Rebuilding something? Run the new version beside the old one. Let the old one decide what's true, log every difference, and count the ones you can't explain. That count is your real progress.
Part 5, the last in the series: https://t.co/qPY1moJOU1
Run it yourself with one command (Python 3.11+ and pytest): components/middle-path/run-suite
Six small suites, 200 tests, ending in middle-path rc=0. 2/2
https://t.co/7Jbf2VrAjl
The middle-path repo now has its first code: a walking skeleton. A story goes from packet to merge decision through every layer. GitHub and the agents are faked, so it shows the shape, not any model's behaviour. 1/2
A green test only shows the code agrees with the fake it ran against. Four cases from one week, including a rule that called 26 of 27 healthy hours unhealthy, and a count from 4,129 review findings. Part 4: https://t.co/ANvuFbO1aF
A hard-to-write test is information: the code's shape is wrong. Tight seams, components, and a merge rule took our average test wait per change from 27–29 min to about 9. Seven of ten changes waited under 3 seconds.
https://t.co/fNdrrmw8bA
What is a local model actually for? I tried to make a 27B model on a gaming PC into a code reviewer. It isn't one: it caught about 13% of real blockers. It's good at checking a claim you hand it, not finding one. Part 2: https://t.co/AZ25TfnezJ
@unclebobmartin The politics of your posts degrades the impact you’ve made in the industry. Bucketing all institutions together when each is unique with their own strengths and challenges is unreasonable. But even if we do, there are many places w broad political views.
@heysarahcarter @WillowCreekCC Not surprised. What your family went through to stand up against that system had to be extremely hard. I sincerely hope you and @steveryancarter understand many people deeply respect your sacrifices and still miss your wisdom and impact.
In toxic cultures, people get promoted for results even if they destroy relationships. Abuse is a price to pay for high performance.
In healthy cultures, no level of individual excellence justifies undermining people. You’re not a high performer if you don't elevate others.
In Product Teams, outsourcing should be viewed as a cost, not a saving.
That isn't to say it's not 'the right' approach for the org (everyone has constraints) but it's almost certainly not the optimal way to create long-term value.
I’ve heard some people in big organizations say they want a 100% say/do ratio each sprint on their agile teams. I’d argue we should be focusing on sustainable delivery of the most value first — even if that value comes in after the sprint is started.