Finalit, rebuilt
Sharper judgment. Deeper memory. Closer to prediction.
Ejder Demir
A whole new interface, and a much harder engine behind it.
Finalit has been rebuilt from the surface down. Everything that used to sit between you and an answer is gone: no panels to open, no mode to choose, no list to work through before you find out what the thing you wrote is going to do.
One draft. One press. One position. That is the whole interface now.
Post-mortem: the Q4 launch slip
This document sets out a summary of recent developments affecting the Q4 launch and provides some background on the wider circumstances, with a view to capturing learnings for future planning cycles.
Timeline. The integration work depended on the partner API landing in July. Their release moved on 14 June and again on 2 July. The second move consumed the two weeks of contingency in the plan, and the plan had no second fallback.
I knew on 3 July and told the sponsor on the 14th, because I wanted a plan to put next to it. The eleven days cost more than the plan was worth.
Unfortunately some resourcing challenges were encountered and certain dependencies did not materialise in the timeframe that had originally been anticipated, which has had a knock-on effect on the wider schedule.
Unaffected. The data migration, the permissions rebuild and the finance reporting views are complete and in staging. None of them were on the partner path and none of them moved.
Remaining. The partner integration and one round of load testing behind it. Both are now built against a fallback that does not depend on a third-party release date.
Cost. Two weeks of delivery, one sponsor conversation that should have happened eleven days earlier, and a fortnight of the team working to a date they had privately stopped believing in.
Going forward we will endeavour to strengthen our planning processes and improve communication so that issues of this nature can be identified and escalated at an earlier stage in future.
Simpler on top
The old version asked you to drive it. Pick a Lens, set a mode, open a panel, read twenty flags and decide for yourself which two mattered. That is a judgment engine handing the judgment back to you at the last moment.
Everything it used to make you configure, it now works out for itself and tells you afterwards, so you can correct it if it read the draft wrong. Nothing stands between pressing the button and being told where you stand.
Much harder underneath
Taking the controls off the surface bought room behind it. The engine now works out the standard your particular draft has to meet instead of marking everything against one rubric, which is why a dissertation chapter and a board paper are never held to the same bar. It reads the piece in separate passes that never look for the same thing twice, and ends on one sentence about what its reader will do.
It is also far less forgiving. Drafts the old version would have called clean now come back refused, with the one thing that would most change the outcome named first. A verdict that is easy to earn is not worth having.
And a new object: the desk
A desk is context you save once and use again. The background to a body of work: who reads it, what they already know, the constraints that never move. Write against a desk and every draft there is judged with all of it, instead of being read cold.
Desks are publishable. Use one Finalit built, use one somebody else made, or publish yours for other people to write against. The desk is what gets shared. The work done with it is not.
And it keeps score
Every judgment is now written down as a call: what Finalit said, what you did about it, and how the work actually landed. A record is what separates judgment from opinion, and it is the half of this rebuild we expect to be held to.
The question is derived from the draft, exactly as the verdict is. An introduction is asked whether it was made. A funding bid is asked whether it was funded. Marked work is asked what it got, on the scale that writer is actually marked in. There is no fixed set of things people write, so there is no fixed set of outcomes to keep.
It already tells you what it will cause
A verdict is not a grade on your prose. It is a call about the person who receives it: that a hiring manager reads to the end and still cannot tell what you want, that a marker finds the argument but never the evidence for it, that a board approves it and changes nothing. The call is made before the work leaves your hands, which is the only moment it is worth anything.
What the record adds is proof. Every call is checked against what actually happened, so how often Finalit is right is a figure you can look at rather than a claim we make. The judgment was always a prediction. Now it has to answer for one.
It is live now.
Open Finalit, paste whatever you were about to send or submit, and press once. The product page is the long version, if you want to see the engine work end to end first.