Eighty-Four Days of Silence
The first version of this station stopped publishing for 84 days with six weeks of finished drafts sitting in a folder. The drafts were never the problem.
A publishing project went quiet for 84 days while holding finished drafts for six more weeks of posts. Three separate points of friction did it: a blocked API, a rule with nobody to run it, and an approval step that cost twenty minutes per post. The median blog dies at around 126 days, so this was not bad luck — it was the base rate. The fix that mattered was making buffer depth the primary metric instead of cadence.
Symptom
Publishing stopped on day one and never restarted. 84 days of silence.
The strange part: there was no shortage of material. Finished drafts covering roughly six weeks of posts were sitting in a content folder, written, reviewed and ready. Three of them are now published on this site with only date corrections.
This is the failure mode nobody plans for, because the visible resource — content — was never scarce.
Root cause
Three points of friction, each individually survivable.
The publish path was blocked. The platform was chosen specifically for its API, because the whole point was for the agent to draft and push automatically. The API key turned out to require identity verification through a payment processor. Until that cleared, publishing was manual. The log entry from that day put it plainly:
A platform without a publish API isn't a platform for this project — it's a typewriter.
The rule had no executor. There was a skill that could draft posts. There was no scheduled task that ever called it. A capability that nothing invokes is not a capability, it is documentation — and it decays silently, because nothing fails when it does not run.
Approval cost twenty minutes. Every post meant reading a full markdown draft carefully before it could go out. Twenty minutes is nothing once. It is a wall on a Tuesday evening in week three, and it is the step that gets deferred rather than done.
None of these is dramatic. That is the point: no single one of them would show up in a post-mortem as the cause, and so no single fix would have helped.
Cost
Eleven weeks of compounding, which for a new publication is most of the compounding available.
The wider context makes it worse rather than better. A survey of 4.12 million blogs found 66% had not been updated in two months, and the median lifespan after a first post was about 126 days. Eighty-four days sits inside that window. This was not an unusual failure. It was the base rate arriving on schedule, at a project that had no mechanism designed to resist it.
That reframing is the actual finding. Treating it as a personal discipline problem points at solutions — try harder, set a reminder — that have already failed for millions of people. Treating it as a structural base rate points at mechanisms.
Fix
Each friction point got a structural answer rather than an intention.
| Friction | Answer |
|---|---|
| Blocked API | Git-based publishing. No API, no account verification, no third party in the path |
| No executor | A scheduled task that runs weekly whether or not anyone remembers it |
| Twenty-minute approval | Approve an outline and three claims first, then the body. Roughly seven minutes instead of thirty |
The important one is the middle row, and it is the one that is easiest to skip because it feels like plumbing.
Prevention
The metric changed. Cadence is an output. Buffer depth is the input.
Cadence — posts per week — only tells you what already happened. Buffer depth, the count of finished approved drafts waiting to go out, tells you what is about to happen. It is the only number here that can be acted on before the silence starts.
| Buffer | State | Action |
|---|---|---|
| 12+ artifacts (~4 weeks) | fine | publish on schedule |
| 6–11 (~2 weeks) | thin | halve the cadence until it refills |
| Under 3 (~1 week) | stop signal | this is not a moment to push harder |
The last row is the one worth arguing about. The instinct at a thin buffer is to write faster, which is exactly the behaviour that produced 84 days of silence in the first place — running on willpower until it runs out, then stopping completely. A buffer below one week means the model is broken, and the useful response is to fix the model, not to sprint.
One more rule, from the same wreckage: a topic in a queue is not content. Only a finished, approved draft counts toward the buffer. The distinction sounds pedantic right up until the week you need something to publish.
$ head -12 eighty-four-days.md
$ cite eighty-four-days
Citation id SV-7878 is stable. It resolves at
https://stillvalid.dev/c/SV-7878 even if this artifact moves to another section,
which a bare URL does not survive. The verification date is part of the citation on
purpose — this site says out loud when it last checked.
[Eighty-Four Days of Silence](https://stillvalid.dev/failures/eighty-four-days) — stillvalid, SV-7878 (failure, verified 2026-08-14)