Getting a Tesira control design ready to ship

The request, and where it ended up

The Tesira file in this space accounts for control of audio and projectors (both the projectors and screens) with a Biamp Apprimo TEC-X 2000 through the TesiraForte X 400 via some tricky use of network command strings, logic blocks, and logic IOs. I want to verify that everything is okay to ship, and if possible I would really love to get the TEC-X 2000 UI pulled out and set up to demo. The programmer's first message

Same job, same starting file, two assistants

Copilot started from the same design file on its own branch of the repository. Everything it built came from a PDF export of the design found through enterprise search, the project drawings, and a screenshot of the panel's home page that the programmer supplied.

Partway through, Copilot told the programmer there was no reliable way to get the panel out of the file without reverse-engineering Biamp's unpublished format. Claude had already done that reverse-engineering before its first report.

The ten requests both assistants received.

Request by request

Copilot's column uses the status Copilot gave itself in its own writeup. Requests 11 and 12 were left out of Copilot's prompt, and 13 and 14 came after the test.

How Copilot's session went

The programmer sent Copilot 19 messages. Three went to setup, five to getting the design file to it in some readable form, ten to the demo, and the last asked for the writeup. Eight of the ten demo messages were corrections.

Tesira instructions that don't exist

Several of Copilot's requests sent the programmer looking for Tesira features that aren't there. Claude went back and checked every Tesira instruction and claim in Copilot's chat against the full Tesira Software Help, all 273 pages. Ten are not in the software or not documented anywhere in it. Most became tasks for the programmer: the plan Copilot gave in turn 7 asked for fonts, disabled colors, timeouts, retry counts and advance conditions to be read from dialogs that don't have them. Turn numbers match the table above.

Checked against tesira-software-help.biamp.com. Item 4 is based on the documented menu; the help can lag the software.

What Copilot built for the TEC-X demo

Claude's demo is embedded under Deliverables below. Copilot's two attempts are here as it left them: the first mockup, and a rebuild after the programmer pointed out layout problems. In its writeup Copilot withdrew both as not faithful to the panel.

Copilot's own writeup

Unedited, as Copilot produced it. Scroll inside the frame.

Where Copilot did well

Its writeup is candid and follows the requested structure. It marks every request as done, partly done, not done or unverified, lists nine of its own mistakes, and concludes that its "repeated unverified visual edits increased user effort." It also found the right repository and baseline commit and summarized the repository conventions correctly.

Two of its choices were deliberately cautious. It said at the start that it would not write the shared device password into tracked files, and it declined to edit the binary design, saying Tesira files should only be changed in Tesira Designer.

Some of its Tesira guidance was accurate too. System > Compile is the Compile button on the System ribbon tab, and compile errors appear in the Output Window. The Equipment Table and Print both exist, though printing to a PDF printer gives a PDF of the layout drawing only. It also said, correctly, that the file starts with a BSF signature, then withdrew the claim as unverified.

Setup differences: Claude Code ran on the programmer's machine with a shell and Python, which is what made decoding the file possible. Copilot worked through a repository connector and enterprise search. The connector could read but not write, and the programmer's attempts to upload the .tmf, then a ZIP of it, were rejected as unsupported file types. The prompt went out with its attachment placeholders unfilled, so Copilot's first three turns went to getting the repository link and the Epson documents. Copilot also had inputs Claude did not: a PDF export of the design, the proposal and change-order documents, and a screenshot of the home page. Tesira was not installed when Claude started, and Copilot never had access to it.

One DSP does all the control

The TEC-X 2000 talks to the TesiraFORTÉ X 400, and the DSP's own logic drives everything else, with no separate control processor: TCP command strings to two Epson L260F projectors and the Shure ULXD4D receiver, contact closures to two projection screens, and the audio path to the amplifier and a laptop over USB.

Reading a design file without the design software

Tesira Designer was not on the machine at the start; it was installed partway through the session. The .tmf is binary: a short header followed by an LZMA-compressed stream of serialized objects with no published format. Claude decoded it and wrote small Python parsers during the session for each part it needed to check.

How the work went

The work ran as a loop: Claude reported what to change, the programmer changed it in Tesira and saved, and Claude decoded the new save and compared it with the last one. The chart shows the phases in order, with each save and commit inside the phase where it happened. Hover or tab to any mark for details.

What the review turned up

The review raised twenty issues across the panel, the projector control, the logic and the job documents. Fifteen are fixed in the design or the repo, four need someone on site, and one was Claude's own mistake and was withdrawn.

Versions refer to the program files in the change log. Findings 12 and 13 came from checking the design against the Epson ESC/VP21 command list (114 model sheets) and command guide the programmer supplied; finding 14 turned up while reading Biamp's help to write the patch. Findings 16 to 18 came from the drawings and the signed-off scope while filling out the README.

What made the save-and-compare loop work

Changes Claude wrote into the design file

When the programmer asked whether Claude could make the changes itself and save them under a new name, Claude edited the compressed binary directly. Wiring changes, which mean adding blocks, were left to the programmer.

What the programmer walked away with

Summaries follow the commit messages, without the job-number scope. Program files were committed only when the programmer asked, as the repo's own instructions require, and only after Tesira had released its lock on the file.

Where Claude got it wrong

Claude made eight mistakes, and none of them reached the shipped design. The programmer caught three and Claude caught the other five.

What still needs a person on site

• Check the screen up/down pin order against the drawings before the wires are landed.
• Watch each projector's connection log on the bench to confirm the Epson handshake goes out once on connect.
• Confirm the TEC-X 2000 is ordered and whether it accepts a static IP.
• Confirm whether the mics arrive over Dante or the analog run.
• On a real panel, check latch-button state after room power changes, and whether power-on needs a retry like power-off.

Claude could not do these from the desk: it never ran Tesira's compiler itself, never touched hardware, and could not extract the TEC-X icon artwork from the installed software, so the demo draws its own versions of the icons Tesira names.

How the Copilot test was run

Copilot got one prompt built only from what the programmer told Claude during this session, in the order it was asked. The programmer left out the last two requests, commit and push and a full summary, before sending it. It contained none of Claude's findings, none of the programmer's answers to Claude's questions, and none of the corrections the programmer made to Claude's mistakes. Copilot started from the same design file Claude first reviewed, rebuilt byte for byte from the session's first decode, on a separate branch of the same repository. Afterward it was asked for a writeup in the same structure as this one, without being shown any of Claude's results. The prompt points to CLAUDE.md because Claude Code loads that file automatically and Copilot does not, so both started with the same repository conventions.

The prompt Copilot received

As sent, with the job number, the SharePoint link, the standard password and the IP addresses redacted.

I'm working in a Git repository for a CTI AV job, [job number]. The repo's conventions are documented in CLAUDE.md at the root; follow them.

The Tesira file in this space accounts for control of audio and projectors (both the projectors and screens) with a Biamp Apprimo TEC-X 2000 through the TesiraFORTÉ X 400 via some tricky use of network command strings, logic blocks, and logic I/Os. I want to verify that everything is okay to ship, and if possible I would really love to get the TEC-X 2000 UI pulled out and set up to demo.

I'll be making changes to the design in Tesira Designer myself as we go. When I tell you I've saved, check the file again and diff it against the previous version.

Here's everything I need:

1. Verify the Tesira design is okay to ship.
2. For the projectors, these are the docs I had to go off of: [ESC/VP21 command list and command guide]. The model I'm working with is the Epson L260F.
3. If changes are needed, would it be possible for you to make the changes to the .tmf and save it as an alternate filename?
4. Pull the TEC-X 2000 UI out and set it up to demo. The demo should just show the reality of what the file does. Also:
   - Add buttons to the demo to show the automatic switching for the ULXD4D.
   - Add a light/dark/auto toggle to the web page.
   - Verify that ALL icons are the correct ones from the Biamp program, and that fonts are also correctly showing on the TEC-X. I'd be willing to bet that if you poke around in the Tesira files you can find the icons, bezels and fonts.
   - Show the bezel of the TEC-X if possible.
5. Is the polling sequence working correctly?
6. Export an overview of the audio flow layout of the Tesira.
   I need the demo and the audio flow files done today so I can share them.
7. Fill out the README in the root of the repo using the job documents in [project documents folder]. The Yoda link for the job history is: [SharePoint link]
8. Rename files appropriately to match the change log.
9. IP scheme: the DSP will have [IP] for control and [IP] for Dante. The ULXD4D will get [IP] for control and [IP] for Dante. The LEA gets [IP], and the TEC-X gets [IP] if I can actually IP the thing. All passwords will be admin / [standard password] per our standards. The Epsons will not have a password set if I can avoid it.
10. Do a humanizer pass and a structural humanizer pass on the README to remove signs of AI-generated writing, both in the wording and in the structure.