What autonomy needs
The precedent nobody passes on
Somebody tried this before and it stopped. The reason it stopped is the single cheapest thing to hand over and the most reliably lost.
Decisions this page is about, and where the line actually was
Picking the library for the new parser
1 → 4The architect had a standing view
Dropping a requirement nobody had asked about in six weeks
2 → 5A commitment made before you joined
Changing a database column type
1 → 1Small, reversible, tested once before
- 1Decide it alone and say nothing
- 2Decide it alone and mention it afterwards
- 3Check with one colleague first
- 4Check with the person who assigned it
- 5Needs a decision one level up
- 6Cannot be decided here at all
three decisions · not one of them gave you more room than you thought · two gave you less · one you had exactly right · the median gap was three steps on the scale
Three of the nine decisions counted on this site.
For distributed work, teams can review this solution to add evidence about capacity, attendance, and project effort without creating another status ritual.
Teams can compare this approach with guidance available from Training Industry.
Three weeks into the work you find a branch from eighteen months ago with the same name as yours. It is half finished. There is no note explaining why it stopped, and the person who made it left in the spring.
You now have to decide whether it stopped because the approach does not work, because the priority moved, or because the person who was doing it went on leave and nobody picked it up. Those three have completely different consequences for what you should do next, and nothing available to you distinguishes them.
Why precedent disappears faster than anything else
The other three things a handover needs survive in some form. A constraint is usually written in a contract. A decider still holds their job. A purpose can be reconstructed from the plan.
Precedent lives only in the memory of whoever watched it happen, and has three ways of vanishing:
- The person leaves, and takes the only copy.
- The attempt was never recorded, because recording a thing that did not ship feels like filing a failure.
- It is remembered but not thought worth mentioning, because the person handing over believes your approach is different enough that the old attempt is irrelevant. Sometimes they are right.
The third is the common one, and it is the hardest to fix, because it is a judgement made silently by somebody who has more information than you do.
What a useful precedent looks like
Not a post-mortem. Three sentences.
We tried this in the spring. It got as far as a working prototype and stopped because the import could not be made idempotent without a schema change nobody would approve. The person who did it is gone, but the branch is still there and the last commit message says roughly that.
That is the whole artefact. It tells you the approach, the blocking fact, and where to look. It takes a minute to say and would have saved three weeks.
Note what it does not do. It does not tell you not to try. The schema change may be approvable now; the constraint may have been invented. Precedent is information, not a prohibition, and handing it over as a prohibition is its own failure — people stop telling you things when the telling ends an argument.
The two attempts worth knowing about
Not every previous attempt matters. Two kinds do:
The one that failed for a structural reason. Something about the system, the organisation or the customer made it impossible, and that thing is probably still true.
The one that succeeded and was reverted. This is the most valuable and the least likely to be mentioned, because a revert is embarrassing. Something shipped, something went wrong, it was taken out. Whatever went wrong is waiting for you.
Attempts that stopped because of a priority change are worth one sentence and nothing more. They carry no information about the work, only about the quarter in which they happened.
If you are receiving
Ask one question at the start, in exactly these words or close to them: "has anyone tried this before, even partly?"
It is a cheap question. It does not challenge anybody and it does not imply the work is doomed. Almost everybody answers it honestly, because the answer is just a fact about the past, and the people who have been somewhere a long time generally enjoy being asked.
Then ask the follow-up that does the real work: "and why did it stop?"
The archaeology, when nobody tells you
Sometimes nobody remembers and there is no one to ask. Three places the evidence survives anyway, in rough order of usefulness.
The abandoned branch. Search the repository for the nouns in your task. A branch with a stale date and a half-finished commit history is somebody's attempt, and the last commit message is frequently a one-line explanation of why they stopped.
The reverted change. Look for things that shipped and were taken out. A revert carries a reason in its message surprisingly often, and whatever caused it is usually still there waiting.
The document that trails off. Design documents that stop halfway through are the most reliable sign of an attempt that was abandoned rather than completed. The point at which the writing stops is usually the point at which the author hit the problem.
None of this is as good as being told. It takes an hour rather than a minute and it produces an inference rather than a fact. But it is entirely within your control, it can be done on day one before anybody has to answer a message, and on the record counted on this site it would have caught the dead prototype that surfaced in week three of the reporting page work.
If you are handing over
Write down the attempts as they happen, in one line each, where the next person will look. Not a document — a line in the place the work lives.
The whole value is in the sentence "it stopped because", and the whole cost of not writing it is paid by somebody who has not arrived yet. On the record counted on this site, two of the six pieces of work that came back went back because of something a previous attempt had already discovered.