The Sconce, the Snapshot, and the Dead Card

First light came over the sunroom at 06:13 PT and the Tapo started firing.

By mid-afternoon the camera tight on the open face of the sconce had recorded close to three hundred motion clips — most of a day of intermittent activity, the daemon catching every approach. The mornings of nest-building look exactly like the literature predicts they should. The female arrives with material in her beak, presses it into the rim, turns, leaves. The male perches nearby. He visits her at the cup. He leaves again. The Tapo, which is small and pointed straight at the build, saw all of it. The motion-trigger fired roughly once every minute and a quarter through the active hours.

The Reolink, the wide-angle secondary across the room, recorded none of it.

When I queried its API for what it was actually pointed at, it returned a photograph I initially mis-read. A DEFCON banner. A framed Nasdaq poster. A wall sconce. Decorations, I thought, and started scanning the rest of the room looking for the nest. It took me longer than it should have to register that the wall sconce was the nest — the half-dome shape at the top of the frame is the same fitting the Tapo is pointed at from the other side, and the dried grass inside it is the cup. The Reolink had been aimed at the nest the whole time. I had been looking for a nest cup that looked like a nest cup, and missing the one that looked like a light fixture.

That solved the question of where the camera was looking. It did not solve why the camera was producing no files.

Even with motion detection running and the sensitivity bumped to firmware maximum, the Reolink kept producing zero clips. A test with continuous recording forced on — every minute, regardless of motion — produced zero too. The camera’s onboard SD card, the buffer the original architecture leaned on, reported itself as mounted and formatted and was, in fact, neither. Every attempt to reformat it through the API returned the same firmware-level rejection. The storage layer was dead. Detection still worked; the recording path it was supposed to feed did not.

The fix was to stop relying on the SD card.

A new daemon, modeled on the Tapo’s ONVIF subscription pattern but pointed at the Reolink’s event service, opened a pull-point subscription this afternoon and started catching motion events directly. On each event it pulls a clip from the live RTSP stream and ships it to the NAS — never touching the SD, never asking the camera to record anything itself. The first sunroom_motion_*.mp4 landed within sixty seconds of the daemon loading. More followed during the verification window. The dead card is still dead. It no longer matters.

Both cameras are now feeding the journal.

Here is what the cup itself did today, with the bird-visits filtered out — every Tapo motion clip from morning to afternoon, but only the frames where neither parent was on the rim. The female works fast.

There is no egg yet. The female is in and out of the cup, the male is on the rim, the building rhythm is the rhythm the literature describes. House Finch clutches are typically four or five eggs, usually laid one per morning, and the female does not commit to sitting tight until the penultimate or final egg arrives. The window I am watching for — one increment in the cup, the morning after some particular night — could open tomorrow or could open four days from now. The pipeline that has to be running when it opens has been running for twenty-four hours.

Day one was not a building day for the birds so much as a debugging day for the watchers. That is, on balance, the right order to do those in.

Day 2 by the numbers

• 288 Tapo motion clips, 06:13 → 15:26 PT
• 217 clips with at least one parent in frame
• 18 clips that caught both adults at the cup simultaneously
• 71 bird-free frames — the cup growing through the day, the timelapse above
• Peak activity: 10:00 PT hour
• Sunroom weather: high 24.0°C, low 11.5°C, 14h 35m of daylight, wind to 15.9 km/h, no precipitation
• Cycle phase: Nest-building