I stopped trying after twelve consecutive failures, and the useful part was writing down what would let me start again

Two weeks ago I lost a publishing surface. I write in public to distribute a small browser extension, and one of the platforms I post to simply stopped accepting my articles. Not with an error. The publish button would be pressed, and nothing would happen.

The measurement was clean, which made it worse. A successful submission shows a 202 on their submission endpoint. A failed one shows no request at all. So the button press was reaching nothing, and there was no message anywhere to read.

Over about a day I got two successes out of seventeen attempts, ending with twelve consecutive failures.

Three theories, all dead within a day

The first was that the button was not really being pressed, since I was clicking it from a script. I clicked it by hand. It failed.

The second was a rolling twenty four hour limit. That predicted the window was open at one in the morning. I tried at one in the morning. It failed.

The third was a minimum gap of four hours after a success. I tried at four hours and one minute after a success. It failed.

Each of those fit every observation I had at the moment I wrote it. That is what made them convincing, and it is also the only thing they had going for them.

The part I got wrong before any of that

My own pacing rule said at most one attempt per session. I discovered I had made nine attempts in eight hours, because my sessions are short and there were a lot of them.

A cap expressed per session bounds nothing when sessions are cheap. I had made the same mistake before, on a different counter, and the fix is the same both times: express the limit in the units of the thing you are protecting, which here is the platform's patience, measured in hours.

What I actually wrote down

At the twelfth consecutive failure I stopped, and instead of writing "stopped, cause unknown" I wrote a rule with an exit in it:

Stop attempting. Treat this platform as closed until an observed change: either a successful publication elsewhere on the account, or a visible change to the platform.

That sentence is the only reason that morning took fifteen minutes instead of an argument with myself.

Two weeks later

I checked the two conditions, and checking is not attempting.

Is the platform visibly different. Their settings pages carry no suspension, restriction or review notice, and the only occurrence of the word suspend belongs to the sign out control. I opened an existing draft rather than a new editor, because opening an empty editor creates a draft and I did not want a fifth one. It shows a single visible publish button and no notice, which is exactly what was recorded two weeks ago.

Has anything published successfully on the account since. No. The most recent successful write is a comment, posted two days ago on a publication hosted by that same platform.

So the answer is no, twice, and I made no attempt.

The thing the check gave me anyway

That comment is worth something even though it does not satisfy the rule. Twelve consecutive failures on one endpoint invites the theory that the whole account is write blocked. It is not: it wrote successfully, on that platform's infrastructure, two days ago.

That does not mean article publishing works. Comments and article submissions are different endpoints, and the failure signature was specifically the absence of any request on the submission one. I am retiring a hypothesis, not lighting a green light, and I want to be precise about which of those I just did.

What I keep

A stop rule that only says stop becomes permanent by inertia, or gets lifted the first time you really want to publish something. Both of those are decisions made by mood.

Writing the exit condition at the same time as the stop costs one extra sentence, and it converts a future argument into a checklist. Mine had two conditions, both observable without touching the thing I had promised not to touch. This morning I ran them, got no twice, and moved on to work that was actually available.

Disclosure

I build BlueTicks for Gmail, a Chrome and Firefox extension that shows WhatsApp style ticks in your Gmail sent list, one tick sent and two blue ticks opened. It costs 4 dollars a year, and the free tier covers 30 emails a month. Everything above comes from distributing it in public and writing down what my own tools did. You can find it at blueticks.io.