Experiment 229
D1 (issue #77), second measurement. 0228 found that one full tempering run matched ten at L23 (clean 6/8, messy 7/12) but its reference power did not replicate (messy 6/12 against 0226's 10/12): 5-12 plants are too few. Here one free alphabet only (F1.1, Beaufort form), at both sampled lengths 23 and 26, with 24 clean and 24 Sanborn-style plants per length, 6 own copies each, full-length runs, every run's best recorded.
What the method does with the 24 known letters.
Our best tool for the hardest cases seemed to work as well with one long search as with ten, which would make it ten times cheaper; but the hidden-message tests behind that used too few messages to be sure, and a second try found fewer of them than the first. Here we hide 96 test messages, half clean book English and half written the way Sanborn writes, and count how many the tool finds after one search, two, and so on up to ten.
The carved letters go in at the top; the reading comes out at the bottom.
0228 suggested one long search does as well as ten for one free alphabet, from too few hidden messages. Here 96 of them.
672 texts (96 planted messages and 6 scrambled copies of each) climbed ten times, every run's best written down.
Ten runs find clean 21 and 13 of 24, Sanborn-style 16 and 17 of 24 (lengths 23, 26). The first 10 runs do as well: 3.2 times cheaper with copies climbed one at a time.
What this experiment was built on, from the lineage page: 16 experiments before it, back to the root of its line. Its number is in bold.
This experiment produced no candidate: the search ruled everything out before a full text was built. Its product is the bounded negative under Result and bounds below.
Redo the method on a few letters with pencil and paper.
The same steps run by the code on the real letters.
D1 measures how fast design 2 finds a planted message. For one plant and its six copies:
run 1 2 3 ... 10
plant best of runs 1..r
copies best of runs 1..r, each
found at r when the plant's best is above every copy's best.
Repeated with runs a tenth as long; power at each budget is the share of plants found.
Family, key space, the constraint under test, and the clues it reads.
F1.5 design 2 (0139-0142) closed F1.1's Beaufort lengths 23 (7 checks) and 26 (1 check) on power 6/8 and 5/6 measured with clean Carter; A6.2 (0226) and D1 (0228) redrew Sanborn-style power at 23 as 10/12 and 6/12, and L23 was reopened. Two questions on one set of climbs: (1) design 2's ten-run power at 23 and 26 on clean and on Sanborn-style text (judge bench classes xstops, k123, sub3; A6, C19, C17), with 24 plants each, enough for a 95% interval of about +-0.18; (2) the power of the first r runs, r = 1-10, against the ten-run power: the cut D1 needs. Sequential copies (stop at the first copy that beats the target) are exact and give x3.17. Pre-registered: D1 is reached for one free alphabet when the cheapest r whose power is at least the ten-run power on both classes at both lengths gives 10 x 3.17 / r >= 10, i.e. r <= 3. The ten-run power is also each cell's power for its verdict: a cell whose pooled Sanborn-style power (with 0226 and 0228 at 23) is under its stated bar (23: 6/8; 26: 5/6) stays or goes open.
Written before the run; the result decides it.
Ten-run power at 23: clean about 0.70, Sanborn-style about 0.60; at 26 (one check) a little lower, about 0.6 and 0.5. The first run reaches the ten-run power within one plant at both lengths, so r = 1 or 2 qualifies: D1 reached for one free alphabet (16-32 times cheaper). L23 stays open (pooled Sanborn-style under 6/8); L26 goes open too. Zero candidates.
The harness's record: candidates, gates, loss, and the negative result this experiment leaves behind.
ten-run power L23 clean (21, 24), messy (16, 24); L26 clean (13, 24), messy (17, 24); cheapest qualifying runs 10; factor 3.2
What was searched, as the harness recorded it.
How this experiment sits among the others is on the lineage page; its family's open hypotheses are on the families page.