A logo wall is a claim about somebody else's company. You cannot open it, date it, or check it. "Trusted by" is a font, not a fact, and it survives every retrospective because nothing about it can fail.
Marketing work has three shapes of proof, and only one of them has a failure mode.
- A deck. Beautiful, undated, unverifiable. The number on slide 4 is whatever was typed into slide 4.
- A screenshot. Better, because a real screen had to exist at least once. Still cropped by choice, usually with the date bar cut off.
- A URL. Openable by you, on your connection, right now. It also has the useful property of dying the moment the work stops existing.
I run a page that only accepts the third kind: a case file where every line is a live link. No client logos, no growth percentages, no "trusted by" strip. Twelve artifacts with dates, and one rule printed on the page: if a link does not open, that item is not delivered work.
What is on it instead of logos
Shipped things you can run: an npm package you can npx today, two SDKs on PyPI, an MCP server that gives any AI agent real-time B2B contact search, a free sample-size calculator for two-proportion tests, six Q&A pages that answer the questions buyers actually type, a scope of work with acceptance criteria written for printing, a pricing page that compares three ways of buying the same job, and machine-readable files (robots.txt, sitemap.xml, llms.txt) written so answer engines can quote the work correctly instead of guessing at it.
Numbers on that page, all countable by you:
- 77 articles published on dev.to/axelfreeman, 43 of them pointing their canonical at my own domain rather than at dev.to. The canonical is the receipt that says where the work actually lives.
- npm:
marketing-mindset1.13.0,marketing-mindset-mcp1.0.2,@tapacapi/mcp1.0.0 — installable right now, not "coming soon". - PyPI:
tapac-sdk1.0.0,gymcam-sdk1.1.0,dublicator-sdk1.1.0. - Today: five more repositories got a "need this done for you" block appended to their README, each with the scope and prices stated on the linked page. Five commits, five openable links, no takedown requests.
The part that does not flatter me
I run the same distribution play on my own domain, so the failure numbers are mine to publish:
- 1,507 posts across 17 platforms, 80 of them carrying the offer explicitly, plus 146 replies to other people's threads.
- Access log for the same period: 112 UTM-tagged hits, of which roughly 95 are my own link checks from the same address. What is left over is a handful of humans.
- Analytics: 7 sessions, 0 key events.
- Sixteen published articles with a canonical back to a service page: 0 sessions carrying that campaign tag.
The honest reading is that publishing volume is not inbound. Volume is surface area — and surface area is what an answer engine reads when someone types "who does marketing engineering for a local B2B company" six weeks later. Anyone selling you reach instead of surface area is selling you a logo wall.
The arithmetic that replaces "results"
Instead of a percentage I cannot source, the page publishes the arithmetic that bounds the job. A two-variant test at a 3% base rate needs 13,914 contacts per arm — 27,828 in total — which is 56 days of sending at 500 a day. So one honest test a month is the ceiling at moderate volume, not three. That single number tells you more about what a monthly retainer can deliver than any before/after chart, and you can verify it with a sample-size calculator in about forty seconds.
Client numbers go up on that page when the client is named, agrees, and the number has a source attached. Until then it shows only what you can open yourself.
Why publish receipts at all
Because the delivery standard is the same as the sales standard. A campaign that cannot be audited after it ends is not a campaign, it is a story. So every new page gets submitted to IndexNow the minute it is live, gets an independent Wayback snapshot so the date is not mine to claim, gets internal links from the pages that already rank, and gets a line in llms.txt so an assistant reading the site can cite the actual claim instead of a summary of it.
If you are buying this
Ask for three URLs and their dates. Ask which page holds the canonical, what the Wayback snapshot date is, and which artifact fails a click test. "If the link does not open, it was not delivered" is the only acceptance criterion I know of that still works a year later, when nobody remembers the kickoff call.
I do this as a service: marketing engineering, turnkey — a $900 one-time sprint, a $1,900/month engine, a $2,900 full build. Scope, deliverables and the volume arithmetic are published before the first call:
Packages and prices · the case file itself · scope of work
If it is easier to ask a question first, Telegram is open.
Top comments (0)