The corrections record

That a contested route “does not accumulate,” that “a direct route is the only one you can act on,” and — as the heading over all three — that of the three routes “only one of them is an asset you hold.”

The 1 September entry below, which had already retired exactly this language: rented and contested routes plainly do accumulate — income, craft, credits, reputation — and what fails to accumulate is renewable access. The correction was made in the instrument and never swept back to the homepage, so the front door of this domain spent five days contradicting the record two clicks away. Found by an external review reading the two pages against each other.

The distinction is now renewable access rather than accumulation, in all three beats and in the opener’s contested reading. The direct route is described as the one whose terms are yours, not the only one worth having.

Two paragraphs, one under the other. For a business whose customers all arrive through partners and hosts — Granted Share 100%, Rent Index 1.00 — the first said the demand “feels like an asset and behaves like a lease.” The second said “demand like this is durable — it rests in your customers, not your ad account.” Separately, any reproduction number below 0.9 printed “growth still needs buying.”

The Read paragraph was keyed to Granted Share alone and never consulted the Rent Index the page had computed six lines above it. A high granted share with a high rent index is the eviction-exposed profile, not the durable one — so the reassuring sentence fired on precisely the position this site exists to warn about. The reproduction line was a separate error of reasoning: an absence of customer referrals does not establish that paid acquisition is required, and on this run it contradicted the same page’s finding that every customer had arrived conferred.

The Read paragraph now branches on both numbers and names both halves when they disagree. A low reproduction number now says what it observes — that customers are not yet producing customers — and never prescribes spending. The reproduction figure also carries the run’s confidence, because at ten customers one referral moves it by 0.10.

That the middle route was “Through something in between — a platform, a gallery, an agency, a publisher, a marketplace,” and that everything in it counted toward the headline finding: “N of ten arrived having already chosen you.”

The row named a channel and asked nothing about orientation, so it swallowed two different arrivals: the collector who asked a gallery for you, and the stranger a feed put you in front of. Only the first had chosen you. This is the same defect as the calculator’s category-search row, corrected three days earlier and never swept to the opener.

The middle row now states the condition before the channel — they had already chosen you, and the way back runs through someone else — and the contested row explicitly names the feed that reaches people who had never heard of you. A standing rule was added: every route row on every surface asks orientation before it asks channel.

Its eyebrow read “the public audit instrument,” its opening ended “so this is an audit rather than an opinion poll,” and the paragraph immediately beneath read “a decision aid, not a measure of trustworthiness, a certification, an audit or a prediction.” It also claimed “no model, no guesswork.”

An audit is conducted by someone who is not the subject; that property is the whole meaning of the word, and it is what the paid review actually sells. Claiming it for a self-score overstated the free instrument and devalued the paid one in the same breath. “No guesswork” was true of the arithmetic and false of the instrument: the inputs are ten human judgements, which is why every score carries an evidence tag and why the reliability question is open.

It is a scoring instrument and a structured comparison. The deterministic claim is now made about the arithmetic specifically, with the judgements named as the reader’s own. The word audit is no longer available to any self-scored instrument on this site.

That “your email list, subscribers, or followers” was a single row scoring granted and direct, and that “search (including searching your name)” was granted and direct too. The Metrics Standard’s definition of Granted Share counted “search-with-intent” as conferred.

Both rows bundled positions that the framework separates. A list whose addresses you can export is direct; followers on a platform are rented, because the platform decides whether they ever see you again — the exact conflation the book spends a chapter correcting. And the search row was wrong on both axes: someone searching your name chose you and arrives somewhere you control, while someone searching a category or “near me” chose nothing and your listing competed for the click. That is contested, and it was being counted as granted demand.

Eleven rows became thirteen at channel table v1.2. The list row splits into a permissioned list you could export (granted, direct) and followers or subscribers on a platform (granted, rented). The search row splits into a search for you by name (granted, direct) and a category or “near me” search (contested). The Standard’s Granted Share definition now reads a search for you by name, published as v1.2.

The vocabulary pass of 2 September renamed this instrument’s owned to direct and stopped there. Renaming the label made the wrong mapping underneath read as corrected. The reusable lesson: when a term is retired, every place that assigns it has to be re-derived, not only every place that prints it.

That the Granted Share proxy for creators was “owned-channel opens + returning viewers”, and that the Creator play computed “GS = owned/returning consumption ÷ total” with Conferral Premium as “revenue per owned-audience member.”

The same argument that corrected the creator audit the day before: nobody owns people, and the property that matters is control of a direct, portable route back. It is worse in a standard than in an instrument. An instrument in retired vocabulary gives one person a stale label; a standard’s definitions are the thing other people are invited to adopt, so the retired model propagates into every measurement built on it.

Direct, in all three definitions, published as v1.1. No formula, threshold, proxy or predicted outcome changed — this is a vocabulary revision and the revisions log says so. A second thing was resolved with it: the calculator had been stamping every result and share card “Conferral Metrics Standard v1.2” while this page still published v1.0, citing a version that did not exist. Publishing the corrected definitions as v1.1 makes that stamp true instead of leaving a phantom version behind.

That the third position was Owned — that the creator “holds the relationship” and that this is “the part that compounds.” Contested attention, it said, was one where “nothing accumulates on its own.”

Three things. Creators do not own people; the property that actually matters is whether they can start the next contact themselves and move the route to another provider. Rented routes plainly do accumulate — income, craft, credits, reputation — and what fails to accumulate is renewable access, which is a narrower and truer claim. And the implied advice was wrong: a session musician, a jobbing actor or a commissioned illustrator can build a durable career across several institutional relationships and never hold a direct audience at all.

The position is now Direct, defined as control of direct, portable access, with ordinary replaceable infrastructure explicitly not counting as a landlord. The compounding claims are gone. The next move no longer assumes conversion is the goal — where no route is portable, it asks whether the creator depends on a single one, and offers a second landlord or firmer terms alongside building a route. The three lease questions — what you keep paying, who can change the terms, what survives if it closes — were added, for rented routes only.

Two of the seven papers treated algorithmic distribution as contested attention — a contest the feed rewards whoever wins. Two others treated it as rented conferral — a floor an intermediary grants, on a relation the recipient does not own. Worse, in Part Three the stale account was in the abstract while the sharper one was already in the body, so the summary of the paper disagreed with the paper.

The same event cannot be two things. Being picked up by a recommendation system is, from the recipient’s side, a conferral: the floor arrives already granted. What makes it different from owned conferral is not that it was contested but that the conferring relation belongs to someone else. The contested/granted distinction was never wrong — it was carrying a second question inside it without saying so.

Part Three’s abstract now states what its body already argued: the platforms began as engines of owned conferral and converted themselves into engines of rented conferral, keeping the appreciation. That version is also stronger, because it explains why the feed felt like a gift for a decade before it stopped — the old version had to treat a decade of users as mistaken. Part Five gains a paragraph: the captured floor is itself rented, which is the reason attention-rich authority is unstable rather than merely fickle. Both papers carry a dated revision note.

That the first-party event beacon records “which band your result fell into,” giving “Design Scorecard: capture-financed” as the worked example.

Scorecard v2.0, published the same day, removed band labels entirely. The beacon now carries the score. Nothing was collecting a band any more, so the page was describing a practice that had ended — on the one page where an inaccurate description is a compliance question and not only a credibility one.

The paragraph now says the beacon carries the single headline value, gives the current example, adds channel names to the list of things never sent, and states that the creator audit sends nothing at all.

The offer above the email field said the profile is written out and sent by hand — usually within a few days. The confirmation that appeared after submitting said something different and fixed: “Sent — your full profile is on its way. The short series follows over the next five days.”

The offer text switches on which fulfilment mode is live; the confirmation did not, and carried the automated-series wording in every mode. The mode in production is the manual one. So anyone who submitted an address was told a five-day sequence was coming that nobody had built and nobody was going to send.

The confirmation now switches on the same setting the offer does, and in the current mode says the profile is written by hand and to give it a few days. If you gave an address before this was fixed and expected a series, that is the reason it did not arrive, and the apology is mine.

A proposed measure, Initiated Share, defined as the fraction of sessions a person began without the product having prompted them — with any session opened inside a stated window of a notification, email, badge or re-engagement message counting as prompted. The stated interpretation was that initiated use was closer to wanted use and prompted use closer to captured use.

User-configured scheduled tasks. A person authorises a product to do work later; the product does it and notifies them; they open the result. Under the v1.0 definition that session counts as prompted, so the measure falls precisely because the delegated service worked as requested. Both ChatGPT and Gemini document this pattern in their own help material. The prompt was the delivery mechanism of a prior grant, not evidence of capture.

The session-origin distribution: eligible sessions reported across eight disclosed origin classes, with no weighted aggregate and no assumption that any class is intrinsically good or bad. Unprompted initiation survives as one class among eight.

That both proposed numbers already existed inside these companies, so publishing them was a matter of willingness rather than engineering.

Working through what a session-origin measure actually requires: a controlled prompt taxonomy, a link from each delivery back to the authorisation that created it, cross-surface identity, exposure states, a versioned attribution window, and an honest representation of unclassifiable cases. That is a project.

The claim now holds for withdrawal only, and says so. Withdrawal events are close to already-instrumented; session origin is engineering work. The page states which half survived.

That the first company to publish one of these numbers makes a commitment a competitor can only match by publishing their own.

A self-defined, unaudited number published once is cheap for anyone to produce. The definition can move, an unflattering period can quietly not appear, and nothing follows from misstating it. First publication alone does not separate a trustworthy product from an untrustworthy one.

The separation comes from the apparatus, not the act: a definition fixed in advance and versioned when it changes, unknowns reported rather than absorbed, bad periods left visible, corrections preserved with dates, and eventually independent assurance. The page now argues that, and this record is part of it.

A total score placed the product in one of four named bands — Conferral-designed, Granted core, contested edges, Capture-financed, Engagement-economy replica — and any tripped red flag capped the result at Capture-financed regardless of the score.

Two problems. The band names are diagnoses of a business model, produced by a rubric with no validity evidence behind it; a label cannot claim more than the instrument can support. And an unvalidated aggregate does not become more meaningful by being overridden — a finding that matters does not need to distort a total in order to matter.

v2.0 reports the criterion profile and earned / available points with no band label at all. Red flags are reported separately as non-compensatory findings, each with its own rationale, and they no longer change the total. Criteria 2 and 8 were re-anchored to match the session-origin correction above. v1.0 is archived rather than rewritten, and three further criteria are openly marked as under revision.