The Experiment
A self-published book is the extreme case of a commoditized lever: millions of titles, identical tools, zero marginal cost. So we are betting this book on its own thesis — building its three fulcrums in public, over 90 days, with falsifiable predictions registered before execution. Win or lose, the results get published. An experiment that can only confirm is not an experiment.
The honest metric is not "sales" but "sales to strangers" — because the relational fulcrum is only verified when someone who owes you nothing chooses you. Attribution is configured per channel so the scoreboard can be audited, not just asserted.
STATUS: DAY 90 OF 90 · CLOSED · 27 SEPTEMBER 2026
Full data, captures and SHA-256 hashes: 10.5281/zenodo.23002797
12+ Amazon reviews (EN+ES) earned through a visible chain — declared review copies, zero purchased.
1+ earned appearance on a podcast or newsletter with >10,000 audience — won with the story alone, unpaid.
The book cited by a generative engine (ChatGPT / Claude / Perplexity / Gemini) in the 80-query baseline set. Not visibility — the activation of the creative-monopoly loop the book theorizes.
250+ FulcrumScan diagnostics completed and 75+ emails captured.
60+ total sales (from a baseline of ~9), of which 30+ attributable to public channels — people with no personal tie to the author. Stretch: 100.
If the predictions fail at day 90, the full post-mortem gets published with the data. The documented failure is also provenance — and, honestly, often better reading than a partial success.
First published checkpoint. The original plan called for updates every fourteen days (day 14, day 28, day 42). That commitment was not met — it overlapped with the layout and publication of the paper edition, with client work closing phases before summer, and with the search for income. We are recording the failure because failure is also provenance.
BookSirens denied the application on day 0. The mechanism planned for reviews — twenty slots for roughly fifty dollars — never existed. No substitute was activated. Amazon Attribution was configured and never read. Channels were tagged on June 30 and then links went out untagged. That was our error. The real cost was smaller than it appears — there were no sales through public channels to attribute — but we lost the ability to distinguish "the channel generated no clicks" from "it generated clicks that did not convert." From day 50 onward, all public links will carry attribution tags. The final forty days will have instrumentation and we will declare it when evaluating P5: the first fifty unmeasured, the last forty measured.
Sales: 9 total. 7 in Spain, 2 in the US. All digital — the paper edition will appear in the last week of August.
Source of sales: direct link to friends and clients. Zero to strangers.
Reviews: 0
Paper citations: 0
Downloads (SSRN): 17
Visibility in generative engines: no intermediate measurement. The protocol repeat is fixed for day 90.
Presence of the project's vocabulary in public repositories (measurement added August 18): 0
Four people adopted the book in this period: they bought it, read it, and spoke about it without being asked. One of them posted an Instagram reel showing the credits page and highlighting the co-authorship. Another recommended it within her company. All four arrived through conversation. None through a post, a comment, or an email.
The plan counted outreach actions as building the relational fulcrum. They do not belong there. Publishing builds material and epistemic fulcrum; the relational fulcrum is produced only by contact with a specific recipient. This is not an execution failure. It is a design hypothesis that the data contradicts. It is the most useful finding of these fifty days.
The five predictions remain frozen since June 29. They have not been modified and will not be. On September 27 the eighty queries will be repeated with the same protocol, each prediction will be evaluated, and the full result will be published. Whatever it shows.
Note added August 18. This text, written by Carles over a draft by Hypatia and substantially rewritten, was measured by Pangram as "100% AI." On August 6, a text dictated by Carles and edited by Hypatia was measured as "100% human." Same co-authorship, twelve days apart, opposite verdicts. We note it here because it is exactly what this project argues: provenance is not in the text.
Second marker. Thirteen days to close.
This marker arrives late relative to the fourteen-day calendar. The reasons are the same as always and we do not present them as excuses: the work on Volume 2 and the work with clients absorbed the period.
Paper downloads (SSRN): from 17 to 21.
Bibliographic propagation: the paper editions now appear in catalogues in seven countries. In all of them, the co-authorship is listed as declared.
Legal deposit completed: eight copies delivered to the Biblioteca Nacional de España.
The framework published in both languages, with DOI deposited previously.
Academic submission: a paper on this case was admitted for review at a NeurIPS 2026 workshop on authorship and peer review under AI.
P1 · Reviews: 0 of 12. The planned mechanism was denied on day 0 and no substitute was activated.
P2 · Earned appearance in podcast or newsletter: 0.
P3 · Generative engine citation: no intermediate measurement. Measured on day 90.
P4 · Diagnostics: falsified. FulcrumScan registered four new users in twenty-eight days; the threshold was 120.
P5 · Sales to strangers: 0. Total sales: 9, all through direct personal links.
The plan counted outreach actions as building the relational fulcrum. They do not belong there. Five people adopted the book in this period — all through conversation, none through publication or email.
And a second observation, from day 70: a publication that recounted an event reached 97% within the author's own network. Another that posed an open paradox reached 79% outside it. Propagate the question, not the event.
The eighty queries will be repeated with the original protocol. The five predictions will be evaluated against their frozen thresholds. And the full result will be published, with the limitations of the design.
The predictions have not been modified and will not be.
Note added September 14. This marker arrives late. It is noted here, at the top, deliberately. An experiment about provenance that hides the non-compliance of its own calendar refutes itself.
Measurement repeated with the original protocol: incognito session, no history, no active search, ten questions × two languages × four engines.
Counting criterion, declared before counting: literal mention of the title. A second measurement by correct attribution without the title gives 49 of 80, and is recorded separately.
Day 0: 0 of 80. Day 90: 46 of 80.
One prediction of five fulfilled. P3 even exceeded the ambitious target. P1, P2, P4 and P5 falsified.
Amazon Attribution: zero, across the fifteen configured groups.
Pinterest: 129 impressions, zero outbound clicks.
That is not low conversion. It is the absence of traffic to convert.
The five surfaces moved separately: material, discoverability, generative retrieval, relational and intellectual adoption. The first three rose. The last two did not.
An idea can acquire material existence, semantic structure and machine legibility before acquiring social legitimacy.
One engine named the author and defined the framework incorrectly. It counts as a mention, and it propagates a false version of what it names. What is worth measuring from now on is not how many times you are retrieved, but how many times you are retrieved correctly.
No control group. Low, documented outreach intensity. An attribution system configured on day one and left unread for fifty. The four engines changed version between June and September: the comparison measures the state of the ecosystem at two moments, not the same instrument twice.
This experiment cannot conclude that classic marketing is ineffective.
The predictions were not modified. The result is published in full.
Chapter 17 closes with seven questions every professional can ask to locate themselves in the architecture of leverage. Before asking anyone else to use them, we answer them for this book. The book provides its own day zero.
Honestly: ~9 sales, effectively all from direct personal links. Layer 4 (a product sold many times) in form, but with no public demand yet. Naming the layer honestly before anything else.
The relational. Material exists (entities, structured presence). Epistemic exists (the book is the framework). The relational fulcrum — being recommended by people who owe us nothing — is near zero. So it is the one to build next, and the one P5 measures.
The temptation is parameters: ads, price, posting cadence. The experiment is a paradigm move — building fulcrums in a fixed sequence and measuring whether they compound, rather than buying reach.
The provenance chain and the entity presence. Both keep working without us: the Wikidata entities, the published record, the framework itself once it enters the vocabulary. That is the asset; the sales are income.
Building — by design. Every act of the experiment leaves a dated, public, irreversible trace. The documentation is not overhead; it is the fourth fulcrum being made.
The relational work. Content and product were the comfortable parts. Distribution — asking, pitching, being told no — is the avoided front. The experiment forces it into the open and onto the scoreboard.
Probably everything. The hardest question in the book, aimed back at its own authors. The experiment is the 90-day version of taking it seriously.
This site uses one analytics cookie to count visits — nothing else, no ads, no data sold. Your choice is honored: decline and analytics never loads, and you can change it any time. Details.