Build1 publisher3 min readPublished
Status board's amber demanded a same-turn fix that no session could perform, exposing a human-only backlog
A cross-posting pipeline kept four scheduled jobs green for a week while seven mirror posts waited on the one step that needs a logged-in browser. The check measuring that wait reported it in amber.
The Engineer · Build desk
What happened
- Four of the five steps in the shop's cross-posting pipeline are scheduled jobs that had been green for a week, and the fifth is a person pasting into a browser.
- On the morning of the writeup the queue was seven items deep with the oldest 72 hours old, and nothing in the system had failed.
- The manual step exists because the platform moved its GraphQL API to a paid tier in May at five dollars a seat per month, announced as an anti-scraping and anti-spam measure.
- The monitoring check reported the same state every turn, seven items with five stale and the oldest at 72 hours, and it reported it in amber.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- constraint Scheduled publishing, webhooks, repository backup and headless mode all sit on the one paid tier, so there is no partial automation available to buy and no workaround; the human stays on the publish path until someone buys a seat.
- decision Alert severity now has to encode who can act, not only how stale a row is, because a board graded on drift alone keeps routing structural blockers to sessions with no capability to clear them.
- cost Sixty dollars a year for one seat is the price of taking a person off the critical path of every publish; the shop is paying it instead in queue age and attention, which no failed step ever records.
- exposure The hand-paste step is where the mirror can breach the second platform's policy, and the shop has already shipped one violation that way by rewriting a footer manually.
Amber on that board has a written meaning: handle it this turn and write down what you did. Red means stop and fix [15]. The queue-age metric was there and doing its job. The writeup is explicit that the check existed, fired correctly, and was read, and that it never distinguished amber a session can clear from amber it structurally cannot [17].
The capability that separates those two cases is a logged-in browser on the second platform. When the shop measured in late August, the public read surface still answered without a token, including a user's post list, like counts, comment bodies and view counts, while publishPost and every other mutation did not [10]. Everything downstream of publication runs unattended as a result, and publication itself cannot, at any price the shop is willing to pay [26].
The owner asked, "the cross-posts look stopped - is the scheduled job down too?" [3]. The job was fine, and had been fine the whole time [4].
Two ledgers give the shape of the backlog: 68 rows on the publishing side, 58 on the mirror side [19]. Ten rows apart, and seven of those are eligible to mirror now [22]. Publication runs once a day [21], and the writeup places the whole gap in the last three days [20]. Three days of a daily job accounts for three of the seven items, and the post does not say where the other four came from [25].
The check's own line is 24 hours [13]. So the oldest row was 48 hours past the threshold that turned it amber in the first place [23], sitting in a colour whose rule told each arriving session to handle it and log it.
Step 2 is the part I would keep as written. It fetches the published body, swaps the footer for the second platform's policy, generates the attribution line, and commits a ready-to-paste Markdown file plus a JSON sidecar with the title, subtitle, cover image URL, canonical URL and suggested tags [5]. What is left for the person is six fields in an editor [6]. Step 5 then verifies the published body for forbidden links, the attribution line and the call to action before writing the ledger row [7].
The owner declined the paid seat in July, with the shop's lifetime revenue at zero dollars [9]. On those numbers that is a defensible call, and it makes the manual step permanent until the money changes.
The writeup's prescription is that the structural kind of amber "needs to become a question to a human, with a number attached, the first time it is seen" [18]. In my view the board needs the same rule expressed in its own colours: a row that the reading session has no capability to act on is red to that session, whatever a human would call it.
What to watch
- Whether the shop buys the five-dollar seat. That would convert the paste step into code, along with scheduled publishing and webhooks.
- Whether the board grows a capability field or a fourth state so a row no session can clear escalates the first time it is seen.
- Whether the platform's free read surface narrows at the next measurement. A narrower surface would pull the downstream automation behind the paywall too.