Platform-native copy means one thing: the copy for each surface is written for that surface rather than trimmed to fit it. The claim is not written three times. It is written once and then carried three times, and the carrying is where it gets changed without anyone deciding to change it.
The ordinary failure is not a bad post. It is three versions of one claim, all of which read well, of which only one still says what the evidence says. Compression removes the qualifier because the qualifier is the longest clause; expansion adds a number because a longer post looks thin without one; a change of register turns “catches” into “prevents” because the shorter word has more force. None of those edits is a decision. All three are made while looking at the surface rather than at the passage the claim rests on.
This piece takes one claim and writes it for three surfaces, naming what changes and why and on what basis, and then states what may not change under any of it. It contains no engagement rate, no follower figure, no reach or click-through, and no best time to post, because it has measured none of those and no number of that kind is a fact about your account. Where a posting time is mentioned below it is marked as a hypothesis with a test attached, and nothing else in the piece depends on it.
Six decisions, four of which are variable
A post is a set of decisions, and the useful discipline is knowing which of them belong to the surface. Of the six in Table 1, four vary by channel and two do not; the two that do not are the two that make the copy worth publishing.
Table 1. The six decisions in a post, which of them vary by surface, and the failure each one causes when it is treated as variable by accident. Constructed for this piece from the structure the Social media role produces: one register per channel, a hook, a body, a call to action and a named asset per row.
| Decision | What it controls | Variable? | The failure when it varies |
|---|---|---|---|
| The claim | The one sentence the post asks a reader to accept | No | It becomes a different claim: “prevents errors” for “catches a variance before payment” |
| The evidence | The passage behind the claim, and its as-of date | No | The version that drops it is unverifiable, and the version that substitutes a statistic is worse than unverifiable |
| Length | How much of the claim and its condition fit before the reader must ask for more | Yes | The clause cut is the condition, because the condition is the longest part of the sentence |
| The opening | The first line, which does a different job on each surface | Yes | The hook invents urgency the evidence does not carry |
| Link handling | Where the link sits, whether it is clickable, and what the reader must do to reach the evidence | Yes | The evidence moves to a comment and the claim then stands alone in the feed |
| Register | Sentence length, person, and how much of the machinery the copy shows | Yes | A change of register becomes a change of claim: “reduces” for “records”, “always” for “within tolerance” |
The four variable decisions in Table 1 are not free. They are constrained by the two that do not, and the constraint is directional: the surface decides the length, the opening, the link and the register, and none of the four may be paid for out of the claim or its evidence. A post that fits by dropping the condition has not fitted; it has published a different claim.
The claim, which is not copy
A claim has three parts and copy carries only the first of them: a sentence, a passage that supports it, and the date on which that passage was current. The passage and the date have to survive the rewrite in a form a reader can reach. On a surface where the link is clickable, that means the link goes to the passage rather than to a home page. On a surface where it is not, it means the evidence is nameable in the words of the post — a document, a section, a revision — so that a reader who doubts the sentence can find what it rests on without asking the author.
Three properties decide whether a version still carries its claim.
The verb keeps its object and its strength. “Catches a price variance before the invoice is paid” and “prevents invoice errors” are not two ways of saying one thing. The first names a check and a point in a process; the second names an outcome the evidence does not establish. A rewrite may shorten the sentence. It may not promote the verb.
The condition stays attached to the claim. Almost every claim worth posting in a business account has a condition under it — a tolerance, a threshold, a period, a scope. The condition is what makes the claim checkable, and it is also what makes it long, which is why it is the first thing a length cut removes. A claim separated from its condition is not a shorter claim. It is a claim about everything.
The attribution stays reachable. Not a gesture at a document (“our methodology”) but a pointer to the place in it: the section, the revision, the month. A reader who has to ask which methodology has been given the same position as a reader who was never told there was one.
There is a fourth property that is a rule rather than a test. A number that the client’s material does not state does not enter the copy to fill a space. A post whose second paragraph is thin is a post with a thin second paragraph; a plausible percentage in it is an error that outlives every correction made to it, because the version with the number is the version that gets screenshotted.
Length: what the reader sees before asking for more
Each surface shows a fixed amount of text and then asks the reader to press something to see the rest. The mechanism is the same everywhere and the amount is not; the number belongs to the product, changes without notice, and is not stated here for that reason. What matters is the consequence: on every surface with a fold, the length decision is not “how much of this can I cut” but “which clause can be seen without the reader asking for more”.
Two rules follow, and they are rules because the fold makes them mechanical rather than stylistic.
The clause that survives truncation is the condition, not the subject. The reader already knows what the post is about, because the surface tells them whose post it is and what the image shows. What they do not know, and cannot infer, is the tolerance, the period or the scope. Put that in the visible part. A post that leads with the subject and hides the condition in the folded text has led with the part the reader could have guessed.
Length is measured in the words the claim needs, not in the allowance. A surface that permits a long post has not asked for one. The test of a version is whether every sentence in it is doing something the claim or its evidence requires; a version that is long because the format is long has four sentences of decoration in it, and decoration is where the second claim gets invented.
The order of work matters here more than the rules do. Write the claim and its evidence first, at whatever length they take. Then set the surface’s allowance against that, and cut from the outside in — the example, the illustration, the second clause of the second sentence — until the condition is in the visible part. A version built upwards from a word count starts with the surface and arrives at a claim chosen to fit it.
The opening: a different job on each surface
The opening is the part of the copy that most obviously belongs to the surface, because what it has to accomplish is set by what the reader has already seen when they reach it.
On a surface with a fold, the opening’s job is to earn the press. It competes with everything else in a feed and it has one line in which to make the claim look worth the rest. On a surface without one, the opening is the structure of the whole post: there is nothing to expand, so the first line sets the frame and the last line lands the claim, and a post that saves its point for a link saves it for a reader who has already gone. On an image-led surface, the picture carries the first impression and the opening connects the picture to the claim rather than restating it; a caption that describes the image wastes the only line it has.
What the opening may not do is promise something the body does not deliver, and the common form of that is urgency. “Most finance teams cannot answer this” is a claim about finance teams, and a post that opens with it has made a second claim, unsupported, in the most visible position on the surface. The opening may promise attention: a question the claim answers, a comparison the claim resolves, a consequence the claim bears on. It may not promise a result.
Links: where the evidence goes, and what moving it costs
The link is where the evidence lives, and every decision about the link is therefore a decision about the evidence.
Where the surface allows a clickable link in the body, put it at the point where the claim is made, or as close to that point as the surface permits. A link at the end of a post is attached to whatever the last sentence said rather than to the claim, and readers follow it to a page that answers a question they did not ask.
Where the surface does not make body links clickable, the copy has to carry the evidence in words: the document by name, the section, the revision. That is not as good as a link and it is far better than nothing, because a reader who wants to check can search for the section while a reader given a bare claim has to trust it.
Moving the link to a comment is a legitimate choice on some surfaces and it changes who reaches the evidence: the reader who does not open comments now has the claim and not the pointer. Where the link moves, the post should say the pointer is there, in the copy, in words a reader can act on without having read the comments first.
A link to the top of a site is not a citation. It says a document exists somewhere behind the sentence without saying where the sentence came from, and a reader who follows it lands on a home page and has to reconstruct the argument the post made in one line.
One claim, three surfaces
The claim, as it comes out of the register: a three-way match catches a price variance before the invoice is paid rather than after it, with the tolerance taken from the client’s own procedure note at section 4.2 and the exceptions recorded in the monthly summary at section 8. It is a method claim, it is boring on purpose, and it has one condition — before the payment run, within tolerance — that every version below has to keep.
The three versions are written for three surface types rather than for three products. The type is what decides the mechanics: whether there is a fold, whether a link is clickable, whether an image leads. The numbers behind those mechanics differ by product and move; the mechanics do not.
The professional feed, 76 words. “Most invoice problems are found after the payment run, when the recovery is a conversation with a supplier rather than an adjustment to a balance. A three-way match moves the check upstream: purchase order, goods receipt and invoice compared before payment, and a price variance outside the tolerance in procedure note §4.2 stops the invoice rather than following it. What stopped, for whom and for how long is recorded in the monthly exception summary at §8.”
The short-form broadcast, 44 words. “A price variance found after payment is a conversation with a supplier. Found before it, it is an adjustment. That is the whole case for a three-way match, and the tolerance is not ours to choose: procedure note §4.2. Exceptions recorded monthly at §8.”
The visual grid, 27 words of caption under a comparison image. “Same invoice, same variance. Found in the payment run; found three weeks later. The difference is where the check sits, not how hard anyone looks. Tolerance: §4.2.”
Table 2 sets the three side by side, with what changed and what could not.
Table 2. The same claim on three surface types: the opening, the length as written, where the evidence sits in each version, and the thing the surface rewards that the others do not. Constructed for this piece; no platform product is described and no performance figure is claimed for any version.
| Version | Opening | Length as written | Where the evidence sits | What this surface rewards that the others do not |
|---|---|---|---|---|
| Professional feed | A consequence in the reader’s own week — “found after the payment run” | 76 words, three sentences, one clause per sentence | Named in the body: §4.2 and §8, plus a link at the claim if the surface makes it clickable | The argument made in full before the reader leaves the post; a claim with its mechanism attached reads as work rather than as promotion |
| Short-form broadcast | A contrast in the first five words: after payment / before it | 44 words, three short sentences, no subordinate clause | Named in the last line, short form: “§4.2”, “§8” | Brevity read as confidence; a claim stated flatly and evidenced in a section number looks like a settled fact rather than a pitch |
| Visual grid | An image that states the comparison and a caption that connects it to the claim | 27 words, two sentences, the image carries the mechanism | Named in the final clause, with the section number as the only text evidence | The comparison read at a glance; the picture does the explaining, so the caption’s job is to fix the reading, not to repeat it |
Read the third column down the three rows: 76, 44 and 27 words, and the same claim, tolerance and two section numbers in all three. Nothing was added in the long version and no evidence was dropped in the short ones. That is the whole test, and it is failed most often at the short end, where the space saved is spent on confidence.
The four ways a version stops carrying its claim
Table 3 is the drift as it actually happens. Each row is a real edit somebody makes for a reason that is good on its own terms, and each one produces a sentence that no longer means what the register says.
Table 3. Four edits that change a claim while looking like a change of style, with the version that results and the test that catches it. Constructed for this piece; the claim under edit is the three-way-match claim above.
| The edit | Why it is made | What the version now says | The test that catches it |
|---|---|---|---|
| The condition is cut for length | The clause “outside the tolerance in procedure note §4.2” is the longest in the sentence | “A three-way match catches price variances” — with no tolerance, which means every variance, which the procedure does not say | Ask which clause a reader could not have guessed; if the cut clause is the one that answers that, the edit has shortened the claim rather than the sentence |
| The verb is promoted | “Catches” sounds modest and the post is about the check | “Three-way matching prevents invoice errors” — a claim about outcomes the procedure note does not make | Put the verb back into the passage’s own sentence; if the passage does not carry it, the version has added an outcome |
| A number is borrowed | The second paragraph looks thin and a percentage is available from somewhere | A fill-rate or error-rate figure with no source, in the most quotable position in the post | Every figure in the copy traces to a row in the register; a figure without a row is removed, not attributed loosely |
| The scope is widened | One supplier-month reads as too narrow to be interesting | “Any invoice run” instead of “one supplier-month, one tolerance” | Read the claim’s noun phrase aloud with its qualifier; if the qualifier was needed to get it past the evidence, it stays |
The four are one failure in four costumes: the post is edited while the writer is looking at the surface instead of the passage. None of them requires bad faith and none of them reads badly. That is why the check has to be a separate pass against the register rather than a re-reading of the draft, and why the check has to happen once per version rather than once per claim: the three versions in Table 2 differ in length and opening by design, and wording is the thing that drifts.
What one claim, carrying three versions, costs
The arithmetic is small enough to follow by hand, which is the point: the cost of the check is the argument for doing it.
One claim on three surfaces, drafted and checked once each, with one revision in this instance because one version failed its check: three drafts, three checks, one revision — seven calls. At the measured rate of $0.00700 a call on Base, that is $0.04900 for one claim carried three versions, or $0.01633 a version.
Table 4 scales it to a month at the counts the Social media role publishes, with the evidence check added.
Table 4. A month of platform-native copy: twenty claims carried to three surfaces each, at the role’s published call counts with an evidence check per version added. Prices are the measured DeepSeek rate on Base on 31 October 2026; counts are the role’s planning assumptions plus the check, not observations.
| Call | Unit price | Calls | Cost |
|---|---|---|---|
| Post copy — one call a version, sixty versions | $0.00700 | 60 | $0.42000 |
| Evidence check — one call a version, against the register | $0.00700 | 60 | $0.42000 |
| Revision where the check flags a version — one version in two | $0.00700 | 30 | $0.21000 |
| Calendar rebuilt weekly — five calls a week, four weeks | $0.00700 | 20 | $0.14000 |
| Creative briefs — three calls a brief, four briefs | $0.00700 | 12 | $0.08400 |
| Pillars, cadence and the monthly read-back | $0.00700 | 8 | $0.05600 |
| Total | 190 | $1.33000 |
The arithmetic reconciles in one line. Drafting at the role’s own count is 100 calls and $0.70000 a month: sixty drafts, twenty calendar calls, twelve for briefs, eight for pillars. Adding the check and the revisions adds sixty and thirty, which is ninety — so the month becomes 190 calls and $1.33000, $0.63000 more, and the whole of the increase is the second pass. That is not a hidden cost of the assembly; it is a cost the role leaves to the person approving the calendar, whose review note is the only place a false claim would otherwise be caught.
Two bounds hold the figure in Table 4. 190 calls at the metered ceiling of $0.00955 is $1.81450, which is $0.48450, or 36.4 per cent, more: the ceiling is what an agent authorises at a thousand output tokens, not what it spends on a sixty-word post. And the 0.50 USDC credited to a workspace as working capital is seventy-one calls at the exact rate, which is 37.4 per cent of this month — so a month at these counts is funded from $1.33, not from the opening balance, and 71 calls buys twenty-eight versions at 2.5 calls a version, not sixty.
The document parts sit outside the total. Sixty rows in the calendar, four briefs, four rebuilds and one export are written, rendered and exported inside the workspace, carry no published per-call price, and are therefore not costed here — which understates the work rather than saving it.
What this piece does not know
The limitation is specific, and it is the one that makes the rest of the piece safe to use.
It holds no engagement figure of any kind. No reach, no impressions, no follower growth, no click-through, no conversion, and no comparison between the three surface types: none has been measured, and the piece is not in a position to measure one, because a post’s performance belongs to an account, an audience and a week. Any number of that kind quoted from a general source is a number about somebody else’s account.
A posting time appears nowhere in it as a fact. If one is suggested at all, it is a hypothesis and it should be written down as one: state the time, the surface and the reason, run it for four weeks, and compare the four weeks against a control period at a different time before treating it as anything. The same applies to frequency and to length: the three versions above are constructed to show the mechanics, not to claim that 76 words does better than 44.
The surface mechanics are properties of products that change without notice. The fold, the clickable link and the image-led layout are described here by type because the specifics move; the truncation point on your surface is the number you have on the day you post, not a number from this page.
And the check has a ceiling. It confirms that each version is faithful to the register: that the claim as written says what the passage says, that the figure is the figure, that the section number is the section that carries it. The four failures in Table 3 are catchable by that check because each leaves a difference between the sentence and the passage. But the check does not confirm that the register is right. A well-documented error becomes three consistent versions of an error, and three consistent versions are harder to spot than one careless one. Nothing in this piece makes a claim true; it makes the claim checkable, which is a different and smaller thing.
One more limit belongs to the reader. Three faithful versions of a true claim can still be of no interest to anyone, and no amount of copy discipline changes that: the piece is about not damaging a claim on the way to the surface, not about whether the claim was worth making.
The decision
Write the claim and its evidence before the first version, and decide at the same time which clause must survive truncation — that is the one decision that has to be made before any surface is chosen, because it is the only one the surface will force later. Then write the three versions in the order long to short, cutting from the outside in, and keep the condition and the section number in all three.
Then check each version against the passage rather than against the other versions, and treat a version that fails as a failure rather than a stylistic variant. There are two permitted outcomes for a version that has drifted: rewrite it to the passage, or cut it. Publishing it with the claim softened is the third outcome that feels like a compromise and is not one, because the softened sentence is still what a reader will quote.
And budget correctly. Sixty versions at these counts is 190 calls and $1.33000, of which ninety calls and $0.63000 are the check and the revisions; a workspace holding 0.50 USDC can fund seventy-one calls, which is twenty-eight versions, and a calendar of sixty rows on that balance will either be short or unverified. The arithmetic decides the calendar, not the other way round.
Where this goes
- The pillar: content and knowledge: from a source you hold to a sentence you can defend — the four stations, the claim list and the verdict
- The role that writes and calendars the copy: studio.x402.press/roles/social_media — pillars, platform-native copy, the calendar row and the export, and the refusal to publish
- The role that checks a claim against the register: studio.x402.press/roles/curator_editor — the citation matrix, the flagged items and the verdict that approves nothing
- Where the calls are priced: DeepSeek — the canonical per-call prices on seven chains, and the two schemes beside them
- What the handshake actually asks for: DeepSeek: per-call chat completions over x402 — the measured amounts and the ceiling behaviour
- The draft that a version must remain faithful to: turning source material into articles — the register, the claim map and the outline that follows the evidence
- Where the drift is caught: editing for one voice — five decisions held constant, and one paragraph edited twice
Method note
The price was read from the live 402 challenge on Base on 31 October 2026 at 09:36 UTC, outside DeepSeek’s peak windows of 01:00–04:00 and 06:00–10:00 UTC: 7000 atomic units under exact, which is $0.00700 a call, and 9550 under upto at max_tokens: 1000, which is $0.00955. Both figures are from the handshake in front of the call rather than from a price list, and the amount the endpoint quotes is the amount a client pays.
The call counts are the Social media role’s published planning assumptions — one call a post, five calls a week to rebuild the calendar, three calls a brief, eight for pillars, cadence and the read-back — with an evidence check per version and a revision on one version in two added to them. They are assumptions, and the added counts are this piece’s own, stated so that they can be argued with.
The claim, the procedure note, its two sections, the three versions and the four drift examples are all constructed for this piece. No client, account, supplier or procedure is described, and no document is cited by a title a reader could mistake for a real one. No engagement, reach, follower, click-through or conversion figure appears anywhere here, because none has been measured; no posting time is asserted, and the surface mechanics are described by type because the specific numbers move with the product. The three versions are an illustration of a method, not a test result.
x402.review is written and edited by N.P. Vincent. The publication is connected to x402.press, which sells the services reviewed here; that interest is disclosed on every piece that touches a service, and it is stated here rather than left to inference.
Filed under Jobs · Tags: social, copywriting, platforms, distribution
