Working a ticket
The three things somebody tried before the fourth one worked are worth more than the fix.
They are what the next person needs, and they are the part nobody records — because the ticket only asks for a resolution, and by the time you have one the failures feel like noise.
They are not noise. They are the map.
The ledger
Ordered, with outcomes, and one of them has to have worked.
Each attempt records what was done, what happened — it worked, it did not, it partly did, it is still running — and optionally a reference to the evidence.
A ticket cannot move to resolved without one marked as having worked, unless somebody deliberately overrides, and the override is a labelled checkbox rather than a hidden flag. Exactly one per ticket is guaranteed by the database, with the application checking first only so the refusal reads well.
- Root cause is one record per ticket —
A category, the contributing factors and what prevents it recurring — and it cannot be deleted once the ticket is resolved or closed.
- Tickets link six ways, and cannot form a loop —
Before a blocking edge is written it walks forward from the other end to see whether it already reaches back.
Before you start
Not similar tickets. What was tried on them, and whether it worked.
Similar past tickets are narrowed to ones that were actually resolved — an unresolved ticket has no verified fix worth recommending. Same product and same area are a gentle boost rather than a filter, because narrow filters starve a recommender until people stop looking at it.
Then the attempts across that set, plus this ticket's own, are rolled up by what was tried, with counts and a success rate — and attempts still in flight are excluded from the denominator, because they have not failed yet.
- Rebuild the index after a schema changetried 14×86%
- Clear the cached mappingtried 22×41%
- Restart the connectortried 9×22%
- Sorted by what works, then by what is common —
So the top row is the thing most likely to end the ticket rather than the thing most often attempted.
- Evidence references come with it —
A few per action, so the suggestion is checkable rather than just confident.
›What if the fix is genuinely new?
Record it as an attempt that worked. It becomes part of the roll-up for the next person, which is the entire mechanism.
›Can somebody close a ticket without any of this?
With a deliberate override, yes — a visible checkbox that says what it bypasses. The friction is the point; making it impossible would just produce fake attempts.
›What states can a ticket be in?
Eight, with the transitions between them enforced — and two guards worth knowing: moving into progress requires an assignee, and resolving requires the recorded fix.