RANK.md

# ABOUT RANK.MD

RANK.md is a working idea: a plain markdown file for AI agents that need to rank things and show their work. It has two parts. The map lists what is worth ranking, as boards: a domain, a category, the question being asked, and the criteria with their weights. The math ranks the items an agent brings: scores per criterion, weights, how sure each score is, when two items are too close to call, and how to merge or compare rankings. Every score points to a source link with the date it was read. It holds rankings and links, never copies of anyone's data.

## WHAT IT COULD COVER

• - the board: domain, category and the question, such as laptops for field engineers
• - criteria: what counts, how it is measured, and its weight
• - sources: a link and a read date for every score
• - ties: the gap under which two items count as tied
• - uncertainty: the chance each item is truly first, not just the order
• - merging: one list from several, and the method used
• - freshness: when the ranking goes stale and should be checked again
• - disclosure: any stake the author has in an item on the board

## HOW IT COULD WORK

• - the map comes first: an agent finds the board for its question, or proposes one with criteria
• - scores come from pages the agent can cite, each with its link and date
• - the math runs in the open: scale each criterion, weight, add, then see how far the order moves when a weight changes
• - close calls: when the gap is under the tie rule, the file says tied and names what would break the tie
• - on its recheck date a ranking is marked stale, not quietly kept

## METHODS IT COULD NAME

Real ranking math, old and well studied:

• - weighted sum: scale each criterion from 0 to 1, multiply by its weight, add; simple and easy to check
• - Borda count: merges ranked lists by giving points for each position; named for the French mathematician Jean-Charles de Borda
• - Kendall's tau: how far two rankings agree, from the share of pairs they put in the same order
• - Elo rating: ratings updated after each head-to-head result; devised by Arpad Elo for chess
• - Bradley-Terry model: the chance one item beats another, fitted from many pairwise comparisons
• - Wilson score interval: ranks by the low end of the likely share of positive ratings, so three lucky ratings do not top three thousand
• - analytic hierarchy process: criteria weights from pairwise judgments of importance, developed by Thomas Saaty
• - reciprocal rank fusion: merges result lists by adding one over a constant plus each item's rank, used in search

## WHO IT COULD HELP

• - buyers at small companies: vendors compared on stated criteria, with every score linked to its source
• - researchers: papers, datasets or tools ranked for a review, with the method written down
• - agent builders: models or tools ranked for one task, recomputed when new results arrive
• - teams with many reviewers: each reviewer's list merged into one, with the disagreements shown

An agent asked for the best option produces a confident list either way. The difference between a ranking and an opinion is whether someone else can recompute it from the criteria, the weights, the sources and the date. Near the top, small weight changes often reorder items; a ranking that shows this beats one that hides it.

## A SKETCH

```md
---
rank: draft
---
# RANK.md
## board: laptops for field engineers (made up)
- criteria: battery 0.40, weight 0.30, repair 0.30
- scale: 0 to 1 per criterion, 1 is best, from linked pages
- laptop A: 0.9, 0.6, 0.8 = 0.78
- laptop B: 0.7, 0.9, 0.6 = 0.73
- ties: a gap under 0.03 is a tie; 0.03 or more sets the order
- order: laptop A, then laptop B (gap 0.05)
- chance_first: A 0.70, B 0.30
- sources: each maker's spec page, read 2026-11-02
- left_out: laptop C, no public repair manual
- stake: none in any maker on the board
- recheck: 2027-02-01
```

## DISCLAIMER

RANK.md is a working idea. It publishes no rankings of its own, endorses no product, company or person, and sells no data. A ranking is an opinion that follows from the criteria and weights someone chose; change them and the order can change. The file states its method; it does not check that a source is accurate or current. Link to sources rather than copying them, and follow each site's terms. Rankings of people for credit, housing, jobs or insurance carry legal duties a file does not meet. The items, scores and dates in the sketch are made up.

RANK.md is an informational page. It is provided as is, without warranty of any kind, and may be incomplete, inaccurate or out of date. Nothing here is financial, investment, legal, tax, insurance, medical, engineering, safety or other professional advice, or an offer or solicitation of any kind. The sketch is an illustration, not a standard or a recommendation. Names of third-party products or companies, if any, are used only to describe them; no affiliation or endorsement is implied. Use at your own risk. This page sets no cookies and runs no scripts or analytics; like any website, its host keeps standard request logs.