Deploy Forager's Field Station

.claude/settings.local.json

@@ -0,0 +1,27 @@
+{
+  "permissions": {
+    "allow": [
+      "Bash(python -c \"import gradio_client; print\\('ok', gradio_client.__version__\\)\")",
+      "Bash(pw-record --list-targets)",
+      "Bash(command -v wpctl)",
+      "Bash(wpctl status *)",
+      "Bash(grep -iB2 \"Audio/Source\")",
+      "Bash(arecord -l)",
+      "Read(//tmp/**)",
+      "Bash(timeout 6 pw-record --target \"alsa_input.usb-RODE_Microphones_RODE_NT-USB-00.analog-stereo\" --rate 48000 --channels 1 smoketest.wav)",
+      "Bash(ffmpeg -hide_banner -i smoketest.wav -af volumedetect -f null /dev/null)",
+      "Bash(wpctl set-volume *)",
+      "Bash(wpctl get-volume *)",
+      "Bash(timeout 6 pw-record --target \"alsa_input.usb-RODE_Microphones_RODE_NT-USB-00.analog-stereo\" --rate 48000 --channels 1 smoketest2.wav)",
+      "Bash(ffmpeg -hide_banner -i smoketest2.wav -af volumedetect -f null /dev/null)",
+      "Bash(timeout 6 pw-record --target \"alsa_input.usb-RODE_Microphones_RODE_NT-USB-00.analog-stereo\" --rate 48000 --channels 1 smoketest3.wav)",
+      "Bash(ffmpeg -hide_banner -i smoketest3.wav -af volumedetect -f null /dev/null)",
+      "Bash(mv /home/alex/narration.wav assets/video/narration.wav)",
+      "Bash(ffmpeg -hide_banner -i assets/video/narration.wav -af volumedetect -f null /dev/null)",
+      "Bash(grep -nE \"^#{1,3} |video|Video|demo|Demo|youtu|\\\\.mp4\" README.md)",
+      "Bash(cat .gitattributes)",
+      "Bash(.venv/bin/python -c \"from huggingface_hub import whoami; print\\(whoami\\(\\)['name']\\)\")",
+      "Bash(.venv/bin/python deploy.py build-small-hackathon/Forager-Field-Notes)"
+    ]
+  }
+}

.gitattributes

@@ -1,7 +1 @@
 *.onnx filter=lfs diff=lfs merge=lfs -text
-examples/chanterelle.jpg filter=lfs diff=lfs merge=lfs -text
-examples/lions_mane.jpg filter=lfs diff=lfs merge=lfs -text
-examples/poison_hemlock.jpg filter=lfs diff=lfs merge=lfs -text
-examples/wild_blueberry.jpg filter=lfs diff=lfs merge=lfs -text
-examples/yarrow.jpg filter=lfs diff=lfs merge=lfs -text
-game/deck_images/water_hemlock_deadly__663372642.jpg filter=lfs diff=lfs merge=lfs -text

FIELD_NOTES.md

@@ -29,15 +29,18 @@ tell that something is about to go wrong. The router would decide a photo was a
 "plant," then I'd run two expert models on it and take whichever was more
 confident. Max confidence wins. Democratic. Elegant.
 
-It also tried to feed people poison hemlock.
-
-Here's the failure, and it's a beautiful one. A deadly plant — say, poison hemlock —
-gets routed to "plant." My high-value forageables expert has never seen hemlock in
-its life, but it has seen ramps, and hemlock and ramps are cousins that have killed
-people who confused them. So the high-value model looks at a deadly plant and
-announces, at 0.9 confidence, that it's ramps. Delicious ramps. Meanwhile my
-medicinals expert, the one that actually knows hemlock is death, correctly flags it
-as deadly at lower confidence. Max confidence wins. The confident idiot
+It also tried to feed people lily of the valley.
+
+Here's the failure, and it's a beautiful one. A deadly plant — say, lily of the
+valley — gets routed to "plant." My high-value forageables expert has never seen
+lily of the valley in its life, but it knows ramps cold — and lily of the valley is
+the ramp's poisonous double: the two push up in the same damp woods at the same
+point in spring, with strap-like leaves so alike that experienced foragers still
+land people in the hospital confusing them. (Ramps are having a real moment right
+now, which only means more hands reaching for that leaf.) So the high-value model
+looks at a deadly plant and announces, at 0.9 confidence, that it's ramps. Delicious
+ramps. Meanwhile my medicinals expert, the one that actually knows lily of the
+valley is death, correctly flags it as deadly at lower confidence. Max confidence wins. The confident idiot
 out-votes the cautious expert. In my tests this leaked deadly-as-edible about six
 percent of the time, and not one of my per-model benchmarks caught it, because no
 single model was ever wrong. The *system* was wrong.
@@ -64,7 +67,7 @@ gadget is a paperweight and their own guessing is faster. And here's the part th
 took me an embarrassing while to accept: you cannot gate your way to zero. I mapped
 the whole curve. Loosen the gate and you get a decisive, useful tool that
 occasionally calls something dangerous safe. Tighten it and you get a safe tool that
-cries "I don't know" at a raspberry. There is no magic threshold that gives you both,
+cries "I don't know" at a blackberry. There is no magic threshold that gives you both,
 because the residual risk isn't low confidence — it's the model being confidently,
 specifically wrong, and no confidence knob catches that.
 

README.md

@@ -24,6 +24,12 @@ Built for the **Build Small Hackathon** — Backyard AI track. The honest fit:
 a forager in the woods has no signal, so a small on-device model isn't a
 compromise, it's the only thing that works.
 
+[![Watch the demo — a tiny on-device model that refuses when it isn't sure](https://img.youtube.com/vi/VsLo-ZmxIaU/maxresdefault.jpg)](https://youtu.be/VsLo-ZmxIaU)
+
+*▶ Watch the 4:45 demo: backyard photos run live through the router and experts.*
+
+💬 **Read the story / join the conversation on X:** [@thefullnacho](https://x.com/thefullnacho/status/2066218609576186274)
+
 ## How it works
 
 ```