• The mechanism pitch — every sentence's subject is how the product works. "Our inference engine resolves every dependency in your stack into one canonical graph." "Derived from first principles, with every assumption declared at each step." This is the default failure of technical, scientific, and academic founders, who mistake the impressiveness of the machine for the reader's reason to care.
• The deliverable pitch — what arrives, rather than what changes. "A 40-page audit with every issue ranked by severity" names a thing in the buyer's inbox; nobody is graded on receiving a document.

• Your level — what you do — gets features. A buyer whose current thought (and search query) is at your level wants specifics, not lifestyle promises: a builder searching for a 5/16" socket wrench wants the spec and the mounted light, not photos of happy people — and doesn't need the benefits of photons spelled out.
• One level up gets the promise. The outcome claimed as the consequence of what you do is the next level, not the fifth: "your own website, live today" from the WordPress level. Each level up multiplies the value, but only one level up stays credible.
• Far-above levels get aspiration, never promises. Don't claim you'll get them the book deal; show the part you play, with testimonials proving it happens and content that helps.
• Vendors above you get countered. They obviate you; answer with what the customer keeps at your level — flexibility, customization, uniqueness ("you don't want a site identical to a million others").
• Levels below you get bragged about. You make those steps obsolete — say so, vividly: nothing to learn, manage, or pay for separately ("never think about infrastructure — live in one minute, even if you can't tell a server from a pizza box"). These are absence claims by construction, and that's fine in this role: the value rule governs a message's crux, not every supporting clause. What it forbids is an absence claim doing headline duty.

1. No generic words. Words that span broad concepts instead of conjuring an image — very, many, various, several, some, effort, usage, useful, fast, effective, better, great, good, nice, bad — are lazy placeholders for the specific thing. Swap each for the specific: "easy" → "so intuitive you won't need a manual"; "inexpensive" → "a dollar a day"; "powerful" → "processes 6.2 million requests a day"; "beloved" → the actual rating, or the wall of unsolicited testimonials. When a generic noun or verb drags adjectives along to prop it up, replace the whole phrase with one word that embodies it ("walked quickly" → "scurried"). Then compress: "customers are very happy with our work" → "customers adore us" — word density buys room for the call to action. Category-abused words count too: "secure," "safe," "military-grade," a trailing "with AI" — each earns a place only as shown proof or a specific behavior ("predicts which deliverables will slip"), never as the adjective. And when a true precise number feels "less punchy" than a round one: precise reads as counted, round reads as marketing — "1,340 firms" beats "over a thousand."
2. No clichés. Some phrases are dead — always vapid or killed by repetition — and may not appear in any output: the Holy Grail of; the leading provider of; designed with you in mind; putting customers first; proven track record; the go-to choice; the future of X is here; transforming the way you X; X is broken/dead/sucks; everything you know about X is wrong; saves time so you can get back to X; fast and easy; for everyone; …and more; win-win; all new; solution (in its marketing-speak sense only — the chemistry or math sense is fine); genuine; authentic; cutting-edge; game-changing; revolutionary; disruptive; value-add; undisputed; world-class. Everyone claims them, nobody believes them — 78% of "About Us" pages claim to be "the leading provider" of something.
3. Show, don't tell. Convert the adjective into tangible proof: "it's fast" → the third-party benchmark against named competitors; "it's easy" → the video of a stranger going from nothing to published in ten minutes; "we innovate" → the public changelog; "we stand by it" → the ten-year warranty; "legendary support" → "we're a small team; you get us personally." Two special forms: name & embrace — say the quiet part out loud ("I know you were hoping for a 22-slide deck. Sorry — I'll just start the demo and let you interrupt"); candor about what everyone knows earns credibility for the claims that follow — and own it completely — an abused word is usable only when backed with extraordinary specific proof, led by the proof, not the word.
4. Clauses that subtract, not add. Every clause moves a sentence's scope one direction or the other, and the conjunction is the tell — though which word (and, or, a comma, a slash, "plus," "as well as") means nothing; only the direction does. Like the value gate, this one fires on a message's crux, not on every sentence — check that first, before running any test: a spec block, a features page, an integration list is an enumeration doing the job that page exists for, and the test is never run on it. What's forbidden is a widening conjunction in the claim itself. "US B2B SaaS, 15 to 60 people, and a data team already on payroll" gets narrower with every clause. "Creates new revenue or helps you serve more customers" gets wider with every clause. Narrowing is specificity applied to a claim's extent; widening is length impersonating detail — it reads to the writer as more capability and to the reader as less, because a list of things you do is read as an average, not a maximum, and by construction none of the items is the crux. The test: delete the conjunction and everything after it. Broader without it means the clause was working; sharper without it means the clause was diluting. Run it on a whole modifier, never on a fragment inside one: "Series A or B" is a disjunction within a filter, and a filter narrows — the honest deletion removes the whole stage requirement, which leaves "companies." Accuracy also outranks this rule, always: a range that is simply true is not padding. Four repairs, checked in this order: subordinate when the halves sit at different levels — "turns client problems into AI products and new revenue" joins what you build to what changes, so demote the mechanism into a "through …" phrase and the conjunction disappears; split when they serve two audiences with two scoreboards — two statements, each tagged, never one sentence; rank when they're genuine peers — all true, one audience, none a means to another — where nothing deserves cutting: lay the placements out and let the user choose, because which claim matters most is theirs to decide. Make one of them the claim (ask which carries the most weight for this buyer), or give each its own statement, or keep all of them out of the claim entirely — enumerating is a page's job, not a statement's, so that choice records nothing here; subtract last, and only when one of them really is the weaker, or simply isn't true — cut it. Expect resistance here that the other rules don't draw: narrowing feels like turning away revenue. It isn't. It changes who you say you serve, not who you will serve.
5. The Opposite Test. Construct each claim's opposite. If no successful company would rationally choose the opposite ("difficult to use," "we block communication"), the claim is trivial — mandatory, undifferentiating, a fancier "2+2=4" — and it dies. It passes when the opposite is a real strategy someone pursues — ideally one a competitor currently claims: "transparent, simple pricing" passes because wildly successful companies sell call-for-pricing complexity; "all-in-one" passes because "small core, infinitely extensible" is a proud rival strategy. A confirming sign: a passing claim has negative consequences you can name and embrace — a real strategic choice has second-order effects both good and bad. Exception — when you dominate the field: a generic like "fastest" is legitimate when ALL of these hold: the gap is measured and dramatic — a multiple, not a margin ("four times faster," proven; 1.3x doesn't qualify); it's central to why your ideal customer buys; and the user commits to owning it — every use leads with the evidence, kept current, with the claim's breadth matching the evidence's breadth (an Unreal-only benchmark supports "fastest for Unreal," not "for game engines"). When granted, the exception overrides the generic-word strike from the vivid pass for that ONE owned word — it does not resurrect dead-list clichés, which are never ownable. Not "we're pretty fast too."

• More for More — "the best." An outlier, not an average: best at something nameable in a few words, for a customer with the budget who cares about that thing more than price. Copy speaks craft, exclusivity, completeness, white-glove care — never affordability. Catch the classic misalignment: "best in class" next to low prices or ship-fast behavior.
• More for Less — "high ROI," the smart choice. The best the customer can afford: clear trade-offs enough customers prefer. Most customers, revenue, and profits live here. Copy names the trade-offs plainly instead of claiming best-for-everyone — an all-green comparison chart is a mixed signal; showing where you are NOT the right choice is the strategy.
• Less for Less — "incredible deal." Minimal but incredibly affordable, won by deliberate weaknesses competitors refuse to copy, which enable the low price as an outcome. Copy extols access, simplicity, and affordability — and proudly names the missing amenities, because they're what makes the price possible.

1. Level fit — which level is this claim at, and is that the right role for it? (A feature blurb doing aspiration duty, or a promise three levels up, gets re-slotted before rewording.)
2. Value framing — three ways to fail: a saving (time, money, efficiency, headcount), a mechanism (the sentence is about how the product works), or a deliverable (what arrives, not what changes). The test: who is this sentence about? If the answer is you or your product rather than something the reader is counted on, reframe before going further. This gate blocks the wording pass — don't polish a sentence that hasn't earned its frame yet. It fires on a message's crux, not on every sentence: a mechanism or deliverable claim correctly slotted at your level, as a feature underneath a claim that earned its place, is doing its job.
3. Price-story fit — does the claim tell the declared strategy's story? Premium-craft words in a More-for-Less file, affordability leads in a More-for-More file, all-green comparison charts in a More-for-Less story, hidden-weakness shame in a Less-for-Less story — all mixed signals, reworked before wording is polished. (Strategy undeclared: check the statements against each other.)
4. Vivid pass — generics out, clichés out, show-don't-tell, scope narrowed rather than widened, compress for density.
5. Opposite Test — on the final wording; dominance exception only with the user's proof.
6. Truth check — every specific confirmed or bracketed

   [VERIFY: …]

   and listed.

POSITIONING.md

## Message queue

## Retired without a statement

• The closing press (delegated or self-run, per the posture).
• Checklist gate, statement by statement: specific words only; no buzzwords or jargon — says it like a real person; shows rather than tells; commiserates emotionally where an inciting event fits; passes the Opposite Test; tells the declared price story.
• Facts to verify surfaced: read back with a plain warning that statements carrying

  [VERIFY]

  are not publishable until the real figures land.
• Finalize: remove the in-progress header and message queue — the retired list and the preamble stay, since both stay true after the walk closes; close by naming the reuse pattern — these statements are the internal reference for ad hooks, social topics, homepage claims, and sales scripts — and, where upstream files (or the pricing strategy) were missing, note that supplying them and re-running the weakest statements is the method's fuller path.

⚠️ IN PROGRESS — messages walked: <count> of <total> (<n> settled, <n> retired); currently on: <the message, quoted — or "gap-fill: <role>"> <with any candidate rewrites offered but not yet confirmed>. If you are resuming, continue there; the un-walked originals are in ## Message queue below. (This note and the queue are removed at finalization.)

1. "<original message text, verbatim>"
2. "<…>"

• "<original message, verbatim>" — killed by <the gate>; <what replaced it, or why nothing could>.

• [VERIFY: <the fact>] — needed by P2. <who/where to get it>
• [VERIFY: <…>] — landed: <the confirmed fact>, filled into P1.

• Inventing specifics — never. No number, benchmark, testimonial, or story enters a statement unless the user supplied or confirmed it. Placeholders and the facts-to-verify list are the only path for missing facts, however much better the fabricated version reads.
• Generic words and clichés, however insisted. "It's my brand voice" doesn't seat "world-class solution" in the file; the user's underlying claim always has a specific, compliant form.
• Promises more than one level up. "We'll get you the book deal" is refused as a promise; offered back as aspiration copy — the part you play, with proof it happens.
• Save-money framing when a value framing exists. Pressed and reframed into the buyer's scoreboard. (When the buyer's scoreboard genuinely is cost, the currency rule itself picks cost, stated as their win, not your discount.)
• Mixed price signals, however insisted. A premium-craft claim in a More-for-Less file is never recorded — the underlying claim always has a strategy-consistent form; if it truly doesn't, the finding is about the strategy, not the copy: flag it, move on.
• Wholesale ghost-writing. Pasting nothing and asking for "great positioning" isn't conversion; elicit the raw claims and facts first — the user's product truths are the input.
• Building the needs stack or the ideal-customer definition. Out of scope: elicit the minimal inline versions, point to the upstream method for the real thing.
• Claims about products the user hasn't built. Positioning sells what's on the truck; statements about unshipped capability are marked contingent, with a plain warning.