Look at this document. Every line carries its brief and its date.
A real working document — the launch checklist for my next app, drawn from my own briefs — shown as is: nothing has been touched up, not the dates, not the gaps.
Sections 1 to 5, plus the recap of hypotheses that closes them. The counters are the document's own, not rounded figures.
Two inks. No third.
Every claim that commits is written in one of these two inks. A sentence in neither is a defect — and it shows.
The brief, the section, the date. Three things a machine can find — or not. The full slug is in the page; the screen shortens it, the check doesn't.
What the briefs don't settle is written as such, hatched rule and all. Neither green nor red: saying you don't know is the contract holding, not an alert.
Three passages from the document. As they are.
One whole section, the recap of hypotheses, and one standalone rule — copied from the document untouched: not the order, not the lines, not the brackets. Only the French has been translated for this page.
- Find my “hammer” first — the channel I can actually work (organic viral content, cold outreach, or paid ads) gets picked BEFORE building, not after [retour-d-experience-sur-14-applications-mobiles-et-saas-deve §reading · 2026-08-01].
- A viable business model from day one — the case of a product its users love but with no revenue model is documented: a hit in usage, a dead end financially [retour-d-experience-sur-14-applications-mobiles-et-saas-deve §reading · 2026-08-01].
- Validate the product / audience / channel fit before development [retour-d-experience-sur-14-applications-mobiles-et-saas-deve §checks · 2026-08-01].
- Be my own user — the projects that last are the ones where the founder is an active user; abandonment for lack of personal use is documented [retour-d-experience-sur-14-applications-mobiles-et-saas-deve §reading · 2026-08-01].
- Copy to validate, plan the innovation to last — strategic copying validates a market, but the plateau that follows demands a line of differentiation prepared in advance [retour-d-experience-sur-14-applications-mobiles-et-saas-deve §x1 · 2026-08-01].
- HYPOTHESIS — to be confirmed: my hammer. The briefs give the decision frame, not my case — it gets confirmed by my own content tests, not by a hunch.
Five rules that commit, five citations. The sixth line has none — so it says so.
The citations pointing to the section shown further down are clickable: they take you to the exact line. The others point to sections this page doesn't show.
- My marketing hammer (§1) — by testing, not by hunch.
- A publishing pace I can actually hold (§2).
- The pricing of my next app — neither brief carries a scale that applies to my case: a whole HYPOTHESIS, still to build.
- iOS-first for me? — the brief justifies iOS on ROI and lower complexity [retour-d-experience-sur-la-creation-d-une-application-mobile §anomalies · 2026-08-01], my case is still to validate.
- My real daily budget (§5).
The document could have invented a price, a pace, a channel. It says it doesn't know — five times.
- Watch the revenue plateaus — the $2,000/month plateau with no innovation angle is documented as a known limit of strategic copying [retour-d-experience-sur-14-applications-mobiles-et-saas-deve §anomalies · 2026-08-01] — cited here so I don't “rediscover” it the day it shows up.
known limit — cited, not rediscovered
The brief already knows this plateau. The document cites it — it doesn't “discover” it a second time.
The audit runs after me. And it counts.
Before this document goes out, a reading tool goes back over its brackets one by one, against the real briefs.
$ verify_citations(checklist.md) 22 citations · 22 resolve · 0 malformed # audited August 5, 2026 — against the briefs as they stood that day
- Deterministic. For every
[brief §section · date], it checks that the brief exists, that the section exists, and that a line in that section really was written on that date; a fabricated citation doesn't pass. - It doesn't judge the sentence. It guarantees the reference is real; whether the sentence actually carries it is what your read-through says.
- The chain stops at the brief. The original audio is destroyed on approval — we promise you the brief and its date, never “proof by recording”.
What the verifier does, in detail
Every citation gets exactly one verdict. Anything that isn't ok is a defect to fix before
handing the document over — never something you let through.
okThe citation holds. The brief exists, the section exists, and a line in that section carries that date.unknown_subjectUnknown brief. The cited slug matches none of your briefs — fabricated citation, or renamed brief.unknown_sectionUnknown section. The brief exists, but the cited section doesn't.no_line_for_dateNo line on that date. The brief and the section exist, but nothing was written there that day. The most common case is a line reworded since: it was re-dated, so the old date no longer resolves.
The “malformed” counter is separate: it flags brackets that tried to be a citation without following its grammar. It's a ceiling to eyeball, not a proof — ordinary prose in brackets pushes it up.
Where the citation lands.
A citation doesn't point to a document an AI produced: it points to a line of a brief — a line I approved, on the date I approved it.
The brief's §reading section, as it reads — rewritten for this page without the third-party app names.
What we don't promise you.
Not infallibility
The contract makes a defect visible, not impossible. A sentence can be badly put: you spot it because its ink is missing.
The reference, not the idea
The verifier guarantees the cited line exists on that date. Whether the sentence is true of that line is your judgment.
A date, not an origin
The date is that of the line's latest wording. Rewording re-dates: the same mechanism is what teaches you that.
A tool that promises you infallibility asks you to believe it. This one shows you where to check.
The deliverable is the top floor. It starts with briefs: the app is the door.
The document on this page was written from my own briefs, then audited by
verify_citations — the reading tool your AI can call.