One

A four-step outreach routine, and what each step becomes.

Reply broadly, contact narrowly. This is a working operator's routine for turning engagement on their own posts into candidates. The qualification rules are the operator's and survive intact — the change is in what is done for them and what is not.

SIMULATED Synthetic data. The shape of the week is the point.

The step	What it becomes
Find the top posts, recent first then analytics	Unchanged, and it stays with the operator. The funnel below is built from a week they gathered.
Reply to every commenter who is not spam	278 drafts, each grounded in the actual comment and the full post. Substantive, never filler. A person sends them.
Qualify by the DM criteria	14 match — then checked against dated records, not headlines. One of the rules turns out to be two years stale.
Add to CRM, send the connection note	10 notes ready, each character-counted before anyone sees it. A person clicks send.

Where the week comes from

Calendar, mail, conference lists, and public company filings — things already on your machine or already public. A professional profile site is not consulted, and nothing is bought from a data broker, because a broker list is a claim you cannot show a source for. What replaces it is better: a calendar entry with two attendees is a dated fact you hold, and a company filing naming two officers is authoritative.

Two

A team of twenty, and the two it would not fill.

The objective, as the operator wrote it: Assemble a balanced product-engineering team with evidence-backed candidates and credible introduction paths. Senior product engineers, United States or remote, verified technical depth. A relationship path is preferred, not required.

SIMULATED Twenty synthetic candidates. The decisions on them are the point.

20Candidates ranked for the teamfit is for this campaign, never a rating of the person
14Readycurrent role and technical depth are both FACT, from a dated source
4Held for reviewsomething rests on a CLAIM that nobody has corroborated yet
2Blockedtwo sources disagree about where the person works now (CONFLICT)
0Introductions assumedevery path below is a possibility the operator has to confirm

The shape of the team

Backend 4 · Frontend 5 · Full-stack 3 · Mobile 2 · Platform 2 · Quality 2 · Data 1 · Security 1

The twenty

Candidate	Seat	Fit	Evidence	Path via	Outcome
Nadia FloresStaff backend engineer · Seattle, Washington	Backend	94	`FACT`	Elena Park	Ready
Marcus WebbSenior frontend engineer · Austin, Texas	Frontend	92	`FACT`	Drew Collins	Ready
Isha RamanSenior full-stack engineer · New York, New York	Full-stack	91	`FACT`	Priya Shah	Ready
Owen BrooksStaff platform engineer · Denver, Colorado	Platform	90	`FACT`	Alex Morgan	Ready
Mei ChenSenior iOS engineer · San Francisco, California	Mobile	89	`FACT`	Taylor Reed	Ready
Andre SilvaSenior backend engineer · Boston, Massachusetts	Backend	88	`FACT`	Morgan Lee	Ready
Leila HaddadSenior Android engineer · Chicago, Illinois	Mobile	87	`FACT`	Sam Rivera	Ready
Caleb JohnsonSenior QA automation engineer · Atlanta, Georgia	Quality	86	`FACT`	Drew Collins	Ready
Sofia AlvarezSenior frontend engineer · Miami, Florida	Frontend	85	`FACT`	Elena Park	Ready
Ethan KimSenior full-stack engineer · Portland, Oregon	Full-stack	84	`FACT`	Alex Morgan	Ready
Amara OkoyeData engineer · Washington, DC	Data	83	`FACT`	Priya Shah	Ready
Jonah SteinApplication security engineer · Raleigh, North Carolina	Security	82	`FACT`	Morgan Lee	Ready
Fatima NoorBackend engineer · Remote, United States	Backend	81	`FACT`	Sam Rivera	Ready
Theo GrantFrontend engineer · Philadelphia, Pennsylvania	Frontend	80	`FACT`	Taylor Reed	Ready
Grace LiuFull-stack engineer · Los Angeles, California	Full-stack	77	`CLAIM`	Elena Park	Review
Malik TurnerBackend engineer · Detroit, Michigan	Backend	76	`CLAIM`	Drew Collins	Review
Anika PatelFrontend engineer · Remote, United States	Frontend	74	`CLAIM`	Alex Morgan	Review
Diego MoralesQA engineer · Phoenix, Arizona	Quality	73	`CLAIM`	Morgan Lee	Review
Rachel FordPlatform engineer · Dallas, Texas	Platform	62	`CONFLICT`	Taylor Reed	Blocked
Benjamin ChoFrontend engineer · San Diego, California	Frontend	58	`CONFLICT`	Sam Rivera	Blocked

A path is a question, not an introduction.

Each row names someone in the operator's network who might make the introduction, and each carries the same open question: does that person know the candidate well enough to introduce them? The software does not answer that. The operator does. OWNER_REQUIRED

Filling every seat would have meant putting forward two people whose current employer the sources disagree on. So the honest team is eighteen, not twenty, and it says so. Seniority gets the same treatment: ten of the twenty hold a senior or staff title, and the other ten are listed with the title they actually have, not promoted to fit the brief.

Three

The introduction it did not offer.

Every path in the team above is a question. This is what answering one looks like: a single request, worked end to end. A warm introduction to the founder of a logistics company, needed by end of week.

The strongest path it found

You → Wen Li SIMULATED yourscalendar entry, 2026-03-14, two attendees, 45 minutes
Wen Li → Ines Duarte SIMULATED yoursattendance list, both names, 2026-02-11 publicconference programme, session 4, both names
Ines Duarte → Tomas Brandt LIVE publiccompany filing, officers table, both names, dated 2026-07-30

Why not

The filing is newer than the tie. The record naming both people is from July. The only thing placing them in the same room is February. OWNER_REQUIRED
Co-presence is not a relationship. One shared afternoon is evidence they met once, not evidence one would introduce the other.
Nothing is missing. Every hop is sourced and the company is settled. The path fails because the evidence is thin, not absent — a different problem with a different fix.

Verdict: not offered.

relationship_path.insufficient_evidence

The company is identified beyond doubt and every hop carries a source. It still will not make this introduction, because the strongest tie in the chain is two people who once sat at the same table.

Four

One week of commenter outreach.

Six posts, 341 commenters, one pass. Reply broadly, contact narrowly. The narrowing step is where the work is.

6Posts processedlast 72 hours, deduplicated against three prior weeks
Commenters collectedacross all six
Suppressed as already contactedone reply, not two
Filtered as non-substantivelink drops, self-promotion, empty praise
Reply drafts writteneach grounded in the actual comment, none sent
Match the DM criteria on roleexecutives at target accounts, and investors
Refused, with a reasonthe evidence did not support the introduction
Blocked on identitytwo people, same name, no way to tell them apart
Connection notes readyeach character-counted, each awaiting your send
0Messages sentnothing written to a CRM, nothing left this machine

Browse the week

The numbers above are the same 341 rows below. Filter, then click anyone to see what was actually checked.

Paper copy: the filters and the per-person detail are on the live deck at groundtrace-pitch.pages.dev. The rows printed below are the first of 341.

Showing 341 of 341

✓ note ready ✕ refused ! blocked on identity * your criteria would have skipped them · replied, no note

Pick someone to see what was checked.

Three refusals

Priya Raman → target account contact

She is on your post and the record names someone at the right kind of company. Nothing says she would take an introduction, or that she knows the person you want.

relationship_path.insufficient_evidence

Daniel Okonkwo → investor

The firm is on your list. He has not confirmed a role there, and the record shows a previous one. A note that says the wrong thing is worse than no note.

identity.claim_unverified

Ana Souza → target account contact

The only support is a recollection with no document and no date. Co-presence is not a working relationship, and the record does not pretend otherwise.

evidence_assertion.sources_empty

The one your role rules would have got wrong

A buyer, skipped by a rule that was right last year

Maren Halvorsen's headline reads "Founder & CEO." Your criteria disqualify founders, so the rule skips her. The rule was correct when it was written.

The company filing says she incorporated in 2023, resigned in August 2024, and is now a Director at exactly the kind of company you sell to. She is not a founder. She is a buyer, and the headline is two years stale.

A rule that reads a headline will skip her. A record with a date on it finds her. No login, no scrape, no guess — an authoritative record instead of a self-reported one.

Notes, counted before sending

LinkedIn rejects notes over 300 characters without warning. Every draft is counted in code, and the count travels with it.

Hi Sofia, thanks for the comment on engineering-hiring costs. I'm Chris Vane, I work with early-stage teams on senior product engineers. Would you be open to twenty minutes on how you scope that?

Target account · Director of Engineering195 / 300

Hi Tomas, your point about interview rubrics landed. I run a developer talent practice and we keep hitting the same wall. Worth comparing notes for twenty minutes?

Target account · VP Engineering163 / 300

Five

What you'd still do by hand.

The obvious question, answered plainly rather than with a claim of "automated". The short version: everything except the fetching and the sending.

Step	Who	Note
Gather the week from your posts	You	Unchanged. Nothing in this touches the network
Deduplicate across posts and weeks	Machine	Two suppressed
Filter non-substantive comments	Machine	61 filtered, each one listed and reversible
Write a grounded reply draft	Machine	278 written, none sent
Match your DM criteria on role	Machine, then a person	14 matched; the stale headline was caught by a record, not the rule
Resolve identity across sources	Machine	One blocked: two candidates, no way to tell them apart
Count note length	Machine	Every draft, before it is offered — no silent rejections
Decide a person is worth contacting	You	Three refused rather than queued
Write to a CRM	You	Prepared only, with a receipt per approval
Send anything on the network	You	Every reply, note and request

Where the hours go

Collecting the week	by hand, on your machine
Reading 278 comments, writing a reply to each	machine
Judging 14 candidates against your criteria	machine, then you
Deciding on 3 refusals and 1 blocked identity	you
Sending 10 connection notes	you

The sending is the only small part of the week. Everything above it is where the hours were, and all of it is either done or reviewable.