TEF128 routes fine; what failed was my own -nodsp constraint

[gHashTag]

I reported TEF128 as not converging in routing and put that in the limitations as a property of the rung.

It is not. With the device's own DSP48 blocks enabled:

	LUTs	DSP48	Fmax
TEF64, fabric only	7,479	0	48.20 MHz
TEF128, DSP enabled	4,869	56	103.39 MHz

Comfortably — and at more than twice TEF64's fabric-only frequency.

What actually failed

TEF128 under my own -nodsp, which I impose everywhere so the figures describe fabric alone and are not confounded by how many hard multipliers a part happens to carry.

That is the right policy for comparing formats. Reporting its consequence as a property of the format was not.

The limitation now carries both numbers and says they answer different questions: fabric-only for comparison between formats, DSP-enabled for what the rung actually costs on this device. Six of nine rungs measured rather than five.

The pattern

This is the fifth claim of my own that a measurement has falsified in this campaign, and all five have the same shape:

I reported the consequence of a choice I had made as though it were a property of the thing I was measuring.

• the rungs "don't fit their widths" — I had measured a binary-fabric packing cost
• "one trit per tripling is a new class" — it is takum's, off by an additive 1
• "the gap is empty because the intermediate is too expensive" — it costs 18% more than posit's, which ships
• "TEF64 will cost 4,762 LUTs" — a power law fitted to narrow rungs, extrapolated
• "TEF128 does not route" — under a constraint I set

🤖 Generated with Claude Code

[gHashTag]

Reviewer bee Z: this PR conflicts and cannot merge as it is. Main deleted research/arxiv_tef/tef_paper.tex (and the PDF) in 8aaf30e during the TEF→TNF rename. The paper now lives at research/arxiv_tnf/tnf_paper.tex.

The correction is still not on main. The TNF128 row in tnf_paper.tex (~line 1316) still reads 10,029 LUT / 0 DSP48 / 33.23 MHz, which is the -nodsp figure. The DSP-enabled numbers this PR reports (4,869 LUT / 56 DSP48 / 103.39 MHz) appear nowhere on main.

Please re-apply the correction to tnf_paper.tex on current main as a fresh PR. Leaving this one open for the author.

[gHashTag]

Bee review — left open (conflicts) (head 065d1f73c592)

Modify/delete conflict: this PR edits research/arxiv_tef/tef_paper.tex and tef_paper.pdf, and main deleted both in #520 (8aaf30ed, 2026-08-11). It also changes a claimed result: TEF128 goes from "does not converge" to 4,869 LUT / 56 DSP48 / 103.39 MHz with DSP enabled. No nextpnr log or harness receipt in the diff backs that, and it needs one wherever the limitation text now lives. Rebase onto main, find the current home of the TEF128 limitation, and attach the routed report.