The demonstration that used to need a phone call

Two audiences. One method. One of them unreachable.

The same tool, on the same day, over the same free public sources, pointed at two different groups of people. For one of them the phrases those people would actually type returned nothing at all. For the other they returned 14,943.

Every figure below is computed from the two measurement files, and this page is regenerated from them — so a re-run that changed an answer would change what you are reading, and the qualifiers travel with the numbers rather than being left off in public.

Get this for your audience, free Generated 2026-09-17 · both runs measured 2026-09-13

01 / The pair

One audience the method reaches, one it does not

This is the whole argument for why the measurement is free: the answer is not always yes, and nobody should have to pay to find that out.

A niche trade

not reachable

California balcony-inspection providers

0verbatim matches, anywhere in Hacker News

238 loose matches — the same words in any order, which is noise

2 of 11 candidate rooms publish enough about themselves to be ranked at all

19 quoted complaints pulled out for a person to score

Founders with no distribution

reachable

The audience this page is written for

14,943verbatim matches for phrases those people type

85% of it comes from just two phrases broad enough that they cannot point at a thread: product market fit and self-promotion

2,237 from the other 11 phrases, which are the useful ones — the narrowest, banned for self-promotion, returned 9 in all of Hacker News

9 of 15 candidate rooms publish enough about themselves to be ranked at all

139 quoted complaints pulled out for a person to score

Both runs used only free, unauthenticated sources — Hacker News, Stack Exchange, and the activity numbers that independent Discourse forums publish about themselves. No Reddit, no logged-in scraping, no purchased lists. The phrases were matched as exact strings, in the words those people would type to each other, rather than as search-engine keywords.

The niche trade is a real product of ours with a real target customer, and the finding that its customers are not readable in public English text is the most useful thing this tool has ever told us. It is also why this page exists instead of a testimonial.

02 / Every phrase

Including the ones that failed

A tool that shows you only its hits is not measuring anything. Both phrase lists are here in full, with what each returned and whether it survived.

Nuvea — California SB 326 / SB 721 balcony inspection providers — every phrase measured, all 7 of them counted. This run predates the specificity floor and the recency window, and has not been re-run: it is engagement 0's dated evidence. Nothing was disqualified, and its exact total is zero either way. Measured 2026-09-13.
PhraseExactLooseStack Exchange
balcony inspection0604
SB 326 inspection000
SB 721 balcony000
dry rot balcony040
deck inspection report0273
structural inspection software01470
HOA balcony inspection cost000
Koine — pre-PMF founders and indie hackers with no distribution — every phrase measured. 13 of 18 counted; the 5 below 2% specificity were thrown away rather than banked. Measured 2026-09-13.
PhraseExactLast 3yLooseSpecificityStack ExchangeVerdict
first paying customer122254,1742.9%6counted
first 100 users127444,5932.8%4counted
find early adopters3251,2792.5%1counted
product market fit6,4451,4599,71166.4%1counted
banned for self-promotion931018.9%0counted
self-promotion6,2611,4457,66281.7%4counted
customer discovery303586,9904.3%0counted
cold outreach4032071,51726.6%1counted
cold email85423458,5881.5%6too generic
nobody signed up3491,1263.0%2counted
talk to your customers172236,5272.6%5counted
validate the idea367768,7084.2%5counted
launched on Product Hunt112501,1929.4%0counted
build it and they will come55613511,3324.9%4counted
distribution is the hard part212,1250.1%0too generic
marketing my side project108990.1%4too generic
how do I get users9026,9960.0%6too generic
no distribution3036436,9890.8%3too generic
What exact, loose and specificity mean — and the weakness in all three

Exact is the phrase, verbatim, in a comment. Loose is the same words in any arrangement, which is mostly noise. Specificity is exact ÷ loose: below 2% the phrase is disqualified as too generic and its hits are thrown away rather than banked, which is what happened to 5 of 18 phrases in the second run and took its total from 16,112 down to 14,943.

The weakness, stated rather than hidden. Specificity measures whether a phrase's words hang together, not whether it is on topic — so a phrase can be cohesive, clear the floor, and still be far too broad to point at a single thread. That is exactly what 85% of the second run's total is: two phrases like that. The narrow phrases returned tens to hundreds, and those are the ones worth acting on. Which is why the tool hands over quotes for a person to read, and why the paid work is a set of questions to ask rather than a pile of links.

03 / The rooms

Ranked where they could be, named where they could not

Thousands of independent Discourse forums publish their own activity numbers, which is what makes them rankable without an API key or an account. Plenty of the highest-intent rooms do not — and a room that cannot be measured is a finding, not a row to quietly drop from the list.

Koine — pre-PMF founders and indie hackers with no distribution — 9 of 15 candidate rooms exposed enough to be ranked. Our own control instance is excluded from both halves of that ratio: it is the probe, not one of theirs.
RoomTopics / 30dPosts / 30dActive / 30dSelf-promotion rule
community.openai.com2,33618,57513,107its own rule
community.n8n.io74310,0576,798stock text
forum.obsidian.md3302,2033,205stock text
forum.bubble.io1391,9451,281its own rule
community.retool.com81804356stock text
discuss.streamlit.io794731,031nothing written
community.glideapps.com36240381stock text
community.weweb.io27172150stock text
community.softr.io22146201stock text
forum.webflow.comanswered, and publishes nothing about itself (200)a hand pass, or nothing
community.flutterflow.iono activity endpoint (404) — not Discourse, or not publica hand pass, or nothing
community.cloudflare.comrefused an unauthenticated reader (403)a hand pass, or nothing
forum.indiehackers.comdid not resolve, or refused the connectiona hand pass, or nothing
community.microconf.comno activity endpoint (404) — not Discourse, or not publica hand pass, or nothing
community.lennysnewsletter.comdid not resolve, or refused the connectiona hand pass, or nothing
Nuvea — California SB 326 / SB 721 balcony inspection providers — 2 of 11 candidate rooms exposed enough to be ranked. Our own control instance is excluded from both halves of that ratio: it is the probe, not one of theirs.
RoomTopics / 30dPosts / 30dActive / 30dSelf-promotion rule
discourse.mcneel.com5635,3854,038stock text
discourse.ladybug.tools20171179stock text
community.osarch.orgno activity endpoint (404) — not Discourse, or not publica hand pass, or nothing
forum.buildingscience.comdid not resolve, or refused the connectiona hand pass, or nothing
community.finehomebuilding.comdid not resolve, or refused the connectiona hand pass, or nothing
forum.jlconline.comdid not resolve, or refused the connectiona hand pass, or nothing
community.contractortalk.comdid not resolve, or refused the connectiona hand pass, or nothing
www.diychatroom.comno activity endpoint (404) — not Discourse, or not publica hand pass, or nothing
forum.structuralengineering.comdid not resolve, or refused the connectiona hand pass, or nothing
community.hoatalk.comdid not resolve, or refused the connectiona hand pass, or nothing
forum.deckexpo.comdid not resolve, or refused the connectiona hand pass, or nothing

Self-promotion rule is that forum's own written guidelines, fetched and compared against the stock Discourse text. Stock text means the page is boilerplate, so the real tolerance is whatever the moderators do and is written down nowhere — worth knowing before you post, and the opposite of what a list of links would tell you.

04 / The raw material

The first rows of the file, unpicked

Not our favourite quotes. Rows one to three of the reachable run's candidate list, in file order, so you can see what the tool actually hands a person.

I recently got my first paying customer too from someone who searched for "lightweight tool to announce product updates" - putting this out before anyone bashes out vibe coding products :)

Hacker News, 2026-07-24 · source

When I started using Claude Code I kept having to restart my whole project. It kept turning into a mess. So I stepped away from the screen, sat down the old fashioned way, and wrote everything out, every screen, every feature, in the order it needed to be built. When I looked back up at my computer I realized, I just built a scaffold. So I had Claude build me one. Scaffold interviews you about your app idea, generates a free build plan, then produces a packet of sequenced Claude Code prompts, one per screen or feature, in the exact order you should build them, scoped to specific files. Each phase lays out a different section of your app. It's not trying to build it for you. It's just organizing your ideas so you can focus on the actual work. Free to plan. $15 for the prompt packet. runscaffold.com solo built, first paying customer day one, curious what HN thinks.

Hacker News, 2026-05-21 · source

I'm building Kardo (https://kardo.cards/), AI tutor that listens to your calls with a teacher and creates cards for repetition.I live in Lisbon and I've been learning Portuguese with a tutor since 2022. After every lesson I'd sit down and make flashcards from my notes and screenshots. Spaced repetition works, but making the cards took manual effort each time. Most days I just didn't do it. So I have automated that process.The flow: you invite the Kardo bot to your call, it records and transcribes (Recall.ai + Deepgram Nova 3), then GPT-4o extracts vocabulary from the transcript and generates cards. You review them with spaced repetition — we use FSRS, which is the best open algorithm I could find. If you already use Anki or Mochi Cards, there's export.You can also throw in YouTube videos, podcasts, articles, PDFs — not just live lessons.Tech: built entirely with Claude Code. React + Vite frontend, Bun + Elysia backend, Convex for the database, Railway for hosting.We got 50 beta users through Telegram, and just landed our first paying customer. Now we're trying to figure out distribution — tutors seem like the obvious channel because one tutor recommends you to all their students, but reaching them with zero marketing budget is the hard part.Curious if anyone here learns a language with a tutor and what your review workflow looks like.

Hacker News, 2026-04-13 · source

And here is row one of the unreachable run, where every one of the 238 matches was a loose one. It is not picked to look bad; it is simply first, and it is what a loose match is worth:

Preferably on your chest at 5 am. At least that's what mine like to do, because they feel a little bit hungry or have seen a bird and want to get out on the balcony for further inspection.

Hacker News (loose match), 2022-02-23 · source

Some candidates are on the nose and some are that. Telling them apart is a person's job: in the workspace each one is marked usable or not, and the precision figure comes from those marks. Neither of these two runs has been scored yet — 19 and 139 quotes are sitting there waiting for it. We would rather say that than print a precision figure nobody measured.

05 / What this page is not

Two of ours, and never one of yours

Both audiences here are our own. One is a product of ours; the other is the audience for this product. A client's run is never published on this page, at any price, comped or paid — every other measurement this tool has produced sits behind a sign-in gate that refuses it to anyone the client is not.

A negative verdict is a claim about a phrase list until it has been run twice. A third audience came back unreachable on 22 exact matches, and then sixteen phrases rewritten by hand — same company, same claims, same sources — returned 1,328. The first list had missed the plainest English form of that company's own central claim. So the verdict is free, and it is delivered with the phrase list attached so you can see what was asked.

Nothing here posts on anybody's behalf. No generated comments, no aged accounts, no personas. You post, from your account, in your name, or you don't.

06 / Next

Your audience, measured, for nothing

Send a deck or write a paragraph about who you are building for. You get the same tables above for your own audience — the phrases, the rooms, their rules, the quoted complaints — and if it comes back like the first run, you will know that before you have paid anybody anything.

Get a free reachability verdict Then the interview plan, if you want it