Every failure has a lever
A platform refusing a post is a normal day's work: a token expires, an account needs a check, a service times out, a pay-per-use balance runs out. What matters is that nothing is lost, that every attempt is on the record, and that whatever needs a person is in front of them with the button that fixes it — in the same place the problem is shown.
any platform, any project: a token, a checkpoint, a timeout, missing credits
every attempt, append-only, one line each; 16 reasons; read every five minutes
Needs you · Open now · History · By day · Reasons; a CSV for the records
reconnect · check now · refresh · retry one or a group, spaced like a person
a direct link (X credits, a Meta check) and a 'done — retry now' button
The ledger#
Since 13 November every failed publishing attempt, in every project and on every platform, is written to an append-only ledger: one line per attempt, with the platform's own words, the account, the episode, a reason from a fixed list of sixteen, and how it ended. Backfilled from the records, the history goes back to 25 October: 1,029 failed attempts on 521 posts, of which 490 are resolved. 721 of the attempts are X waiting for its API credits; of the rest, the biggest kind is Instagram saying it could not process a video, which led to the fetch-first upload described in the previous chapter.
| Reason | Failed attempts | What it means |
|---|---|---|
| X credits | 721 | X's pay-per-use balance is used up: posts wait, nothing is lost |
| Media processing | 119 | the platform could not process the video; most go out on a retry |
| Account unlinked | 81 | an Instagram account was unlinked from its Page: the owner relinks it |
| Timeout or network | 36 | the platform did not answer in time: retried with backoff |
| Meta checkpoint | 10 | Meta asks the account's owner to confirm it in the app |
| Other | 62 | tokens, rate limits, posts that may already be live |
The page, and the levers#
The portal's Failures page reads the ledger every five minutes and shows it five ways: what needs you (grouped by account), what is open now, the full history with filters and a CSV, by day and by reason. Every row and every group has a lever: Reconnect the account through its sign-in, Check now, refresh a Meta or LinkedIn token, or retry one post or a whole group — spaced like a person would, stopped at the platforms' daily caps. Where the fix can only happen elsewhere — topping up X, confirming an account in Meta's app — the lever is a direct link and a 'done, retry now' button. The morning report e-mailed at 08:00 has a section on the last day's failures.
ChannelForge has the same, for every workspace: a Failures tab under Publishing for its members, and an admin page for the operator, with the same levers and the same rule — a group retry is spaced, capped and never posts in a burst.