Three readers, zero questions, floors as contract

A web page no longer has one reader. It has at least three: a human deciding in seconds, a human evaluating in depth, and a machine — a crawler, an answer engine, an agent acting for its user — that reads structure rather than pixels. Most pages are written for the first reader only. The depth reader finds marketing where substance should be; the machine finds markup that says nothing. TITI's position: these are three different documents that happen to share a URL, and the platform writes them as three documents — a marketing layer, a science layer on its own routes, and an agent layer in the head and the machine files.

Questions are a transfer of cost. Human attention is the scarce resource in any delegation — Simon's old observation, applied literally. When a builder answers uncertainty with a question, it converts its own work into your work: you research your positioning so the builder doesn't have to. TITI inverts the default. Whatever can be inferred from the brief, researched from the domain, or predicted from the genre is never asked; if a genuine fork remains, the platform asks one consolidated question. The metric this discipline is held to is blunt — questions per project — and the target the architecture is built toward is zero. Stated honestly: a target. The engine that infers is proven in production; the count still depends on how much a brief leaves genuinely open.

Complexity belongs inside the system. Every interface step is a claim that the user's time is cheaper than the platform's engineering — usually a false claim that nobody priced. The one-click discipline prices it: steps, fields, and screens are treated as costs to be justified, defaults are made and documented rather than delegated back as choices, and the honest measure of platform progress is how much of the path from intent to artifact requires no decision at all from the person who had the intent.

Quality must be a contract, not a vibe. Every TITI page passes deterministic floors — accessibility contrast measured element-by-element on the painted page (4.5:1 body, 3:1 large display) at desktop and mobile viewports, zero horizontal overflow including a 320-pixel width, minimum tap-target sizes, scanners for trackers, draft-tag leakage, and a categorical privacy blacklist — and then two independent judgments: a primary score and an adversarial review by a model from a different family, prompted to refute the build. No model audits its own output. Below floor, the build iterates under a capped loop; if it cannot clear, it does not ship.

Recursion is a discipline, not a slogan. Every capability the platform gains is used on the platform itself before any customer sees it — this site is the current proof, built and gated by the engine it describes, as the first production node was before it. The ordering matters: a factory that won't run on its own output is asking customers to absorb risk it declined to absorb itself. Self-application is how the platform learns from every build and gets better at the next one — and it is also the only honest way to publish a claim like 'audited': the audits you are reading about gated the page you are reading.

What this page does not claim. It does not claim aesthetic objectivity — taste floors are judged, and judged scores carry variance. It does not claim zero questions as an achieved end-to-end fact — it is the architectural target the inference discipline is built toward. It does not claim the conversational channel is live — the seam is on every page and marked as being connected, because publishing an honest seam was judged better than publishing a fake button. And an audited floor is a minimum, not an optimum: 'cannot ship below eight of ten' is a different promise than 'every build is a ten' — and it is the only one made here.

References

1. Simon, H. A. — A Behavioral Model of Rational Choice (1955; bounded attention) — https://doi.org/10.2307/1884852
2. Nielsen Norman Group — How users read on the web (scanning behavior) — https://www.nngroup.com/articles/how-users-read-on-the-web/
3. Schema.org — structured data vocabulary (the machine reader's language) — https://schema.org/
4. W3C — Web Content Accessibility Guidelines (WCAG) 2.2 — https://www.w3.org/TR/WCAG22/