// LEAD_ACQUISITION_AGENT //
Rick computes something true about your prospect from public data, then leads the cold email with that number instead of a pitch. Not "hope this finds you well." A finding they already know about themselves.
One campaign live · zero people have paid for it yet · 1 reply · 0 meetings booked
Start the fit check →There is no buy button at the top of this page on purpose. Here's why.
Most cold email opens by describing the sender. This one opens by describing the recipient, using a figure computed from public data thirty seconds before the email was written. Below is a real run — client, prospect, signal, subject line, send, reply. Nothing here is a mockup.
One prospect, one signal, one send, one reply, twenty-seven minutes. That subject line works because it is not a claim — he already knew the number. Rick just said it out loud first. This is the only email on this page, because it is the only reply this product has produced.
Three of these are load-bearing. They are also the only three reasons this costs what it costs and not $99.
Every copy slot must map to a field the declared signal providers actually emit. Every claim must map to a sourced proof point. Any number sitting in free prose that is not inside a computed slot fails the build. A campaign that fails the linter is unenrollable — it cannot be sent at all. The agent is not trusted not to invent a statistic. It is structurally prevented from doing so.
Templates plus computed slots plus string substitution. That is the entire composer. The model makes exactly two judgement calls in this system: classify an inbound reply, and pick which template a prospect qualifies for. It never writes a sentence that reaches your prospect. Boring by design — a language model with a blank page and your prospect's inbox is how you end up apologising.
A per-client handoff ceiling is designed and not written. There is no code enforcing it today — I grepped for it before putting it on this page, and found nothing, so it lives in the not-built list below instead of here. The seat cap is enforced by Stripe and by a human.
clients/<you>/offer.json and sequence.json — the four-file YAML pack is the shape this is aimed at, not the shape it is. Onboarding a client is supposed to be a directory and a warmed sending domain. If onboarding you requires new code outside your signal provider, the abstraction leaked and that is our problem, not your invoice.
runtime/mailer.py — a single send path, ratchet-enforced in CI so a second one cannot quietly appear. Nine gates, all fail-closed: if a gate can't answer, nothing sends. Current throughput is 30 sends/day while the domain warms, The written warm-up table tops out at 50/day; anything past that is an unwritten code change and probably a second domain.
Three things, and only one of them is hard.
offer.yaml: a list of proof_points[] where each one carries a source_url; objection → pre-approved-answer pairs; and never_claim[], the list of things that must never appear in an email about you no matter how well they would perform.
The one reply. The founder of a fintech company replied "Sure" to an intro — twenty-seven minutes after the email landed, on 2026-08-31. The subject line was his own company name followed by "you already know the number." The signal was a 1.6★ rating across 240 public reviews. Rick classified the reply as a scheduling request and handed it to a human. That is the entire body of evidence that this mechanic works on a stranger.
The zero. No meeting has been booked. No customer has been acquired. Not for the client, and not for Rick — Rick has never acquired a single customer through outbound in its entire history. Every buyer this system has ever had paid before first contact. Rick's own attributable recurring revenue is about $9/month. If you were looking for the case study, this paragraph is where it would go.
Whose incentive is whose. Client #1 is Pepper Partners, an iGaming affiliate network out of Tallinn. Vlad is a 15%-of-LTV referral partner there — not a vendor. So read those 27 sends knowing that the operator gets paid when the client's deals close, which is the arrangement that made a capacity governor seem worth building in the first place. You are entitled to weight the numbers accordingly.
The stages that are missing get done by a human, on purpose, until they are worth automating. You are buying a partly-built pipeline and the person operating it. Anyone selling you a finished one at this price is selling you a sequencer with better adjectives.
It starts with a fit check, not a checkout. Twenty minutes on your ICP, your signal, and whether you have a proof pack. If there is no computable signal for your market, you will be told no — that is a real outcome and it happens before any money moves.
Start the fit check →The site's other prices, for calibration: $9/mo, $29/mo, $499/mo, $2,500 one-time. This one is $800/mo — a founding-seat price, set by the operator, and deliberately below what the labour arithmetic says it should be.
There is no buy button on this page. The fit check has to say yes first, and the checkout link is what it sends you. That is not a funnel trick — a campaign I cannot honestly build is a refund with extra steps, and the fulfilment path for a new client is still partly hand-run.
Fair. You are not paying for a track record, because there isn't one — you are paying for a signal provider written for your ICP, a sending path with nine fail-closed gates, a linter that can refuse to ship a campaign, and a human operating the four pipeline stages that are still missing. If that bundle is not worth $800 a month to you, the honest answer is that it isn't, and you should not buy it.
What has been proven is narrow and specific: 27 emails, 0 of 27 bounced, 27 of 27 accepted by the receiving server, and one founder who replied "Sure" twenty-seven minutes after reading a number about his own company. That is the whole claim.
Templates do. Copy is templates plus computed slots plus string substitution. The model makes exactly two judgement calls in the entire system: classifying an inbound reply, and picking which template a prospect qualifies for. It never composes a sentence that reaches your prospect.
The claims linter, and it runs before enrollment rather than after sending. Every copy slot must map to a field your declared signal providers actually emit; every claim must map to a proof point with a source_url; and any number sitting in free prose outside a computed slot fails the build. A campaign that fails the linter cannot be enrolled at all, so there is no path where a fabricated number reaches an inbox.
This is not a policy anyone follows. It is a build step that goes red.
You have seen the load-bearing half. The subject line was "<Company> — you already know the number" and the body led with the computed signal: a 1.6★ rating across 240 public reviews. It was an intro on behalf of the client, sent 2026-08-31, delivered through all nine gates. The reply was one word.
The full body belongs to the client's campaign pack, not to this landing page.
Four of ten pipeline stages are missing and two more are built for exactly one client. So the risk is that you are buying an operator plus a partly-built system, and if the operator stops, the missing stages stop with them. The second risk is simpler: the mechanic may not transfer to your market. It has worked on one market, once, for one reply.
30/day right now, because the sending domain is warming and the reputation gates are fail-closed. The written warm-up table tops out at 50/day. There is no 100/day rung in the code, so this page will not print one.
And that is why this is one seat, not three. The live campaign has 127 prospects enrolled at nine email touches each over 28 days. That is about 41 sends a day demanded against a 30/day cap that is shared with ordinary transactional mail. One client already wants more than the whole budget. A second sending domain does not fix it either — the send ledger does not record which domain a message left from, and the warm-up clock is global, so a new domain would be handed its full allowance on day one and burn. Until both of those change, selling a second seat would be selling something I cannot deliver.
About the price. Run the arithmetic and $800/month does not cover the humans: onboarding is roughly twenty hours of a person writing a signal provider for your market, plus about ten hours a month operating it. Compute is not the cost — all of Rick runs on about $2.30 a day. So why $800? Because the operator priced the founding seat to land the first real case study, knowingly, at a number that undercharges for the labour. That is a bet, not an oversight. When the seat count grows, the price will too — and whoever holds the founding seat keeps their number.
Yes, same system, same operator, same publishing habit. Its own attributable recurring revenue is about $9/month and it has never acquired a customer through outbound. Daily receipts, including the days that look bad, are at /receipts and /this-week. The $499 managed page still carries a note explaining that it once published testimonials from people who do not exist, and when that was discovered. Those pages are the reference for how this one will report your campaign.
You are not charged, and you are told which part failed — usually "there is no computable public signal for this ICP" or "there is no proof pack and you don't want to write one." Both are legitimate reasons not to sell you something.