annotated-bad-post

What this file is. The same source material as second-post.md, written the way most agency blogs would write it. It is fluent, positive, and structurally tidy. A lazy review would pass it. Every paragraph below is followed by a margin note naming the doctrine law or voice rule it violates, and what the shipped post did instead. This file teaches faster than the instruction files do; read it once before operating the engine.

⚠ Nothing in this file may be published or quoted. Several of its numbers are INVENTED to demonstrate the failure mode. checks.py fails it with 25 hard errors (the margin notes quoting banned vocabulary add to the count, fittingly); the rubric scores it 13/50 with four hard-gate failures.

How We Leveraged AI to Transform Email Outreach for a Leading Education Company

⟿ MARGIN: fails headline rules + voice.md §8. "Leveraged", "Transform", "Leading" — three banned words in one headline, zero specifics, and it's the title of ten thousand vendor posts (differentiation gate: instant fail). The shipped headline names an earned detail instead: "A cold-email tool for people who can't connect WhatsApp Web."

In today's fast-paced business environment, effective outreach is more important than ever. Companies everywhere are discovering that artificial intelligence can revolutionize the way they communicate with prospects. In this post, we'll explore how we helped one of Europe's leading education companies unlock the full potential of their outreach.

⟿ MARGIN: fails Law 4, Law 8, voice.md §9.13. Three sentences, zero information. The scene- setting opener the spec explicitly bans, a banned "revolutionize/unlock", and "In this post, we'll explore" instead of just starting. The shipped post's first 25 words contain the thesis and a number.

Their team was struggling with time-consuming manual processes that were holding them back from reaching their goals. Sending cold emails required significant effort and expertise, which many team members simply didn't have. They knew there had to be a better way.

⟿ MARGIN: fails Law 2 (specificity is the product). "Time-consuming manual processes" could describe any company on earth. Who struggled? How many centres? Which tools failed them? The shipped post: 100-plus centres, five named tools, and a persona blunt enough to hurt ("struggles to connect WhatsApp Web to a phone"). Also "significant" — banned vague quantifier.

We implemented a cutting-edge AI solution powered by state-of-the-art large language models. The platform seamlessly integrates with existing workflows and empowers every team member to create professional email campaigns in minutes. It's not just a tool — it's a complete transformation of how outreach gets done.

⟿ MARGIN: fails Law 1, Law 8, voice.md §8 + §9.1, and the em-dash ban. Four banned words, the not-just-but construction, an em dash, and not one verifiable claim. Note what's missing: ANY mechanism. The shipped post explains canonical markers and a function that throws on unknown tokens — a paragraph an engineer could re-derive the system from. This one could have been written without ever seeing the system.

The results speak for themselves. Response rates improved by up to 40%, and the team now saves over 15 hours per week on average. One manager told us: "This tool has completely changed the way I work — I can't imagine going back!"

⟿ MARGIN: HARD-GATE FAILURES, Law 6 (no fabrication). The record contains NO measured business results: the 40% and the 15 hours are invented (orphan numbers — checks.py catches both against the intake), "up to" is the banned weasel range, and the quote does not exist in quotes[] (fabricated quotes are the engine's cardinal sin). The shipped post says plainly: "Business results are not yet measured", and explains the sign-off gate — which reads MORE credible, not less.

Of course, no project is without its challenges. There were some technical hurdles along the way, but our experienced team was able to overcome them quickly thanks to our agile approach and deep expertise.

⟿ MARGIN: fails Law 3 (show the hard part) — hard gate. This paragraph gestures at difficulty while revealing nothing: no named obstacle, no failed attempt, no mechanism, no cost. It is the opposite of the shipped post's affordability-band section, which names a reputational landmine and shows the structural fence built around it. A post whose hard part is "hurdles we overcame quickly" didn't have one, and fails the gate.

The rollout was a huge success, and the client couldn't be happier with the outcome. The platform is now an integral part of their daily operations, driving efficiency, scalability, and innovation across the entire organization.

⟿ MARGIN: fails Law 1 + voice.md §9.7 (tricolon of abstractions). "Efficiency, scalability, and innovation" is the exact banned tricolon. "Couldn't be happier" is a feeling asserted on the client's behalf with no quote to back it. And note there is no limitations section anywhere — the shipped post names concurrency caps, an enqueue-only spend check, and unmeasured results, because naming a limit is the strongest trust signal available (Law: honest limitations, rubric criterion 8).

Ready to transform your outreach? Get in touch today to learn how AI can revolutionize your business — the future of email is here!

⟿ MARGIN: fails voice.md on five counts. Banned "transform/revolutionize/the future of X is here", an exclamation mark, an em dash, a rhetorical question, and a hard sell. The shipped close: one upward reframe ("That is what most operational AI should look like") and one quiet line. Also missing entirely: the transferable insight (Law 10) — this post gives the reader nothing they could use without hiring anyone, which is why it reads as an ad, and forgettable.

Why a lazy review passes this, and this engine doesn't

It is grammatical, upbeat, "on message", and shaped like a case study. Nothing in it is embarrassing sentence by sentence. The failures are all failures of ABSENCE: no specifics, no mechanism, no named hard part, no limits, no traceable numbers, no insight to take away. Absence is invisible unless a gate forces you to look for presence. That is what the rubric's ten criteria and the checker's orphan-number pass are for.

Scorecard against the rubric: specificity 1 · hard part 1 (gate fail) · stakes 2 · readability 2 (banned patterns, em dashes) · voice 1 · momentum 2 · insight 1 · limitations 1 · differentiation 1 · traceability 1 (gate fail: two orphan numbers, one fabricated quote). Total 13/50, four hard gates failed (fabrication, no hard part, banned patterns, orphan numbers). Compare second-post.md: same raw material, 47/50.

Addendum: the second face of slop (added 2026-07-24)

The post above fails by ABSENCE: no specifics, no mechanism, no limits. Operator review of this engine's own early essays surfaced the opposite failure, which passes every gate above and still reads as AI: the OVER-COMPOSED essay. Every section ends on a coined aphorism, tricolons appear five times ("closed, enumerated, and tested"), every paragraph lands a punch, confidence never varies, and the close generalises to all of humanity. Fluent, dense, and unmistakably machine.

The tells and their budgets now live in voice.md §11 (≤2 aphorisms per essay, ≤2 deliberate tricolons, varied paragraph endings, at least some workmanlike stretches, no grand universal closes), and checks.py warns on punch-ending and triad saturation. The fix is subtraction: cut the aphorism, keep the sentence before it. It was usually already the point.