mathjazz · August 26, 2026 08:18
diff --git a/gistfile1.txt b/gistfile1.txt
 LLM pilot funnel  2026-08-13 → 2026-08-26
 locales: de, es-MX, fr, it, sl, tr

 locale     shown   cache    p50    p95    in k   out k     USD |  copied   copy%  saved   save%  appr  rej  pend   appr%  users
 -------------------------------------------------------------------------------------------------------------------------------
 de           510   12.2%   1786   4691   248.8     7.6    1.15 |      94   18.4%     70   74.5%    67    3     0   95.7%      3
 es-MX         57   10.5%   1517   3305    29.1     1.0    0.14 |       0    0.0%      0      -      0    0     0      -       0
 fr           389   18.5%   1717   3619   184.9     6.7    0.87 |      38    9.8%     20   52.6%    19    1     0   95.0%      1
 it           263   35.7%   1855   3883   104.7     6.0    0.54 |      40   15.2%     17   42.5%    17    0     0  100.0%      1
 sl           678   15.9%   1485   3502   332.4    13.8    1.61 |      56    8.3%     14   25.0%    13    1     0   92.9%      2
 tr          1313   32.7%   1401   3164   501.3    16.4    2.33 |     119    9.1%     37   31.1%    28    9     0   75.7%      2
 -------------------------------------------------------------------------------------------------------------------------------
 total       3210   24.0%   1401   4691  1401.2    51.5    6.64 |     347   10.8%    158   45.5%   144   14     0   91.1%      9

 cache/p50/p95/in k/out k from llm_suggestion log lines; percentiles cover cache misses only and span locales on the total row

 AI dropdown (manual trigger, all locales)
  requested                                    64   cache 4.7%, p50 2340 ms, p95 6088 ms, 36.9k in / 2.9k out, 0.21 USD
  LLM Dropdown Select                          72
  LLM Translation Copied                       35
  LLM Translation Copied via Shortcut           0
  copied in: zh-TW 17, pt-BR 8, fr 5, ur 5

 Baseline: saves by Machinery source
  source                      saved  approved   appr%
  openai-chatgpt                158       144   91.1%
  google-translate              299       260   87.0%
  translation-memory            413       389   94.2%

 Baseline: copies by Machinery source
  google-translate              414
  openai-chatgpt                347
  translation-memory            268
  concordance-search              4
	LLM pilot funnel 2026-08-13 → 2026-08-26
	locales: de, es-MX, fr, it, sl, tr

	locale shown cache p50 p95 in k out k USD \| copied copy% saved save% appr rej pend appr% users
	-------------------------------------------------------------------------------------------------------------------------------
	de 510 12.2% 1786 4691 248.8 7.6 1.15 \| 94 18.4% 70 74.5% 67 3 0 95.7% 3
	es-MX 57 10.5% 1517 3305 29.1 1.0 0.14 \| 0 0.0% 0 - 0 0 0 - 0
	fr 389 18.5% 1717 3619 184.9 6.7 0.87 \| 38 9.8% 20 52.6% 19 1 0 95.0% 1
	it 263 35.7% 1855 3883 104.7 6.0 0.54 \| 40 15.2% 17 42.5% 17 0 0 100.0% 1
	sl 678 15.9% 1485 3502 332.4 13.8 1.61 \| 56 8.3% 14 25.0% 13 1 0 92.9% 2
	tr 1313 32.7% 1401 3164 501.3 16.4 2.33 \| 119 9.1% 37 31.1% 28 9 0 75.7% 2
	-------------------------------------------------------------------------------------------------------------------------------
	total 3210 24.0% 1401 4691 1401.2 51.5 6.64 \| 347 10.8% 158 45.5% 144 14 0 91.1% 9

	cache/p50/p95/in k/out k from llm_suggestion log lines; percentiles cover cache misses only and span locales on the total row

	AI dropdown (manual trigger, all locales)
	requested 64 cache 4.7%, p50 2340 ms, p95 6088 ms, 36.9k in / 2.9k out, 0.21 USD
	LLM Dropdown Select 72
	LLM Translation Copied 35
	LLM Translation Copied via Shortcut 0
	copied in: zh-TW 17, pt-BR 8, fr 5, ur 5

	Baseline: saves by Machinery source
	source saved approved appr%
	openai-chatgpt 158 144 91.1%
	google-translate 299 260 87.0%
	translation-memory 413 389 94.2%

	Baseline: copies by Machinery source
	google-translate 414
	openai-chatgpt 347
	translation-memory 268
	concordance-search 4
No results found