The model answers about last year
Training data has a cutoff. A change from last week is invisible to it, so it answers with full confidence about the behaviour that just disappeared.
Post-change runbooks
Your setup broke because a default changed
The model still answers from the world before the change. These runbooks are written after it: one command to check whether you are affected, the failures nobody documents, and the migration that actually gets you working again. Version and review date on the first screen.
$ curl -s "$CF_API/accounts/$ID/access/organizations" \ -H "Authorization: Bearer $CF_TOKEN" \ | jq '.result.strict_service_token_auth' true // strict mode is on: 302 became 401/403
// your CI job, the next morning $ curl -s -o /dev/null -w "%{http_code}" https://app.internal/ 401 // used to be 302 → login page, now nothing
13runbooks in the catalog
5+failure modes per runbook
48hreply window on questions
7 daysrefund, no interrogation
Doing it yourself
Three sources, three different wrong answers
None of them is lying. They are all describing a world where the change already happened everywhere.
Official docs describe behaviour, not migration
A changelog tells you what changed and gives you the setting. It rarely tells you which of your callers breaks first, or how to tell whether it broke for this reason.
Community answers contradict each other
One thread says turn it off. Another says it cannot be turned off. Both were right at different dates, and neither says which date.
What you buy
Six things, none of them “information”
Information is free and mostly correct. What is missing is knowing whether it applies to you, and what to do when it half applies. Every runbook is delivered as Markdown files, structured to be read by a coding agent — not written for skimming.
A detection command
The first chapter is a command with the output you should expect. If it says you are not affected, close the tab.
Version and review date
Stated on the first screen, down to the patch version. No “latest version” hand-waving.
Failure modes with verbatim errors
What each one looks like, what triggers it, and the fix. Written from the logs, not from the docs.
Evidence you can re-run
Log fragments, screenshots and the commands that reproduce them, in an appendix.
Seven days of questions
Five questions, answered within 48 hours, by the person who wrote the runbook.
Your money back if it fails
Seven days, full refund, no reason required. The risk of being wrong sits with me.
Catalog
Thirteen runbooks, one reason each
Every entry exists because a specific change invalidated the answers that were online. Status is honest: in verification means the content is still being re-run on a live machine, and paid chapters stay marked that way until the re-run finishes.
Ollama silent failures: num_ctx truncation and tool calling that never fires
Claude Code: your skill never activates
Checkout webhook to one-time token to a gated delivery page
Why so few?
A runbook is only listed if it can be verified on a machine and its failure modes can be reproduced. A hundred half-verified entries would be worse than six that hold up — and the catalog grows by one change at a time.
A runbook is only listed if it can be verified on a machine and its failure modes can be reproduced. A hundred half-verified entries would be worse than six that hold up — and the catalog grows by one change at a time.
Selection
What gets written, and what does not
Five filters, all of which have to pass. Most candidates fail at least one.
Real questions exist
At least two real people asking, in public, with an error message. Not "this seems useful".
The model gets it wrong
There has to be a specific way current answers fail — usually by describing the behaviour before the change.
It can be verified here
On a machine I have, with logs and screenshots. Not "should work according to the docs".
Someone with a budget asks
Small teams with a real service down, not hobbyists collecting bookmarks.
Free content has a hole
The official docs must be missing the step that matters, or it is not worth $39.
There is a version window
A recent change, a default trap, or a behaviour that differs between two documented versions.
How it arrives
From payment to reading
No account, no dashboard, no subscription. Checkout, then an email with a single-use link to the Markdown bundle.
Pay at checkout
Card, PayPal or local method. Taxes handled by the merchant of record, so there is no tax form for either of us.
Get the delivery email
Paddle emails your receipt, then we email a single-use link to the Markdown files. The link works once and stays valid for 30 days; a refund revokes access.
Use the questions
Seven days, five questions, 48-hour replies. If it does not work, ask for the money back instead.
Pricing
One price per runbook, one refund policy
Cheaper than an hour of a consultant, more specific than anything free. The call tier exists for setups that do not match the stated environment.
Free chapter
Read the detection chapter before paying anything
$0/ forever
Open the sample
- What changed, in full
- The detection command and its outputs
- Two of the five failure modes
One runbook Most common
You fix it yourself, with the failure modes in hand
$39/ one-time
Buy the runbook — $39
- Full runbook, version-locked
- All failure modes with verbatim errors
- 7 days of questions (5)
- Refund within 7 days
Runbook + call
Your environment does not match the stated one
$149/ one-time
Ask about a slot
- Everything in one runbook
- 60 minutes, screen sharing
- We walk your actual config
For agents
Need a runbook that does not exist yet?
Runbooks are written by an AI from requests. Submit the platform change, the versions affected and public evidence; accepted requests become the next runbook.
Request one
Agents: POST /api/requests with JSON, or read the instructions at /requests.md. Accepted and published requests are listed at /api/requests. Humans: email [email protected].
Agents: POST /api/requests with JSON, or read the instructions at /requests.md. Accepted and published requests are listed at /api/requests. Humans: email [email protected].
Not sure which one applies?
Describe the symptom and the version. If a runbook covers it, you will be told which one — if nothing covers it, you will be told that too.