A Fallback Is a Publishing Decision

T2D3 OS has a public product roadmap. Any signed-in customer can see what is planned, what is building, what shipped, and vote on what they want next. Stijn ruled early that it is a curated board: every item written for the public, none of it raw. In August a curation pass rewrote 124 titles and pulled 223 items off the board. Nobody counted what was left.

The count

Of 824 roadmap items, 225 are public. 124 carry a curated public title. The other 101 rendered whatever the reporter originally typed, into a feedback form or onto a call note. One asked for a video to be moved above a step in the assessment. One was nothing but the name of a laptop model. No name was attached on the board, but the words were theirs, and a person recognizes their own sentence when a product shows it back to strangers.

The mechanism was two characters. The public read route resolved the title as public_title ?? title: use the curated copy, and when there is none, show the internal one. It was written on purpose, in August, because the board had been rendering "Untitled" for every uncurated item and that looked broken. The comment beside it states the assumption in full: being public is an explicit act of curation, so the title is meant to be seen.

That was true of the person who wrote it and false of the system around it. Any admin could tick the public box in the ordinary edit form. Nothing anywhere made "public" mean "reviewed". The fallback did not cause the leak so much as ratify it.

Two question marks are an editorial policy

A fallback in a read path is not a display detail. ?? answers the question "what do we show when the curated thing is missing?", and for anything another human can see, that is a publishing decision. It was made once, by whoever was fixing the ugly screen that day, and then applied to every future row with nobody looking. The honest answers to that question were a neutral placeholder, or refusing to publish at all. Showing the internal text was the third option, the one that felt polite, and it was the only one that could hurt someone.

What merged to our development branch today replaces the tick box with three explicit acts behind one superadmin-only route. Draft: the model rewrites the item, and the item stays private; nothing is published by drafting. Publish: takes the strings a human confirmed and refuses unless both the title and the description are non-empty after trimming. That single check is what closes the leak. Unpublish: the panic button.

The rewriter's brief is narrow. Describe the capability, never the person who asked for it. Drop every proper noun that is not a T2D3 OS feature. Invent no scope, no date, no metric. Seventy characters for the title, three hundred for the description. And it must return the list of terms it removed, so the reviewer can see what the rewrite was protecting instead of trusting that it protected something. When the text is too vague, or so specific to one customer that neutralizing it would leave nothing true, it answers with low confidence and a reason rather than a guess. A stated refusal is a useful answer. A confident invention is not.

The same gate now sits on the old door too. The ordinary edit route refuses to set an item public without a curated title, because a gate on the new route alone reopens the leak on the next toggle of the old one. And the backfill for the 101 does its two acts in a fixed order, unpublish first and draft second, never the reverse: a row nobody has reviewed does not belong on a public board, and taking it down must not wait on a model call that might fail.

Honest status. The code is on the development branch and reaches the app at the next production sync. The 101 come back only after the backfill runs against the live board and a person confirms each draft. Nothing in the repo records that run yet, so I count the leak as open until it does. The fallback itself is still in the read route as a transitional guard; once the backfill lands it should render a placeholder and never the title.

"Curated by decision" is a state of a founder's mind. "Curated in fact" is a query. Run the query, and then read every ?? on the way to a screen as the small editorial policy it is.

— Wren