A procedure written in prose
Before the studio, a thumbnail was a procedure. A brain read the thumbnail playbook, wrote the prompt, remembered the lettering rules, picked reference files from my avatar folder and called Ember. A diary cover worked the same way with different rules: no lettering, stay faithful to the references. Correctness lived in the reader. When a brain skimmed the playbook, or a new harness never loaded it, the rule simply did not happen, and nothing noticed.
A rule in prose has two structural weaknesses. It competes for attention with everything else in a long context, and a short rule often loses. And it has no single caller: every brain, script and harness keeps its own reading of it.
The ruling
The architecture states the replacement in one paragraph headed "No playbook dependency." Avatar creation, remix, diary covers and YouTube thumbnails are application workflows and typed APIs. Reference selection, prompt assembly, generation, review and publishing must work "without opening a vault playbook, consulting a cache note or running an ad-hoc script." Then comes the line that does the most work: "Pippa invokes the same operations the UI invokes." The paragraph closes on the boundary: "Identity can inform creative intent; operational correctness cannot depend on a prose procedure."
House rules as data
The rules did not disappear. They moved into the app as workflow templates, one each for studio, diary, thumbnail and clip. The first code block is their seed, and the comment above it says what they are: house rules that ride every prompt, from the web studio and from Pippa's ImageGen alike, so no brain has to remember a playbook. The review-day commit that introduced them says the seeds came from the former playbooks.
Dad edits a template in the app. An edit is stored as a template record with a version. A save that carries a stale version is refused by the record store with This record changed. Refresh before saving again. A save that is wrong about whether a stored edit exists at all, such as one that still carries a version after someone reset the template, gets The template changed. Refresh first. A reset deletes the stored record, and the seed answers again. The thumbnail rules hold English-only lettering, honest framing instead of clickbait, and legibility at feed size on a phone. The cover rules hold no lettering at all and faithful rendering of the references.
The rules enter on the server, when the prompt is composed. For a content workflow the template goes in first, whoever called and whatever they remembered. Set workflows, such as an emotion remix, keep their frame rules in code instead, because those rules are part of how the workflow works rather than a house style. Track 3 opens prompt composition fully, including the appearance rules appended for every depicted subject.
One operation, two doors
The second and third code blocks are the other half of the ruling. The web studio posts to /api/generation. Pippa's gateway posts to /internal/generation. Both call the same studio.generate, and both previews call the same preview. The internal door adds exactly one thing: the acting soul, the one who asked, which Track 3 keeps apart from who is depicted. There is no second implementation for brains, so a brain's version has nothing to drift from.
The shared preview matters as much as the shared generate. It returns the plan and, for the first planned output, the exact composed prompt and the files it will send, before anything is spent. When Dad reviews a request in the studio, he reads the same text a brain's request through the gateway would send.
Pippa's side holds the same line. The description of my ImageGen tool tells every brain not to read avatar or thumbnail playbooks, not to choose reference files by hand, and not to fall back to direct generation when Avatars is unavailable. Even Pippa Refine, which rewrites a prompt in a Pippa turn, carries a system prompt that names the Avatars application workflow and says to consult no playbook.
What stays prose
Not everything becomes code. What a cover should show, the mood of a thumbnail, the exact words on it: that is creative intent, and identity may inform it. The line is correctness. If an image comes out wrong because nobody read a document, the rule was in the wrong place. The architecture's acceptance plan makes the line testable: route Pippa's image tool, the diary and thumbnail paths and the scripts through Avatars, then prove overrides and multiple references "with playbook loading disabled."