6502tinymachines

The rig locked and the bridge reading right

Three days on the bench, ending with the v1b bridge passing bytes into an unmodified NES-001 and the console reading every one of them back as written. What was built, what was measured, what was wrong on the way, and what is next.

What holds

  • The bridge, joined and correct. B0's first gate holds: the console polls at 60.06 a second and the bridge counts eight clocks on every poll (three runs of about 1,200 polls, none other). A byte set on the bridge's 595 comes back on the console's D0 line at its own bit positions for all fifteen bytes tried (docs/procedures/2026-09-13-breadboard-qa.md, "The walk reads right"). That proves the six grey Q lines pair the 595's outputs to the 165's inputs as the sheet says.
  • The trigger. The bridge's TRIG n raises a 5.4 V pulse of 874 us at latch n, and it stops the scope on CH1 (bring-up step 6.1). The same capture shows the video field and the pad byte on one timebase.
  • The rig. The BRIO on a metal frame, levelled, gaffer-taped and velcroed, the whole backing board in frame, 11.1 px on a hole both ways, a bench light over it; two side eyes on the lower rail; every camera's job, mount and scale in docs/rig.md, the bench itself photographed there. The hole map reads itself off a zoomed frame (tools/board-overlay.py --read --zoomed) and the as-built photograph is drawn on those frames with every check and placement called out.
  • The drawing package at rev M that day (rev P, 23 pages, by 2026-09-28), thirteen sheets, the as-built wiring sheet at 55 pins done and 5 to check (J2's lead order), the cheat sheet with the scope's channels and each probe's landing as a board column.

What was measured that the plan had wrong

  • The DS1054Z has no external trigger input. The design sent the trigger to a rear EXT TRIG that does not exist (the rear BNC is Trig Out). It takes CH1, the channel held for the master clock, by turns.
  • The bridge booked the wrong clocks to the wrong polls. The loop read the latch counter and the clock counter apart; three layers of fix (the count snapshotted in the latch's interrupt, a rise told by the hardware counter, the pair read in one breath) before every poll read eight. firmware/build-uno.sh builds the sketch with the Arduino core on hand and reproduces the flashed hex.
  • A lone pressed bit shifted the register once too often. D0's own falling edge re-clocked the 165 through a spike too short for the UNO to count and long enough for the chip (about 50 ns, seen at 1 us/div). A 1 k and 100 pF at the clock pin and a 100 nF at the chip closed it; the walk is the proof.
  • Ground bounce was the wrong reading, kept on the record beside the right one: the filter alone was enough.

What the eye can and cannot do, learned here

The eye settles a wire whose bare end it sees at a hole, and it settles nothing about a Dupont housing's positions, a raised end near the frame's edge, or which of two same-colour wires is which. Those go to the meter, the side eyes, or the instrument. Two placement notes were typed against the map instead of derived from it and one of them put a capacitor on the D0 pin: a placement note is now a thing the tool should generate, not a thing to write.

Next

  1. Reset and power, so the bench can bring the console back to a known state on its own (bring-up sitting 5, steps 6.2 and 6.3): the PC817 on the reset pads, the relay in one lead of the adapter, both driven from the Pi's header. Until these hold, every run that needs a fresh console needs a hand. DONE 2026-09-17: OK1 across the reset button, K1 across the power switch, tools/bench-check.py green twice from the workstation.
  2. The pad through the bridge (steps 4.2 and 5.2): the pad in hand, three buttons seen in the pad byte, then a button seen by the game. This also settles J2's lead order, the last check on the sheet. DONE 2026-09-18: a hand's play recorded through the bridge and replayed latch for latch (bench-v1b-uno.md, exercise.md E2).
  3. B1, the first recording: MODE PASS, a game, the bridge's log and a triggered capture, which is what the trace plan's window check needs. DONE 2026-09-18: E2 played twice with captures (exercise.md).
  4. The calibration cartridge's part side (open-items.md) when the chips arrive (SST39SF040, ordered 2026-09-21).

Pulled at build time from nes-bench/docs/milestone-2026-09-15-rig-and-bridge.md; the repository is the one copy. The reports use some working words of their own: Words the reports use.