listen · detectors

Detector kinds

The five kinds of monitoring rule.

In Detectors

Detector kinds: the five ways a monitoring rule decides it found something.

A detector is a monitoring rule: where to watch and what to match for one trigger. Every one of the 74 detectors in the library is one of five kinds. The kind decides what the rule can read, how noisy it will be, and what the signal it emits carries.

Triggers → Detector catalog → Detector kinds → Signal catalog

Pick by where the moment shows — Words, a number, a dated record, a changed page, or your own events.

Different kinds stack — Two kinds on the same person is the strongest evidence.

Your data first — If the person is already in your funnel, no scraping is needed.

what it is

Same trigger, different evidence.

Industry names: phrase detectors are social listening or keyword monitoring; thresholds and events are intent data and trigger events; change detection is competitive monitoring; your-data detectors are product signals or first-party behavioral triggers. They differ in what they read, not in what they produce: every kind emits a signal with its own detector id, a link or record, and a date.

Phrase — Words in a post or comment, in the buyer's own language. · 30 detectors in the library

Threshold — A number on a public record crosses a line. · 18 detectors in the library

Event — A dated record appears: a launch, a job post, a funding round, a review. · 11 detectors in the library

Change — A page you watch changes: pricing, changelog, status page. · 1 detector in the library

Your data — An event in your own analytics, app, forms or Stripe. · 14 detectors in the library

choosing a kind

Start from where the moment becomes visible.

Where does the — moment show up?
words in a post or comment
phrase — asks, complaints, vents
a number on a public record
threshold — upvotes, rating, MRR
a dated record appears
event — launch, job post, funding
a page you watch changes
change — pricing, changelog, status
your own systems
your data — app, forms, Stripe

two kinds beat one

Most triggers can be caught more than one way. Two detectors of different kinds firing on the same person (a low-vote launch listing and a "launched, crickets" post) is the strongest evidence a cold signal can carry. A judge can sit on top of any kind; it changes precision, not the kind.

the kind spec

What each kind reads, and what it costs.

The object every detector shares is on the Detector catalog. What changes by kind is the shape of match and the fields it emits. The sample is a threshold detector written out in full.

Kind	Reads	Match form	Noise	Cost to run	Typical window	In the library
phrase	posts, comments, threads	phrases + negatives + judge	high	feed polling plus judge calls per candidate	3 to 14 days	30
threshold	launch sites, reviews, revenue boards	field compared to a line	low	page or API reads, few judge calls	7 to 30 days	18
event	job boards, news, listings	new record of a type, filtered	medium	listing reads, dedupe across boards	14 to 60 days	11
change	competitor pages	diff between snapshots	low	one snapshot per page per day	1 to 7 days	1
your data	app, analytics, forms, Stripe	condition across your events	lowest	your own events, no scraping	1 to 60 days	14

Field	Type	What it holds
`kind`	enum	phrase \| threshold \| event \| change \| owned
`reads`	field list	Which fields of the source item the rule can see.
`match`	rule	Phrases for phrase; field + operator + value for threshold; record type + filters for event; watched fields for change; a condition across events for owned.
`judge`	optional	A question with a rubric and a minimum confidence, for any kind.
`emits`	field list	Always detector_id, a link or record and a date; plus what this kind can read.

a threshold detector, as data

{
 "kind": "threshold",
 "id": "det.tried-it-and-it-failed.uneed",
 "version": 1,
 "source": "uneed",
 "reads": [
 "listing.name",
 "listing.domain",
 "listing.launched_at",
 "listing.upvotes",
 "listing.spotlight"
 ],
 "match": {
 "field": "upvotes",
 "op": "<",
 "value": 10,
 "within": {
 "field": "launched_at",
 "days": 7
 },
 "boost_if": {
 "spotlight": true
 }
 },
 "judge": {
 "question": "Is this a software product sold by a small team, not an agency or template?",
 "min_confidence": 0.7
 },
 "strength": "strong",
 "fresh_for": "7 days",
 "exclude": [
 "agencies",
 "templates",
 "non-software listings",
 "icp.not_if",
 "lead_rules.never"
 ],
 "emits": [
 "detector_id",
 "link",
 "date",
 "owner",
 "field_values"
 ]
}

the five kinds

Each kind, in depth.

Phrase — industry: social listening, keyword monitoring · 30 in the library

Words in a post or comment, in the buyer's own language.

use when: The moment is said out loud: an ask, a complaint, a vent, a hiring post.
avoid when: The moment never gets written down (a quiet churn), or the words are too common to separate.
noise: cheap to build, noisiest to run
strength: weak to strong; strong only for direct asks
window: 3 to 14 days
emits: detector_id, link, posted_at, author, matched_text
example: "what do you use for" {job} on Reddit, 7 days, strong
breaks on: Sarcasm, vendors answering themselves, very common pain words.

mechanism

Pull new items from the source on a schedule
Fill placeholders from the brief: {job}, {category}, {competitor}
Keep items that hit any_of and all_of phrases, drop none_of
Send survivors to the judge with the trigger's rubric
Score strength, check freshness and exclusions, emit

Threshold — industry: threshold alert, rule-based scoring on public records · 18 in the library

A number on a public record crosses a line.

use when: A public number shows the moment: upvotes on a launch, a star rating, MRR, review count.
avoid when: The number isn't public, or the line is arbitrary and you haven't checked it against real examples.
noise: little noise, limited to sources with numbers
strength: medium to strong
window: 7 to 30 days
emits: detector_id, link, date, owner, field_values
example: upvotes < 10 within 7 days on Uneed, strong
breaks on: Records with no reachable owner; vanity metrics that don't mean pain.

mechanism

Read the record and its fields (upvotes, rating, MRR, launch date)
Compare against the line, e.g. upvotes < 10 within 7 days of launch
Attach the owner (maker profile, company domain)
Optional judge: is this a real product in the ICP?
Emit with the field values frozen

Event — industry: event monitoring, intent data, trigger events · 11 in the library

A dated record appears: a launch, a job post, a funding round, a review.

use when: The record itself is the moment, and its date tells you the window is open.
avoid when: Records arrive long after the moment (news lag), or the event is too common to mean anything alone.
noise: medium; depends on how clean the listing is
strength: medium to strong
window: 14 to 60 days
emits: detector_id, link, event_date, owner
example: new job post titled {role} at a company in the ICP's size range, strong
breaks on: Agencies posting for clients; duplicate listings across boards.

mechanism

Watch for new records of a type
Filter by the ICP's size and category
Read the event date, not the capture date
Join to a person or company
Emit with event_date

Change — industry: web change detection, competitive monitoring · 1 in the library

A page you watch changes: pricing, changelog, status page.

use when: One change moves many people at once: a price rise, a plan removed, an outage, an acquisition.
avoid when: You can't connect the change to people yet. Pair it with a phrase detector for the spike that follows.
noise: low noise once layout is ignored
strength: strong as an event, weak per person
window: 1 to 7 days
emits: detector_id, page, before, after, seen_at
example: {competitor}/pricing differs from yesterday on price or plan names
breaks on: Cosmetic redesigns; A/B tests showing different prices to different visitors.

mechanism

Snapshot the page on a schedule
Diff against the last snapshot, ignoring layout
Classify the change: price, plan, end-of-life, outage
Emit one event for many people
Arm phrase detectors on the competitor's name for the next few days

Your data — industry: first-party behavioral triggers, product signals · 14 in the library

An event in your own analytics, app, forms or Stripe.

use when: The person is already in your funnel: they replied, signed up, stalled, paid, failed to pay, cancelled.
avoid when: Never the wrong tool when the data exists; the limit is permission, not detection.
noise: most reliable; no scraping, no third-party terms
strength: usually strong
window: 1 to 60 days
emits: detector_id, person_or_account, event, at
example: invoice.payment_failed from Stripe, strong, 1 day
breaks on: Test accounts and your own team; seasonal usage dips.

mechanism

Subscribe to your own events (webhooks, app events, form posts)
Evaluate a condition across events, e.g. signed_up AND no activation within 24h
Check permission: what may we send?
Emit with the account or person id
Hand to the lifecycle play

all 74, by kind

The whole library, filtered by kind.

Every detector from the Detector catalog.

Trigger	Source	Kind	Match	Strength · fresh
Asking for a recommendation cold	Reddit	phrase	`"what do you use for" {job}` · `"best tool for" {job}` · `"any recommendations" + {category}`	strong 7 days
Asking for a recommendation cold	X	phrase	`"recommend" {category}` · `"what do you use for" {job}`, with a product link in the author's bio	medium 3 days
Asking for a recommendation cold	Threads	phrase	the same asks on Threads	medium 3 days
Asking for a recommendation cold	Hacker News	phrase	Ask HN titles with `{category}` + "recommend" or "what do you use"	strong 7 days
Asking for a recommendation cold	Indie Hackers	phrase	forum posts asking which tool to use for `{job}`	strong 7 days
Asking for a recommendation cold	Discord, Slack and Skool	phrase	the same questions in member communities	strong 7 days
Comparing options cold	Reddit	phrase	`"{competitor} vs"` · `"alternative to {competitor}"` · `"{competitor} or {competitor}"`	strong 14 days
Comparing options cold	X	phrase	the same comparisons, plus `"switching to"`	medium 7 days
Comparing options cold	G2	event	a new review on a `{competitor}` page that names an alternative it was compared with	weak 30 days
Comparing options cold	Google Alerts	phrase	`"{competitor} alternative"`	weak 30 days
Unhappy with their current tool cold	G2	threshold	new review of `{competitor}` with `rating ≤ 3`	strong 30 days
Unhappy with their current tool cold	Capterra	threshold	new review of `{competitor}` with `rating ≤ 3`; note role, industry and company size	strong 30 days
Unhappy with their current tool cold	Trustpilot	threshold	new review of `{competitor}` with `rating ≤ 2`	medium 30 days
Unhappy with their current tool cold	Apple App Store	threshold	`rating ≤ 2` on `{competitor}`'s app	medium 30 days
Unhappy with their current tool cold	Google Play	threshold	`rating ≤ 2` on `{competitor}`'s app	medium 30 days
Unhappy with their current tool cold	Shopify App Store	threshold	`rating ≤ 2` on a competing app; the review names the store	strong 30 days
Unhappy with their current tool cold	Chrome Web Store	threshold	`rating ≤ 2` on a competing extension	medium 30 days
Unhappy with their current tool cold	Reddit	phrase	`"cancelling {competitor}"` · `"switching from {competitor}"` · `"{competitor} is too expensive"`	strong 7 days
Unhappy with their current tool cold	X	phrase	the same phrases on X	medium 3 days
Competitor shake-up cold	Competitor pages	change	`{competitor}/pricing` or changelog changes: a price rise, a removed plan, an end-of-life notice	strong 3 days
Competitor shake-up cold	Competitor pages	event	the status page reports a major outage	medium 1 day
Competitor shake-up cold	News and funding	event	acquisition or shutdown news naming `{competitor}`	strong 7 days
Competitor shake-up cold	Reddit	phrase	a spike of posts naming `{competitor}` with angry words, after one of the above	medium 3 days
Tried it and it failed cold	Uneed	threshold	launched in the last 7 days with `upvotes < 10`; a paid Spotlight slot makes it stronger	strong 7 days
Tried it and it failed cold	PeerPush	threshold	`publishedAt` in the last 7 days and low votes	strong 7 days
Tried it and it failed cold	Product Hunt	threshold	launched in the last 7 days, low votes and no comments	medium 7 days
Tried it and it failed cold	Hacker News	threshold	a Show HN in the last 7 days with `points < 5` and no comments	medium 7 days
Tried it and it failed cold	BetaList	event	listed in the last 14 days, read with a post about no traction	medium 14 days
Tried it and it failed cold	Other launch sites	threshold	a paid or featured slot with single-digit upvotes (Smol Launch, Fazier)	strong 7 days
Tried it and it failed cold	X	phrase	`"launched"` + `"crickets"` / `"0 signups"` / `"no sales"`	strong 7 days
Tried it and it failed cold	Reddit	phrase	the same phrases in founder subreddits	strong 7 days
Tried it and it failed cold	Meta Ad Library	threshold	ads running 60+ days for a store with few reviews	medium 30 days
Trying again cold	Product Hunt	event	the same product or domain launching a second time	strong 14 days
Trying again cold	Other launch sites	event	the same domain listed twice across launch sites	strong 14 days
Trying again cold	X	phrase	`"relaunching"` · `"launching again"` · `"v2 is live"`	medium 7 days
Venting about the problem cold	Reddit	phrase	pain phrases in the buyer's own words, from audience research	weak 14 days
Venting about the problem cold	X	phrase	the same phrases on X	weak 7 days
Venting about the problem cold	Threads	phrase	the same phrases on Threads	weak 7 days
Venting about the problem cold	Discord, Slack and Skool	phrase	the same phrases in member communities	medium 14 days
Doing it by hand cold	Reddit	phrase	`"template for" {job}` · `"spreadsheet"` or `"google sheet"` + {job} · `"I built a script"`	strong 14 days
Doing it by hand cold	X	phrase	the same phrases on X	medium 7 days
Doing it by hand cold	Indie Hackers	phrase	posts sharing a homemade workflow for `{job}`	medium 14 days
Doing it by hand cold	Discord, Slack and Skool	phrase	template and spreadsheet requests	strong 14 days
Set up, but no results cold	Revenue boards	threshold	`customers > 0` and `MRR = $0`, or flat for 3+ months	strong 30 days
Set up, but no results cold	Shopify App Store	threshold	a listing live 6+ months with 0 or 1 reviews	medium 30 days
Set up, but no results cold	Apple App Store	threshold	an app live 6+ months with very few ratings and a recent update	medium 30 days
Set up, but no results cold	Meta Ad Library	threshold	ads running 60+ days for a store with very few reviews	medium 30 days
Crossed a threshold cold	X	phrase	`"just crossed"` · `"just hit"` + users, customers or revenue · `"we raised"`	medium 14 days
Crossed a threshold cold	News and funding	event	funding for a company in the ICP's size range	strong 30 days
Crossed a threshold cold	Revenue boards	threshold	MRR crosses a line that matters, like $1k or $10k	medium 30 days
Hiring for the job you do cold	Job boards	event	a new post titled `{role}` at a company in the ICP's size range	strong 30 days
Hiring for the job you do cold	Hacker News	event	the monthly "Who is hiring?" thread, filtered for `{role}`	medium 30 days
Hiring for the job you do cold	X	phrase	`"hiring a" {role}` · `"looking for a VA"`	medium 14 days
New in the role cold	LinkedIn	event	job change into `{role}`, or a new company page	strong 60 days
New in the role cold	X	phrase	`"quit my job to go full-time"` · `"just opened my practice"` · `"day 1 at"`	medium 30 days
Taking stock cold	X	phrase	`"month 1"` · `"MRR update"` · `"year in review"` with numbers	medium 7 days
Taking stock cold	Reddit	phrase	the same posts in founder subreddits	medium 7 days
Taking stock cold	Indie Hackers	phrase	milestone posts with flat or zero numbers	medium 7 days
Taking stock cold	Revenue boards	event	a new revenue-board listing	weak 30 days
Engaged with your content warm	Replies and mentions	your data	a reply, mention or comment from someone who isn't a lead yet	medium 3 days
Engaged with your content warm	Forms and email list	your data	a new newsletter subscriber	weak 7 days
Hand-raise warm	Replies and mentions	your data	a reply under your offer post that contains a URL or asks for the offer	strong 2 days
Hand-raise warm	Forms and email list	your data	a lead-magnet submission, or answers to waitlist survey questions	strong 2 days
Checked your pricing warm	Site analytics	your data	an identified visitor views `/pricing` twice in 7 days	strong 3 days
Signed up, didn't start warm	Your app	your data	`signed_up` and no activation event within 24 hours	strong 1 day
Trial ending, not converted warm	Stripe	your data	`customer.subscription.trial_will_end` (3 days before the end) with low usage	strong 3 days
Went quiet warm	Your app	your data	no login or key action for 30 days after some use	medium 30 days
Payment failed warm	Stripe	your data	`invoice.payment_failed`	strong 1 day
Using it less warm	Your app	your data	key actions down by half or more over 4 weeks against the 4 before	medium 7 days
Outgrowing their plan warm	Your app	your data	hits a plan limit, or adds a second product, location or teammate	strong 7 days
Outgrowing their plan warm	X	phrase	the customer announces a second product or location	medium 14 days
Just had a win warm	Your app	your data	a milestone event: first result, first payment caught	medium 7 days
Just had a win warm	Replies and mentions	your data	a happy reply, mention or survey score of 9 or 10	strong 7 days
Cancelled warm	Stripe	your data	`customer.subscription.deleted`, with the reason from `cancellation_details` if your cancel flow asks	strong 60 days

failure modes

How each kind goes wrong.

Kind	Failure	How to catch it
phrase	Matches the words, misses the meaning: sarcasm, quotes of other people, vendors answering themselves.	A judge with a rubric, plus none_of negatives learned from every false match.
phrase	Vocabulary written by the founder, not the buyer.	Take phrases from audience research in the buyer's own words; review misses monthly.
threshold	The line is a guess (why 10 upvotes?).	Pick the line from 30 days of real records, then check precision on a sample.
threshold	Records with no owner you can reach.	Mark the detector research-only; route its signals to content, not outreach.
event	News lag: the event is old by the time it is captured.	Freshness on event_date, never on capture time.
change	Layout changes or price tests read as a price change.	Diff only the fields that matter: prices, plan names, end-of-life text.
your data	Test accounts and your own team fire triggers.	Exclude internal accounts by id; label them in the app.
any	The source changes format and the detector silently returns nothing.	Source health: alert when a detector that normally fires goes quiet.

data to capture

Record the kind with every match.

So precision and yield can be compared across kinds, not just across detectors.

record

kind and detector version on every match
per run: items scanned, candidates, judge calls, matches, cost, duration
for threshold rules: the field values at capture, frozen
for change rules: the before and after snapshot

label

founder real or not on a sample per kind
signal became a lead, a reply, a join, a payment

trains

precision and yield per dollar, by kind and source
which kind combinations predict a reply (stacking weights)

test

phrase vs threshold for the same trigger, on the same weeks
line value for a threshold (under 5 vs under 10 upvotes), offline on stored records

Detector kinds: the five ways a monitoring rule decides it found something.

Same trigger, different evidence.

Start from where the moment becomes visible.

What each kind reads, and what it costs.

Each kind, in depth.

Words in a post or comment, in the buyer's own language.

A number on a public record crosses a line.

A dated record appears: a launch, a job post, a funding round, a review.

A page you watch changes: pricing, changelog, status page.

An event in your own analytics, app, forms or Stripe.

The whole library, filtered by kind.

How each kind goes wrong.

Record the kind with every match.

Where to go next.