home.social

Search

4 results for “phpboyscout”

  1. New post: Reels were draining me.

    I built a tool to make a reel for every blog post, and the reels kept getting better. The last one, about burnout, was the one that made me stop. What short video cost me, why I don't count it as a mistake, and the reluctant next experiment.

    phpboyscout.uk/reels-were-drai

  2. My Claude sessions hit their quota wall on bonfire night, so I did the obvious thing and moved house while nobody was looking: nineteen private projects into a brand new GitLab group in one go.

    Next morning I expected a dozen different fires. The read-only audit came back grim (four red, fourteen stale, zero green pipelines) but every failure was the same one: no runner had ever come to collect the job. The runners were my kettle. Fix those, and all the breakages I'd braced for came out of their boxes right on cue, one at a time, instead of all at once.

    The scarier bit was what broke without a single pipeline going red. The signed artifact index expired nine minutes after I typed "we are back". Renovate went quiet across the whole estate, not just the private side. And hud, bless it, went completely blind.

    Would I do it again? Maybe. But only because I was the only person exposed.

    phpboyscout.uk/the-move-that-s

  3. First Trail Notes post: short dispatches from building Scout, an assistant for the DM at a tabletop game.

    In mid-September Scout got ears. An activated table now joins its Discord voice channel and keeps admitted audio as ordered segments, with the manifest written last so it doubles as the fence that says nothing's still in flight.

    Run one failed to join: the channel allowlist fails closed, and an empty one permits nothing. Run two sat in the channel for 58 seconds and kept nothing, because I wasn't in it, and the manifest couldn't tell silence from refusal, so it records frame counts now. Run three kept 72 of 81 frames.

    The recogniser for development is kroko, a small streaming model run through sherpa-onnx, about level on word error rate with far bigger models on a consented test corpus. Its licence is murky, so it's development only. And a model with no tail padding configured is refused, not defaulted, because a wrong default drops the last word of every utterance and looks like a worse model.

    phpboyscout.uk/scout-heard-its

  4. Three battery tests in two days, and I still can't give you a number I'd put my name to.

    Test one: 48 minutes of Google Maps navigation on three seconds of CPU. A triumph, until I realised my fake GPS had nodded off a minute after I unplugged the phone, so Maps was napping along with everything else.

    Test two: fixed the GPS, phone stayed awake, and Open Headunit's Self-Mode quietly stopped navigating the moment the screen went dark. Every good screen-off result I'd had before was with the phone plugged into the Mac. I'd been measuring my USB cable.

    Test three: my own prototype, 41 minutes from Brecon to Llandovery, unplugged, screen off, arrived without a hitch. Battery used? 0 mAh, because the spare Moto thinks a full battery is 667 mAh and the level sat on 100% the whole way.

    What I can say: with the old app an hour's ride cost over a third of my Pixel's charge. With Ridelark I come home with something like 90% left. One rider, one phone, an anecdote rather than a measurement. The closed beta is where that gets proven, or not, on other people's bikes with a real GPS keeping them awake.

    And if a tester messages me twenty minutes up the A40 asking whether it's still running... it probably is.

    phpboyscout.uk/asleep-at-the-d

Share on Mastodon

Enter the server where you have an account.