This is the desk as a console, gossip no. 031, in the hand ruled on 2026-08-28. The seven screens are the ones the second edition carries. What changed is the room they sit in: a tab edge you can move through in any order, receipts set in the margin against the row they belong to, a ledger along the bottom that counts what is filed, and a keyboard you can drive the whole intake from.
Screen 7a asks for a handle per platform and marks each one as yours or as a room you watch. On leaving the field the page calls the platform's public endpoint where the census records one live, and prints what came back. Where no endpoint answers, the row prints unverified and says which of the two reasons applies. No login is requested for anything, and nothing is written anywhere.
7aYour accounts. Handle rows, the ours or a room we watch mark, live verification against two keyless public endpoints, the anchor tally.7bYour known metrics. Each row lands as one evidence item on the reported rung, which the template gained today. A row without a date or a source is not filed.7cYour files. Listed in the trail as examined or abstained, so a file that went unread is visible rather than absent. 20 MB a file.b2The two changed screens. Messages become the tester's preloaded phrases; how you will know it worked becomes the falsification row, on the open rung.031The console. The tab edge with a state mark per screen, receipts as marginalia at desk width, the ledger, the keyboard map, the paper grain, and the opening. The receipts themselves are unchanged.laterOAuth, the Supabase columns, the private Storage bucket. Deferred on purpose. Handle plus public read gets most of the value without a registered app per platform.One row per handle. Your own accounts become the anchor set of the read. The rooms you watch become the harvest scope. Leave a handle field to verify it.
Numbers you hold: from your dashboard, from a platform export, or as your own estimate. Every row lands in the read on the reported rung, which decode-fieldnote-3d/1.2 added for exactly this: filed by you, dated, and not observed by us. We print the number, the source and the date, and we say we did not check it.
No login is requested for anything. A number without a date or a source is not filed, because the rung is what carries the date and the source.
A deck, a prior study, an analytics export. Every file you attach is listed in section 8 of the read as part of the trail, marked examined or abstained, so a file that went unread is visible rather than absent.
This screen used to collect messages and stop. It now feeds the MCS tester directly: every line filed here becomes a preloaded phrase in your report, so the first thing you see in the tester is your own line, scored, with the rule it met or broke and the source of that rule.
Nothing is scored here. The scorer reads the crowd rule sets an analyst writes, and those do not exist until the read does.
This screen used to ask how you will know it worked. It now writes the falsification row of your read, on the open rung: the test that would prove the read wrong, defined and not yet run. A read whose every crowd has one is a read that can be checked against the world later.
This build has no schema behind it. Every row on the five screens lives in this tab only. The Supabase columns, the two RPCs, the private bucket and the export into a DECODE_DATA file follow once the screens are approved.
1desk submission lands in the intake table2the compiler writes one data file per engagement3the validator exits 0 or the report does not mount4the field serves the field, the report, the codebook and the testerThe surface enum widens to fourteen. It was twelve (tiktok, ig, yt, sp, pod, irl, x, discord, reddit, li, sub, wa) and none of the twelve has a keyless public endpoint this page can call: Reddit returns 403 with no cross-origin header, YouTube needs a key, and the rest have no public read. So the enum as ruled contained no surface a handle could be checked against, and every row on this screen would have printed unverified, taken as filed.
Mastodon and Bluesky are the two lanes the census records live and keyless, which makes them the only two this screen can verify by observation. They are in the enum now, and a read may select them among its six. Landed in decode-fieldnote-3d/1.2 with the nine-slot palette and the select-six rule.
What that costs: two platforms are in the vocabulary because they are legible to us, not because they are where the most audiences are. The gap between those two facts is the reason a read selects six rather than being handed them.
The desk is the intake for an Audience Decode: the seven screens collect what a business already knows about its own audience, and each row files as one evidence item on a named rung of the measurement ladder. The instrument the rows feed is recorded at /decode/, its arithmetic at /decode/method/, and a worked read at /decode/field/.
collectsHandles, metrics, files, phrases, one falsification row. Five of the seven screens collect; the opening and the handoff do not.verifiesTwo lanes, by observation. Mastodon and Bluesky answer a keyless public read, so a handle on either is checked live. Every other surface prints unverified, taken as filed at the same size and in the same voice.rungsNothing is promoted. A supplied figure files as reported, dated and unchecked by the house. A falsification row files as open. Neither becomes measured because it was typed in.refusesA metric with no date or no source is not filed. The refusal renders like a success, so a reader sees what was declined rather than an absence.writesNothing. This build has no schema behind it. Every row lives in the tab and leaves with it. The specimen band prints schema writes: 0 for that reason.not thisThe crowds, the rules and the weights. They stay with the analyst. No screen fills them in, and no intake produces a read on its own.