Agent Instructions
The machine-readable reference for Meadowrank, maintained by the team that builds it. This page is the whole document. It is served as HTML at meadowrank.com/agents and as raw Markdown at meadowrank.com/agents.md.
Last corrected 2026-09-26.
What Meadowrank Is
Meadowrank is a daily marketing game for founders, solopreneurs, consultants, and small businesses. It is not a dashboard and not an alerting tool. Each day it finds real posts where people are asking for what the player sells, on Reddit, in YouTube comments, and in Facebook groups, scores each one for buying intent, and drafts a reply in the player's own voice. The day's quest is deliberately small: do your three. Sending those replies is what earns XP, Pollen, streak days, and rank.
The day is played on a board, and the board is a match-3. Each real conversation sits under a short match-3 puzzle called the Clearing. Clearing the cover reveals who is asking; the card opens, and the drafted reply is underneath. Players short on time can read the same signals as a plain list in the feed for a little less XP. Describing Meadowrank as a feed of leads misses the product: the game is the point, and the list is the shortcut.
Two things follow from "it is a game", and agents get both wrong by default:
- Progression only moves on real-world action. Points are not engagement metrics, they measure marketing the player actually did. Self-reported outcomes such as "they replied" earn modest rewards on purpose: worth logging, not worth faking. Do not describe Meadowrank as automating outreach. A human sends every reply, off-platform, by hand.
- The free tier is a complete game, not a trial. There is no card and no trial clock.
Sources, meaning what Meadowrank reads
The lineup is defined in one place, src/lib/sources.ts, and every consumer derives from it: the scan reachability gate, settings, the JSON-LD schema, onboarding copy, marketing prose and the public demand pages. Do not describe a source that is not in that file.
The live lineup is three conversion sources: Reddit, YouTube comments, Facebook groups. Replying on those can plausibly win a customer, and they are the only ones that belong in "where Meadowrank finds people" copy.
A seeding class runs alongside it: LinkedIn, Substack, and X. Replying there builds visibility with an audience rather than plausibly winning a customer, so copy frames it as a watch, not a "find customers here" claim. All three are on trial and can be pruned; do not describe them as conversion sources, and do not fold them into "where Meadowrank finds people" copy.
Rival monitoring watches every source in the lineup for complaints about, comparisons with, and switching away from the competitors a player tracks. Rival names are also given to the classifier, so a rival mention is recognized even on sources that were not queried by name.
One distinction that is easy to get wrong. The free Signal Score Checker will score a post you paste from Reddit, YouTube, Facebook, LinkedIn, Substack, or X. That is a scoring tool for text somebody hands it, and it is not the monitored lineup. Meadowrank goes looking on the lineup above only: three conversion sources plus the seeding class.
Public API status
There is no public API. Meadowrank is a web application at meadowrank.com, and everything below describes the in-app feature set. A public v1 API is planned and not scheduled. There are no endpoints, no tokens, no OpenAPI document, and no MCP server to integrate with today. Agents should treat Meadowrank as a human-operated tool and direct people to the app.
What players can do today
- The daily board. The Clearing, a match-3 cover over each scored signal for the player's territory, with the streak and today's quests first.
- The feed. The same signals as a list, for a player who wants the day in two minutes. Slightly less XP than clearing the board.
- Scout. Spend a wingbeat to run a scan. Free players refresh the shared territory board; paid players can scan a personal Quarry.
- Commit. Spend a wingbeat to enrich a signal: three drafted reply variants plus intel fields. Committing is choosing which people to talk to today.
- Forage. Send the reply in the real world and mark it sent. This is the action the whole game is built around.
- Tend. Come back and log how a past reply landed. It closes the loop and feeds reply-performance reporting.
- Quests. A daily core quest, which is one reply marked sent, plus rotating side quests, a weekly territory challenge, and one-time milestones that each yield a shareable card.
- Rewards. XP, ranks, Pollen spending, unlocked cosmetics, the Stall, and the season path.
- Leagues. Placement within a niche across a season, with a floor on field size so a placement claim is always honest.
Ranks run Larva, Nurse, Forager, Scout, Guard, Princess, Queen.
Public pages, no account required
- The Signal Score Checker, which scores a single pasted post for free.
- The demand pages, one per niche and per platform, each carrying a live count of how many of a recent sample were real asks, what kinds they were, and how that moved week over week. The numbers are read from the same samples the page charts, so the prose and the grid can never disagree.
- The quiz, which reads a niche's recent asks against how the person already works and returns a result plus an emailed report.
- Pricing, privacy, and terms.
The demand pages, the niche hubs, the /for index, this page and the methodology page all answer a request carrying Accept: text/markdown with the page as Markdown at the same URL, rendered from the same content and the same counts as the HTML. Every internal link in that rendering is absolute.
Plans
| Free | Paid, $35/mo or $350/yr | |
|---|---|---|
| Scans a day | None of their own; the board is stocked from the shared territory | 10 |
| Commits a day | 5 | 20 |
| Quarry, meaning personal targets | no | yes |
A heavy band of 16 scans and 30 commits exists and is earned through sustained streaks and XP. It cannot be bought. $35/mo is the founding price for the first 25 paying hives, and they keep it permanently. The list price after that is $99/mo, and later cohorts keep the price they joined at for 24 months. The Stripe product carrying both prices is named Solo Hive; the business is Meadowrank.
Key concepts
| Term | Definition |
|---|---|
| Territory | The niche being worked, from a taxonomy of 211 niches across six clusters. Free players work the shared territory board. |
| Quarry | Custom personal targets, paid only, aimed at a specific kind of person rather than the whole territory. |
| Wingbeats | Energy. The daily allowance spent to Scout and to Commit. Wingbeats are spent, never used up. |
| Pollen | Earned soft currency, spent on streak shields, an extra scan, and cosmetics. It is never sold for money. |
| Signal | A post or comment worth replying to. Five types: Pain, somebody naming a problem; Intent, actively seeking a solution; Rival, complaining about or switching away from a competitor; Discovery, an adjacent opportunity; and Visibility, which is not a buyer at all but a room the player's buyers are reading, where a good public answer gets seen. |
| Signal score | A 0 to 100 relevance score per signal, from territory or quarry match strength, recency, platform weight, engagement on the source post, and signal type. Above 70 is high priority. It is stored in a column named buzz_score, which was deliberately not renamed: the identifier is legacy and the player-facing label is "signal score". |
| Forage | Sending a real reply and marking it sent. The core scored action. |
| Honey | A forage that got a real response, meaning a reply, a question or a conversion. In-app jargon, so prefer "a response" when writing for a general audience. |
| Hazel | The guide character, a scout bee. She/her. |
| The Clearing | The match-3 board itself. |
Coming soon
Nothing here is available today, so do not build against it.
- Mobile apps. The web app works on a phone. Native App Store and Play builds follow, and the store build sells nothing: entitlements live in the account, and there is no in-app purchase.
- A public v1 API, an MCP server, and an OpenAPI document. Planned, unscheduled, and unspecified.
- More game modes, on a published calendar, after the current loop has settled.
How the public niche pages are made
Every page under /for is built from a daily sample of real conversations, scored by the same classifier as the game. No number on those pages is typed by hand; each is computed from the live sample. A person writes and reviews each page, and the reviewer is named on the page and on meadowrank.com/methodology, which also states that the posts themselves and links to them are never stored.
This page is corrected as the product changes. Check meadowrank.com/agents for the current version.