Method — Field

What each part of the page is for.

Anchor page 01

Coverage

Every factor as a column, shaded by how many established players sit at each level. Dense bands mean crowded ground; empty bands mean open water; faint bands mean the evidence is thin, not the market.

Ledger

Inside each established player's card, one row per factor with the anchor and that established player marked on the same track, so the gap between them reads directly, without a legend.

Strategy check

Focus, divergence, and whether the tagline matches the curve it actually draws, read as a target and a measure rather than a paragraph.

ERRC grid

Eliminate, Reduce, and Raise are read from what an established player already ships. Create is labeled a hypothesis, because no evidence can point at something that doesn't exist yet.

Fact strip

Five hard numbers per established player, each carrying its own source and date. A number nobody discloses is written as unknown, with a reason code, never left blank.

Craft read

At most three cited claims about the product itself — an onboarding step, a pricing page, a changelog cadence — never a star rating.

Moves strip

What changed and when, laid on a shared time axis, each move dated and sourced.

Confidence glyph

Four marks, filled or outline: source count, independence, recency, and agreement. Shape carries the meaning, not color.

Window and new

Week, Month, or Year filters moves and new established players to that span. Anything first seen inside it gets a dot and its date, not a fresh curve.

Weekly, headless, buyer-side.

Brain 02

Six Paths discovery

Established players don't come from one competitor list; each anchor gets one discovery pass per path in Kim and Mauborgne's Six Paths framework — alternative industries, strategic groups, buyer groups, complementary offerings, the functional-versus-emotional appeal a buyer is actually paying for, and the trends its category rides.

Every factor on the canvas is defined from the buyer's side — what a buyer can observe or pay for, never a vendor's own claim about itself.

Three tiers, one page, in order.

Tiers 03

Established players

Who the buyer already weighs the anchor against. The bar is medium: one full-weight independent source plus the company's own site, with at least one page read this run. High marks two such sources. Full record — facts, moves, reads, canvas levels, an ERRC grid.

Long tail

Which slice of the job is being served to buyers the anchor doesn't reach — Explee's shape, not Apollo's. Thin bar: the candidate's own site plus at least one weighted point of independent evidence, or one full-weight source alone. A light record: one distinguishing fact, the job it's hired for, its fit against the agentic, early-stage, and no-SDR axes, pricing and funding when fetched, and a status. Below the bar, a claim carries the inline phrase Thin evidence, 2 sources: YC directory, integrations page, Wikipedia's citation-needed pattern for a record that hasn't cleared the established player bar yet.

New entrants

What's new this week, read straight off a feed before it has earned either bar. Low by construction, on the strength of one item, until a later run clears it into the long tail.

Sources

The long tail is read from the YC directory, a candidate's own integrations or partner page, marketplace and app-store listings, and trade press. G2, Crunchbase, and TechCrunch articles are named sources Field does not fetch — G2 and Crunchbase block the request at the runtime layer no matter what their robots.txt allows, and TechCrunch disallows Claude's crawlers outright. A citation from any of the three reaches the page only if the maker supplies it by hand.

Confidence is a checklist, not a guess.

Confidence 04

The bars

High is two independent, non-affiliated, full-weight sources agreeing on the same buyer job, one of them read this run. Medium is one. Thin is the company's own site plus weaker sources that add up to one. The anchor's own pages and the company's own pages never count as independent, however many of them get read.

Verification

A second pass checks that each citation actually entails the claim it supports. If it doesn't, the claim becomes unknown rather than staying on the page.

Unknown

A missing fact is written as unknown, with a reason code, not left blank and not guessed at.

New companies, read before they're established players.

Entrants 05

Feeds

Three feeds get read first, every week: YC Launches, Crunchbase News, and Product Hunt. Six more follow: Launch HN, TechCrunch's startups feed, Newcomer, Latent Space, BetaList, a16z's portfolio, and The Information's headlines.

A handful of those outlets name Claude's crawler in their robots.txt and restrict or block it while leaving a generic agent open. Field reads those outlets as feed metadata only and never fetches the linked article, rather than treating an unnamed gap as permission.

An item counts as new when it clears at least one test, cited: founded or launched within eighteen months, this is its first public launch, or this is its first funding round.

A cheap keyword pass against the anchor's jobs to be done and canvas factors drops most items for free; a model call then judges what survives against that same frame, so an entrant has to actually compete, not just share a category.

A new entrant starts at low confidence, by construction, on the strength of one feed item, and waits on the review queue rather than the public page. It graduates to an established player, and onto the canvas, once two independent sources agree or the maker approves it directly.

What stays manual, and private.

Manual 06

The queue

A source robots.txt disallows is dropped, not routed around; it returns only if the maker supplies a citation by hand. Every medium-confidence established player and every new entrant waits in the admin queue for the maker's own approve or reject, a single keystroke, not a majority vote. What arrives there stays private until that moment, and nothing below the confidence bar reaches the public page.

How the map gets made.