Skip to content

verification.md

Latest commit

 

History

History
95 lines (72 loc) · 4.23 KB

verification.md

File metadata and controls

95 lines (72 loc) · 4.23 KB
Raw

Hardware verification

The firmware is deployed and working on one real panel: it enrols, is claimed, and renders the dashboard with live calendar and weather.

Nothing below has been formally worked through. Stage 1 is substantially covered by that deployment, but the rest is not: deep sleep, the 304 no-flash behaviour, KEY1 wake, failure backoff, battery-only operation and provisioning have never been deliberately exercised. The sleep-current measurement in sleep-current.md is still an unfilled form.

Do not tick optimistically. A box that fails is a bug to fix, not a note to carry forward — and "it seemed to work" is not the same as having checked.

Date tested: ____________ Firmware: ____________

Before powering anything

  • WiFi antenna clipped onto the U.FL socket on the XIAO module
  • USB and battery both disconnected while handling the ribbon
  • 24-pin ribbon inserted the correct way round, latch fully closed
  • EE04 jumper set to 24 Pin

The antenna is easy to overlook and does not fail cleanly. Without it the panel associates roughly one boot in three, and the visible symptom is an HTTP read timeout rather than anything that looks like a WiFi problem.

Stage 1 — fetch and display (commit "fetch and display server-rendered frames")

USB powered, secrets.h created from the example.

  • Serial at 115200 shows [inkpanel] 0.1.0 device=esp32-xxxxxx
  • WiFi connects and prints an IP
  • [net] GET ... -> 200
  • Panel draws the enrolment screen, showing the server URL and this device's ID
  • Enrolment text is legible with no missing horizontal bands
  • Claim the device in the web UI, add a calendar URL and location
  • Reset — panel now draws the dashboard
  • Both outer borders reach all four edges; image is not mirrored or rotated
  • Black is even, not blotchy
  • The two Spec 2 slots show their hatched empty state, not blank white
  • Smallest text (13px weather detail, footer) is readable at arm's length

Stage 2 — deep sleep (commit "deep sleep, RTC-persisted ETag")

  • After drawing, serial prints [sleep] 900 seconds and goes quiet
  • Device wakes on its own and prints wake=timer
  • Second wake logs 304 and the panel does not flash
  • Press KEY1 — wakes immediately, logs wake=button
  • Change a calendar event, wait one interval — panel updates
  • Stop the server — panel keeps its image, log shows increasing backoff (900, 1800, 3600, capped)
  • Restart the server — device recovers on its next wake without a reset
  • Disconnect USB, run on battery only — still wakes and refreshes

If the panel flashes on a 304, the ETag is not surviving sleep. Check that storedEtag is RTC_DATA_ATTR and that the server's ETag header is being captured by collectHeaders.

Stage 3 — provisioning (commit "on-device WiFi provisioning")

Delete secrets.h and re-flash, with Erase All Flash Before Sketch Upload enabled once.

  • Serial reports the portal starting
  • inkpanel-setup appears in the phone's WiFi list
  • Joining it triggers a captive-portal prompt (or http://192.168.4.1 works)
  • Network list is populated and readable
  • Saving reboots the device and it fetches a frame
  • Wrong password fails gracefully and retries rather than bricking
  • Submitting with an empty server address is rejected, not saved
  • Hold KEY3 and reset — returns to the portal
  • After a factory reset the device re-enrols under the same device ID

Stage 4 — power

Requires the Task 1 measurement in sleep-current.md.

  • Deep sleep current matches the recorded figure
  • Battery percentage in the footer tracks a discharging cell
  • Low battery (below 3.5 V) lengthens the interval to 6 hours

Stage 5 — endurance

  • Runs 48 hours unattended with no intervention
  • Record battery voltage at start and end: ______ V → ______ V
  • Extrapolated runtime: ______ days
  • No ghosting or blotching after two days of refreshes

Known gaps at time of writing

  • The golden image was generated on Windows and will not match one rendered in the Linux container. Regenerate before wiring CI to it.
  • The Docker image has never been built.