FOH key light balance (LEKO FOH 1 ignores DMX)

Rendered from the working notes KEY-BALANCE.md on 2026-10-11.

FOH key light balance (2026-10-10, live desk on the lighting PC)

Answer

• LEKO FOH 1 (fixture 1, DMX 1.33) does not respond to the desk. We hold real control of LEKO FOH 2 and of both FOH SPOTs. LEKO FOH 1 did not react at 80 %, and did not react with all eight of its channels at full. Confidence: high.
• The desk is not the cause. Both lekos receive identical DMX on every channel. So do both FOH SPOTs, apart from their pan and tilt. There is no curve, look, override or baseline difference between the units of each pair. Confidence: high.
• The cause is at the fixture. LEKO FOH 1 is listening on a different DMX address, is not getting DMX through its link, or has failed. Whatever light it gives comes from something other than its patched channels 33 to 40. Confidence: moderate on which of the three. Someone at the unit can tell in one look.
• Remedy. On LEKO FOH 1's menu, set the address to 33 and the mode to "Direct, Strobe off, Fan auto" (8 channels). Then check its DMX in and through cables. Re-run the round-3 test afterwards.
• FOH SPOT pair. At the same DMX, FOH SPOT 1 (201) reads dimmer than FOH SPOT 2 (202). Its pool mean is 18 % lower and its 90th percentile 33 % lower. The two pools land in different places, so throw and angle account for part of that. Confidence that the unit itself is dimmer: low to moderate.

Fixtures

• Group 21 "LEKO" (the desk's KEY group) is fixtures 1-6. Group 22 "FOH SPOT" is 201 and 202. The lead's brief called these "KEY" and "FOH".
• Sources: unstable/data/patch-full.json, rig.json, groups.json, console/fixturetypes/channel-maps-2026-09-14.txt, dmx-baseline-overrides.json.
• Neither type has a dimmer-curve or mode channel in the console profile. The desk renders dimmer linearly: 80 % = 204, 55 % = 140.
• The one difference within each pair is in the baseline overrides. 201 and 202 have different pan values (121/214 vs 126/147) and the same tilt (56/227). Their pools land in different spots.

Raw DMX before the test (universe 1)

The before values changed because another agent recalled looks on the desk at 19:39:09 and later. Each round restored to whatever it found.

Method

• Each light was driven through /api/cmds with Fixture N / Attribute "Dimmer" At Decimal8 v in the desk's own grammar. This is a raw programmer value on the dimmer channel alone.
• Attribute "Dimmer" At Default deletes that value and nothing else, so the desk's look and fader levels come back untouched.
• Camera-safe mode was off (/api/safety), so no key-floor line was added.
• After every state, /api/dmx/frame was read back to prove the value was on the wire. Every value landed exactly (job logs in key-balance/).
• Round 2 alternated A (one on, other 0), B (swapped) and Z (both 0) four times. Luma = mean(on frames) minus mean(Z frames), per pixel. This cancels the rest of the rig, which held still (12 frame reads across 10 s showed zero changed channels).
• Round 1 used single-shot states. Other activity on the rig changed the background mid-run (spot beams and a gobo come and go in round1-contact-sheet.jpg), so it is kept only as evidence that the values landed.

Luma (round 2, 900x506 frames, 0-255 luma, added over both-off)

• Repeatability of total added light per repetition, in thousands of luma units. LEKO FOH 1 gave 303, 205, 212, 174, which is the size of the frame noise between two both-off frames (477 to 520). LEKO FOH 2 gave 2336, 2301, 2245, 2223. FOH SPOT 1 gave 1491, 1480, 1348. FOH SPOT 2 gave 2156, 1955, 2068, 1923.
• LEKO FOH 2's difference image (diff_L2.png) is a broad front wash: mic-stand shadows, a lift on the drum kit, no holes outside the gobo pools. The both-off frames hold mean luma at 37.5 to 37.8, so this is real light, not a camera exposure shift.
• Round 3 set all eight of LEKO FOH 1's channels to 255 vs 0, three times, with LEKO FOH 2 at 0. Added mean luma was 0.21, 0.12, 0.12. Pixels above 10 numbered 775, 654, 667, the same as both-off-vs-both-off noise (628, 790). The unit does not respond to any channel in 33-40. See diff_L1_fullfootprint.png.

Verdict

• Same frame values, very different light. The fault is at the fixture, not in the desk data. Confidence: high.
• LEKO FOH 1 is outside the desk's control. The likely causes are a wrong address on the unit, a DMX link break before it, or a failed unit. Confidence: moderate on which. The light Greg sees from it is not coming from its patched channels.
• If the unit is set to another leko's address (1, 9, 17, 25 or 41), it would still come up with Group 21 but would follow that fixture's values. That fits "consistently darker". This is inference: the camera cannot separate it from the five other lekos.
• A camera view from the FOH position cannot rule out that LEKO FOH 1 is aimed entirely out of shot. But it would have to throw no spill at all onto the stage. Confidence that this is the explanation: low.
• FOH SPOT 1 vs 2: the same DMX on dimmer, shutter, colour, gobo and focus. The pools differ by 18 to 33 %. Part of that is throw geometry, because the pools land in different places. Re-focus both onto one mark and compare by eye before trimming.

Desk-side change (written, committed, not deployed)

• Commit 81c09df on branch two-screen in /Users/gregspero/trc-lighting/unstable, not pushed. It adds data/dimmer-gain.json support, { "202": 0.85 } = fixture id to an intensity trim from 0 to 1. It scales only the rendered dimmer, in Dmx/Engine.cs Render(), and is loaded in Program.cs next to the baseline overrides. With no file, nothing changes.
• It compiles (dotnet build -c Release, scratch copy). It is not deployed to the PC. No gain file was created.
• This trim cannot fix LEKO FOH 1, because no DMX value reaches it. It is for the FOH SPOT pair, if a same-mark comparison still shows 202 brighter.

Restore proof

• /api/rig/state showed 1 and 2 at 55 and 201 and 202 at 0, both before and after round 2.
• No other fixture, look, master or setting was changed by these tests. The rig changes between rounds came from another agent's look recalls (rig version changed at 19:39:09).