Documentation

What is Vesta?

Vesta is a decision-support tool that helps you compare options against multiple criteria — price, quality, risk, strategic fit, and anything else that matters to you — and arrive at a ranked, defensible recommendation.

Instead of relying on gut feeling or a spreadsheet full of weighted averages, Vesta guides you through a structured process:

Define what matters (criteria)
Define what you're choosing between (alternatives)
Express how much each criterion matters (by ordering your criteria, or by comparing them in pairs)
Resolve trade-offs head-to-head (duels)
Get a transparent, auditable ranking

AI analysis is woven throughout to flag inconsistencies, explain results, and suggest improvements — but you remain in control at every step.

Quick start

1 — Create a project

From your project list, click New project. You have four options:

Blank: give it a name and build from scratch
Template: start from a ready-made model for common use cases (vendor evaluation, hiring, tool selection…)
Import: load a JSON file exported from a previous project
AI Model Builder: describe your decision in plain language and the AI structures it with you in an open-ended conversation, then guides you through weighting it — requires your own Anthropic or Groq API key, and runs directly from your browser to that provider

2 — Add criteria

Open the Criteria tab. Add each factor that matters in your decision. For each criterion, pick the right type (numeric, rating scale, label-based, or yes/no) and optionally set a veto threshold to automatically disqualify options that cross a hard limit.

3 — Add alternatives

Open the Alternatives tab. Add each option you're considering. Fill in the value for each criterion. Vesta handles the rest.

4 — Set weights

In the Weights stage, say which criteria matter more than others. In simplified mode you drag them into order of importance; in pairwise mode you compare them two at a time. Either way you never have to juggle all of them at once, and you can switch between the two whenever you like.

5 — Resolve trade-offs

The Trade-offs stage shows head-to-head comparisons between alternatives on individual criteria. Confirm automatic results or override them when you have contextual knowledge the numbers don't capture.

6 — Read the ranking

The Ranking tab shows your final ordered list with per-criterion score breakdowns and optional AI insights explaining why each alternative landed where it did.

Criteria

Criteria are the factors you use to judge each alternative. Vesta supports four types, so you can model almost any real-world attribute.

Quantitative

A numeric value with a unit (price in $, weight in kg, response time in ms). You choose whether higher is better or lower is better. Two alternatives are considered tied when their values are within 3% of the full range — this prevents spurious precision from skewing results.

Ordinal

A rating on a fixed numeric scale (e.g. 1–5 stars, 1–10). Displayed as a slider or star control. Higher is always better. Alternatives with the same score are tied.

Mapped

A set of named labels that map to numeric scores — for example Small → 2, Medium → 5, Large → 10. Useful for categorical attributes like company size, support tier, or contract type. Each label can optionally be marked as a veto: any alternative with that value is automatically disqualified.

Boolean

A yes/no criterion. You define which answer is preferred and, if needed, which answer triggers a veto. Useful for compliance checkboxes, feature presence, or legal requirements.

Veto thresholds

A veto is a hard disqualification rule. If an alternative violates a veto condition (e.g. price > budget, or a required feature is absent), its total score is forced to zero and it appears at the bottom of the ranking labelled "vetoed". Its potential score (what it would have scored without the veto) remains visible for reference.

Alternatives

Alternatives are the options you're deciding between — vendors, job candidates, products, strategies, or anything else.

Entering values

For each alternative, enter the value for every criterion. The input control matches the criterion type: a number field for quantitative, a slider or stars for ordinal, a dropdown for mapped, and a toggle for boolean.

Additional properties

Attach non-scored metadata to any alternative — a website link, a location, or a free-text note. Properties are shown in the ranking view for easy reference but do not affect scoring.

The starred alternative

Star one alternative to mark it as your preferred fallback (the option you would choose if you had to decide right now). This activates the Vesta adjustment: the weight system subtly shifts to emphasize the criteria where the starred alternative is strong, giving it a fair chance against rivals that may outperform it on criteria you actually care less about. The starred alternative also stays visible in the ranking even if it violates a veto.

Weights & comparison

Vesta uses the Analytic Hierarchy Process (AHP) to derive criterion weights. Rather than asking you to assign a percentage to each criterion, which is cognitively hard and prone to error, it asks you to compare them. There are two ways to do that, and you choose which when you create a project or at any point afterwards.

Simplified mode

Drag your criteria into order of importance, most important first. Vesta turns that order into weights, giving the top of the list meaningfully more weight than an even split would. This is the default for new projects, and it is the quicker of the two: ten criteria are one pass rather than 45 separate comparisons. An ordering also cannot contradict itself, so there is nothing to reconcile afterwards.

Pairwise mode

Compare criteria two at a time, on a scale from equal importance up to extreme importance in either direction: "Is price more important than delivery speed, and if so, how much more?" It takes longer, and it is the only mode that lets you say how much more rather than just which comes first, so it is the one to use when the difference between first and second place matters.

Switching between them

Switching is free and loses nothing. Both modes read and write the same underlying comparisons, so moving from one to the other changes what you are looking at rather than what your project says: your weights, scores and ranking are identical either side of the switch. A hand-tuned matrix shown in simplified mode appears as the order those weights are already in, and comes back untouched if you switch away without changing anything. Only the weighting itself changes when you actually reorder or move a slider.

When answers conflict

In pairwise mode it is possible to give answers that cannot all hold at once: price beats speed, speed beats support, support beats price. Vesta points out which three judgements are involved, but it does not treat it as an error, because it is not one. Preferences that go in a circle are a real thing to hold, and they are often what makes a decision hard rather than arithmetic. Your weights are computed regardless, as the closest single fit to everything you said. Revisit a judgement if one of them was a slip, or leave the tension in place if it is real.

If you have categories, the whole process runs twice: once to weight the categories against each other, and once within each category. The final weight of any criterion is its within-category weight multiplied by its category's weight. Simplified mode covers both levels, so it is never a narrower model than pairwise.

Trade-offs

When two alternatives score the same on a criterion, the numbers alone can't separate them. That's the only situation where your subjective judgment is actually needed — if one alternative already outperforms the other, you said so when you entered the values.

Pick any two alternatives and Trade-offs surfaces only those ties, ready for you to break them.

Breaking a tie

For each tied criterion, drag the nudge slider toward the alternative you prefer. The further you drag, the stronger the preference you express. This creates a small score offset applied to the final evaluation — just enough to break the tie in your favour without distorting the overall ranking.

Reviewing decided criteria

Use the "Show decided criteria" toggle to see how the two alternatives compare on criteria where there is no tie. These rows are read-only — the math already has a clear answer and no input from you is needed or possible.

Ranking

The Ranking tab shows your final ordered list. Every score is traceable: expand any alternative to see exactly how much each criterion contributed to its total.

How scores are calculated

Each criterion value is normalised to a 0–1 scale across all alternatives (0 = worst in the set, 1 = best). Normalised scores are multiplied by their criterion weight and summed. If duels introduced nudges, small offsets are added. The result is a total score between 0 and 1.

Category breakdown

When you use categories, the ranking view also shows a score contribution per category, so you can see whether an alternative wins on cost but loses on quality, for example.

Veto warnings

Vetoed alternatives display their potential score alongside their actual score of zero, so you can see what they would have achieved had they cleared the disqualifying condition.

AI features

AI analysis is optional and complementary — it never changes your data, it explains and challenges it. All insights are marked stale automatically when relevant data changes, so you always know whether they reflect the current state of your project.

Ranking insights

Available from the Ranking tab. Generates three focused cards:

The winner — why did the top-ranked alternative win?
The kingmaker — which single criterion had the most decisive impact on the final order?
Stress test — what would change if a key criterion weight shifted?

Trade-off guidance

Available from the Trade-offs stage. The AI reads your head-to-head results and flags potential inconsistencies — for example, if you ranked A above B in three criteria but B above A in a fourth that partially overlaps. It helps you spot logical contradictions before they silently distort your ranking.

Project audit

Available from the Audit tab. A full health check of your project:

Dissonance patterns — preferences that contradict each other across duels
Weight suggestions — adjustments that would better align weights with your revealed duel preferences
Omitted variables — criteria you may have forgotten given the nature of your decision
Confidence score — an estimate of how internally consistent the project is

Key concepts

AHP — Analytic Hierarchy Process: A method for deriving weights from comparisons rather than from direct percentages, which people are notoriously bad at estimating. Vesta offers two ways to supply those comparisons: simplified mode and pairwise mode.
Simplified mode: Weighting by dragging your criteria into order of importance instead of comparing every pair. Faster — ten criteria are one pass rather than 45 comparisons — and an ordering cannot contradict itself. It is the default for new projects, covers categories as well as criteria, and can be switched on or off at any time without losing anything.
Pairwise mode: Weighting by comparing criteria two at a time and saying how much more one matters than the other. Slower than simplified mode, and the only mode that captures the size of a gap rather than just the order.
Normalisation: Converting raw criterion values to a 0–1 scale so that a price in dollars and a rating out of 5 can be compared on equal footing. The worst value in the set becomes 0 and the best becomes 1.
Veto: A hard disqualification rule. An alternative that crosses the threshold is scored 0 regardless of how well it performs on other criteria.
Head-to-head: A comparison of two alternatives, one criterion at a time. The Trade-offs stage focuses on tied criteria — cases where the computed scores are equal and you need to express a subjective preference to break the deadlock.
Nudge: A small synthetic score offset applied when a duel result is tied. The nudge preserves the tie at the raw-value level while letting you express a slight subjective preference.
Starred alternative: An alternative you mark as your preferred fallback. The Vesta weight adjustment shifts emphasis toward criteria where this alternative is strong, and it remains visible in the ranking even if vetoed.
Hierarchical AHP: A two-level comparison used when criteria are grouped into categories. Category weights are determined first, then criteria are compared within each category. The final weight of a criterion is category weight × within-category weight.
Stale insights: AI-generated analysis that was produced for an earlier version of your project data. Vesta marks insights stale automatically when relevant data changes so you always know if they are up to date.

Working together

A decision is rarely one person's work. Invite people to a project with Share, and give each of them a role describing what they are there to do.

Roles

Role	Can	Cannot
Viewer	Read the decision	Change anything
Evaluator	Add alternatives and fill in their evidence	Change criteria or weights; see the weights or the ranking
Modeller	Define criteria and categories, and weight them	Score alternatives; see the alternatives or the ranking
Adjudicator	Resolve trade-offs, pin the pick, read the outcome	Add alternatives or criteria; set weights
Facilitator	Manage who is involved	Edit the model itself

Each role is also blinded from whatever would bias its own contribution: an evaluator scores alternatives without seeing the weights or the standings, and a modeller builds and weights the frame without seeing the alternatives it will judge. Those stages still appear in the sidebar — the process is the same for everyone — but their contents are withheld with a note saying why. This is about bias, not confidentiality: nothing is encrypted, and a project export still contains the whole model.

Someone can hold more than one role — an evaluator who also adjudicates, say. The person who created the project owns it and can do everything, including changing privacy and deleting it. An owner never holds a role as well: ownership already covers everything, and a second entry saying otherwise would only contradict it.

Who can do what to the project itself

Roles cover the decision. These cover the project as an object, and they live on the project's Settings page — reached from the gear at the foot of the stage sidebar, or from the ⋮ menu next to the project name. Each section appears only if you hold the operation it needs, so what you see there depends on your role.

Action	Who	Why
Export	Anyone who can open it	Scores are worked out in your browser, so opening a project already gives you all of it. Restricting the download would stop nobody.
Manage people	Owner, facilitator	The facilitator role exists for this. They cannot change their own roles, or it would promote itself.
Undo and redo	Anyone who can edit	Offered only where there is something you could undo, so a viewer or a facilitator is not shown a button that would do nothing.
Import data	Owner only	It replaces the whole model at once — criteria, options, weights — which is broader than any single role is trusted with.
Rename	Owner only	The name is how everyone else finds the project, including in the invitation they were sent.
Public or private	Owner only	Making a project public lets anyone with the link read it, signed in or not.
Delete	Owner only	Nothing to add.
Leave	Anyone but the owner	An owner leaving would strand the project, so transfer it or delete it instead.

An owner can hand the project to a facilitator who has opened it at least once, from the Share dialog. They become the owner outright and you keep access only if they invite you back, so it is not something you can undo yourself. A facilitator cannot change their own roles — they can give and take away everyone else's, so without that the role would promote itself.

Inviting

You invite by email address, and there is nothing to accept. The project appears under “Shared with me” the next time that person signs in with that address. If they have never used Vesta, the invitation simply waits for them.

The flip side is that a mistyped address never matches anyone, and looks exactly like someone who has not opened the project yet. If a colleague says they cannot see it, check Settings → Access: it shows “Has not opened it yet” against anyone who has never arrived.

What people can see

Roles control what someone can change, not what they can read. Anyone you invite can see the whole project, and Vesta hides the Ranking and Visuals tabs from evaluators only so that entering data is not coloured by the standings so far. Treat that as a focus aid, not confidentiality: do not invite someone you would not show the results to.

Leaving

You can leave any project you do not own, from Settings → Danger zone or from the projects list. Nothing is deleted — the owner keeps the project and can invite you again. Projects shared with you never count towards your own project limit.

History

Every change to a project is recorded: what changed, who changed it, and what the value was before. The owner reads it at Settings → History.

What an entry says

Each entry names the action rather than the fields it touched — “Set the weights”, not “changed categoryComparisons” — then lists each value that moved, from what to what. A score reads Option Suburban Family · Price · value 560000000 → 560000; a weighting reads Practical moderately more important.

Some changes pull others along with them. Deleting a criterion also removes every score and comparison that named it, and those follow-on changes are listed separately, marked knock-on, so a small act with a large consequence looks like one.

Why only the owner

The log carries before-and-after values for every field, so anyone reading it can see the parts of the model their role withholds. Giving it to an evaluator would undo the reason the Ranking tab is hidden from them; giving it to an adjudicator would hand the person judging the result the record of how it was built. It is the owner's, and no role grants it.

Recorded by the server

The record is written where the change is applied, not by the browser that asked for it — so it cannot be skipped, edited or reordered from a client, and the time and the person are the server's account of them rather than yours. Undo is recorded too: undoing a change does not unmake the fact that it happened, so both appear.

The history outlives the project. Deleting a project removes the project; the record of what was done to it is kept separately.

Plans & limits

Vesta is free to use. A paid plan is planned for the future and will raise or remove the limits below. Projects shared with you do not count towards your own project limit — only the ones you create.

Projects

Limit	Free	Paid (coming soon)
Projects per account	3	Unlimited

Per-project limits

Limit	All plans
Categories	10
Criteria	20
Alternatives	20
Property keys per alternative	10
Name length	4–30 characters
Description length	160 characters

AI features

Every AI feature — the AI Model Builder, pairwise comparison guidance, duel insights, and audit insights — requires your own API key from either Anthropic or Groq, and calls that provider directly from your browser. Your key and your conversation never reach Vesta's servers. None of them are subject to a Vesta-imposed limit; usage is billed to your own account with whichever provider you connect. Connect both and Vesta falls back to the second automatically if the first is rate-limited or unavailable.