Gtm

GTM

Overview

The GTM harness for this portfolio. Read gtm/GOAL.md and gtm/RULES-OF-ENGAGEMENT.md before acting — they are required_sources for every verb, not background reading. gtm/SWARM.md explains why it's shaped this way.

The one thing to know first

The bottleneck is not that we lack customers. It's that three doors are closed.

The product with world-class distribution has no way to pay us; the product with a world-class wedge cannot be seen. Open doors. Don't knock on others'.

Task Index — pick the right verb

There is no outreach verb. Not an oversight — see GOAL §5 and RULES §7. The swarm's MCP token is minted without outreach:send, so it is a capability this harness does not possess, not a rule it chooses to obey.

Usage

scripts/gtm-verify.sh            # re-check every number in GOAL.md §2

/gtm audit --all                 # highest-value autonomous loop
/gtm research "who ships Clay alternatives in 2026"
/gtm measure omnivoice

The three rules that carry the weight

1. No primary-source URL, no action. A vendor blog, an SEO article, or your own recollection is not a source. If you can't produce one, refuse and escalate.

⚠️ You cannot verify a statistic with AI search — search summaries reproduced a known misattribution live, twice, during this project's research. They confirm errors rather than catch them.

2. Check the artifact, not the claim about the artifact. Every real finding in this project came from running the thing:

• The README's own clone URL 404s anonymously — every user who followed our quickstart failed on line one.
• resubird.com/pricing returns "Page Not Found" while linked from nav and footer.
• The first-run demo filled 0 of 175 cells — one missing dict key.
• A research agent's #1 priority was "ship audio watermarking." It shipped months ago, default-on. Two weeks nearly went to building a thing that existed.

3. Data gaps are not zeros. A KPI with no measured value is a DATA GAP — leave it untouched, never write 0. A zeroed metric is indistinguishable from failure and corrupts every decision downstream, including the planner's.

Hard stops

Full list in RULES §1. The ones you'll actually hit:

• Never call the Clay alternative "Yupcha." yupcha.com is a live AI hiring platform. The name is spent. Use a placeholder and escalate.
• Never market OmniVoice as an outbound/cold-calling tool, and never demo with a famous voice. Our legal posture (general-purpose, local, offline, open-source) is strong, and our own marketing copy is the only thing that can destroy it.
• Never email stargazers. 8,509 supporters → 8,509 detractors, in one send.
• Never solicit upvotes anywhere, including off-platform. Bannable, and the HN trigger is pattern-based — one founder lost posting privileges for "repeatedly posting from a single site," with no vote manipulation at all.
• Never claim a metric we haven't measured or a testimonial we didn't receive. FTC v. TruHeight: unsubstantiated efficacy claims + seeded reviews = $4M judgment, suspended to $750k. ResuBird's 4.8/1250 vs 4.8/12 markup is that thread.
• Never publish autonomously. Ever.

Don't bother

Evidence in GOAL §5. llms.txt (97% of 137K domains got zero requests; zero AI bots ever probed for one) · schema markup for AI citation (causally tested, effect zero) · GEO tactics (fail replication; the top methods literally instruct fabrication) · chasing stars (HTTPie lost 54,000 stars overnight with zero revenue impact) · Product Hunt (91.2% of 500 launches got <100 users) · MCP directory listings as acquisition.

What this harness is for

It does: research · audit · draft · measure · prepare. A human does: anything that speaks to a stranger in the founder's voice.

The channels that work for a solo OSS founder — the Show HN that has never happened, the right subreddit, median 8.5h response latency → 71% higher odds a newcomer resolves an issue — are precisely the ones a swarm can't do for you. Automating them is how you lose them. This exists to buy back the hours so a human can spend them where they compound.