling you something the numbers could not.

The matrix is the output, not the input. You never fill in a grid; the grid is what the sprint leaves behind.

How it got here

An earlier version simply added up rating × weight. Because the app asks how much of something an option has, a 5 on Cost counted as five points in favour, so the expensive, exhausting option routinely won. The engine was rebuilt around criterion direction and standard multi-criteria decision analysis methods to fix that properly.

The core algorithm

Ratings + direction→ Weights→ TOPSIS score→ Monte Carlo→ Verdict

1 · Direction: benefit or cost

Every criterion is either a benefit, where more is better, or a cost, where more is worse. It is the simplest step and the most important one: get it wrong and every later step optimises for exactly the wrong thing.

2 · TOPSIS: closeness to the ideal

TOPSIS — the Technique for Order of Preference by Similarity to Ideal Solution (Hwang & Yoon, 1981) — scores each option by how close it is to an imaginary perfect option and how far it is from an imaginary worst one.

1  normalise   r_ij = x_ij / sqrt( sum_i x_ij^2 )     per criterion column
2  weight      v_ij = w_j * r_ij                      weights sum to 1
3  ideal       A+_j = best  v_ij   (max if benefit, min if cost)
               A-_j = worst v_ij   (min if benefit, max if cost)
4  distance    d+_i = sqrt( sum_j (v_ij - A+_j)^2 )
               d-_i = sqrt( sum_j (v_ij - A-_j)^2 )
5  closeness   C_i  = d-_i / ( d+_i + d-_i )          from 0 to 1

C is the Close to ideal number. A weighted sum only rewards what an option piles up; TOPSIS also measures how far it is from the worst case, so one lopsided high score cannot carry an option on its own.

A worked example, using the default criteria with equal weights:

Option	Want it ↑	Costs me ↓	Regret ↓	d⁺	d⁻	Close to ideal
Cappuccino	4	2	2	0.052	0.211	80%
Mocha	5	4	4	0.211	0.052	20%

Mocha wins Want it by one point but loses Costs me and Regret by two each, so Cappuccino sits far nearer the ideal.

Two edge cases are handled explicitly. A criterion that every option scores identically carries no information, so it is left out of the distances. And if nothing differs at all, the app falls back to a plain direction-aware average rather than inventing a 100%.

3 · AHP: weights from pairwise duels

The optional weight duel uses the Analytic Hierarchy Process (Saaty, 1980). Each answer fills one cell of a pairwise comparison matrix on Saaty’s 1–9 scale — Equal, A bit more, Clearly more, Far more and No contest map to 1, 3, 5, 7 and 9 — with the reciprocal in the mirrored cell.

weights       w_i = gm_i / sum_k gm_k,   gm_i = ( prod_j a_ij )^(1/n)
consistency   CI  = ( lambda_max - n ) / ( n - 1 )
              CR  = CI / RI              RI:  n=3 0.58   n=4 0.90   n=5 1.12

The geometric mean of each row closely approximates the matrix’s principal eigenvector, which is Saaty’s own method, without needing a linear-algebra library. A consistency ratio above 0.10 is Saaty’s threshold for judgements that contradict each other; that is the Inconsistency figure. Comparisons you skip count as equal.

4 · Monte Carlo: does it hold up?

Borrowing the idea behind Stochastic Multicriteria Acceptability Analysis (Lahdelma, Hokkanen & Salminen, 1998): after scoring, the app rebuilds the decision 2,000 times. In each run every rating shifts by a random amount of up to ±1 (kept within 1–5) and every weight is scaled by a random factor between 0.75 and 1.25. TOPSIS runs again and the winner is recorded. Holds up is the share of runs in which your winner still wins.

It answers the question a single score cannot: had you been in a slightly different mood, would the answer have changed?

5 · The behavioural layer

Alongside the maths, every snap call is timed. The slowest answer is flagged only when it is over 2.5 seconds and more than 1.8× your own median, so the threshold adapts to how fast you are. It feeds the roast banner, the verdict note and the gut-check list — as a prompt to reflect, not a lie detector.

Honest limitations

Criteria are assumed independent. If two criteria really measure the same thing — say Costs me and Regret both come down to money — that concern is counted twice.
Ratings are coarse. 1–5 is an ordinal scale treated as though the gaps between steps were equal.
Rankings can shift when the option list changes. Normalisation and the ideal points depend on which options are present, so adding or removing one can occasionally swap two others. This is a known property of TOPSIS, called rank reversal.
The robustness jitter is a heuristic. ±1 and ±25% are reasonable defaults, not a measurement of your actual uncertainty.
It will not decide for you. That is what the override is for.

Technical details

One file: A single HTML file of HTML, CSS and vanilla JavaScript. No framework, no dependencies, no build step, no server.
All the maths runs in your browser: TOPSIS reruns on every rating change. The 2,000-run Monte Carlo takes a few milliseconds on a laptop, runs once per reveal and is cached, so it never slows typing.
Sound is synthesised, not recorded: Every tick, chime, fanfare and record scratch is built live with the Web Audio API from oscillators, filtered noise and volume envelopes. There are no audio files.
Voice and images: Voice of Fate uses the browser’s built-in Web Speech API where available. The receipt image is drawn on a canvas and saved as a PNG.
Storage: One localStorage key. Saves from older versions are migrated when they load — criteria without a direction get one inferred from their name.
Design: An old-ticket palette in flat colours with no gradients. The perforated edges are drawn with CSS masks on layers behind the text, so text is never masked, filtered or rotated — which is what keeps it sharp. Bungee for display type, Karla for reading, Courier Prime for the printed-ticket details.
Accessibility: Every text and background pair meets WCAG AA contrast in both themes, everything works from the keyboard, and reduced-motion preferences are respected.
Network: The only outside requests are for the page itself and its web fonts from Google Fonts.
Browsers: Current Chrome, Edge, Firefox and Safari.

Set the Table

The Gut Sprint

—

The verdict

—

Receipt of Destiny