ENG-812 says “In Progress”. The code merged four days ago.
Nobody typed anything wrong. The work finished and the board did not hear about it.
This is the whole product in one ticket, running on demonstration data, with no signup.
driftENG-812 · Reject checkout when the basket is empty
wrong for 4d 02h
Mon 09:16The code side. PR #1204 merged to main, checks green, one approval.
Mon 09:16GroundTruth saw it and matched it to ENG-812 by the ticket key in the branch.
nowThe tracker still says In Progress. Every report built from it since Monday has been wrong, including the one that went to your leadership.
What GroundTruth would do · move ENG-812 to Done, because three of three acceptance criteria are met by the merged diff
Mode
Shadow mode. GroundTruth reads everything and writes nothing. Press Apply and you will see
exactly what it would have changed. This is how every install starts, for as long as you like.
2/5 acceptance criteria still unchecked though work looks complete
85%
flag only
ENG-820
Needs identity API from INIT-4 (PR #1177, not merged) - Checkout blocked on Identity
80%
flag only
ENG-777
In Review but PR merged 4d ago
85%
In Review → Done
#1206
PR #1206 is not linked to any ticket (untracked work)
80%
flag only
ENG-731
In Progress with no PR and no update for 14d
70%
flag only
EPIC-51
All children of EPIC-51 are done but EPIC-51 is "In Progress"
80%
In Progress → Done
stop
And one it will not touch.
Ticket assignee "Sam Okoro" is not a PR author (Priya Rao). Confidence 50%, which is below the threshold,
so nothing is written and nothing is guessed. It goes to a person with the evidence attached.
Most tools in this category would flip the ticket anyway and let you find out later.
You set the threshold. Above it, GroundTruth acts. Below it, a human does.
The record, as you make it
Every automated change appends here, carrying the finding behind it, the confidence, the mode, and a hash of
the entry before it. Change an old entry and every later one stops matching.
Nothing written yet. Press Apply on anything above.
Now do it on your own repository
Everything above runs on demonstration data. This does not. Give us a public GitHub repository and we will
read its merged pull requests and tell you how many of them say which ticket they were for.
Public repositories only, and we read the pull request list and nothing else. We can see whether your code
says which ticket it was for. We cannot see your tracker, so we will not pretend to know whether those
tickets are right.
What it looks like once it is running
Demonstration data from a fictional team of 22 developers, not customer results. We publish
no customer numbers we cannot name.
6%
Jira drift
↓ from 41% (14d)
8m
Time-to-accurate
99.2% within SLA
99.8%
Write precision
audit-chain provable
87%
Auto-resolved
no human touch
4.2d
Blocker lead time
earlier than sprint review
143
Chases eliminated
this month
Drift over time
Percentage of tickets out of sync with GitHub. Baseline 41% → 6% today.