Experiments · E208

Can the search still find good alloys if each must mix five elements?

Not yet. It was never run: it waits on an energy-model refit that passes its tests, and a trial run failed.

In the log: and E209 — pre-registered before either runs (2026-09-20 21:16)

recordedDate 2026-09-20 21:16, as written in the loggenerator · fly brain3 predictions · 0 result paragraphsEXPERIMENTS.md lines 13076–13080, lines 13082–13091
exp E208 diagram
What E208 did and how it came out, drawn from this record and the files it names (book/assets/diagrams/exp/E208.svg).

Pre-registration

  1. (1)
    every find carries ≥ 5 elements — a property of the projection, so a failure is a bug and not a result (already observed on a 3-round smoke run: element_counts {5: 6} for elitist and {5: 3} for the fly).
    no verdict written against it
  2. (2)
    Finds fall by ≥ 40 % against the unconstrained 284 / 246: five-element compositions really are rarer below the qualifying bar, because the best five-element composition scores −30 meV against the best binary's −87 (E134).
    no verdict written against it
  3. (3)
    The fly's lead over elitist on finds survives the constraint — still ≥ one seed-sd. If
    no verdict written against it
The pre-registration, as written

E208 and E209 — pre-registered before either runs (2026-09-20 21:16)

Both on the pure reward, eight-element space, 200 × 4 × 2 seeds, scored against E201b (fly 136.9 ± 2.1 AUC_Q, 284 ± 22.6 finds, depth −66.8; elitist 145.3 ± 15.5, 246 ± 4.2, depth −73.7). Both sit behind flags that are off by default.

E208 — the operator's element family. FORAGER_MIN_ELEMENTS=5 FORAGER_ELEMENT_FLOOR=0.05 on the full eight elements, elitist + fly. Predictions: (1) every find carries ≥ 5 elements — a property of the projection, so a failure is a bug and not a result (already observed on a 3-round smoke run: element_counts {5: 6} for elitist and {5: 3} for the fly). (2) Finds fall by ≥ 40 % against the unconstrained 284 / 246: five-element compositions really are rarer below the qualifying bar, because the best five-element composition scores −30 meV against the best binary's −87 (E134). (3) The fly's lead over elitist on finds survives the constraint — still ≥ one seed-sd. If (3) fails, the fly's advantage was an advantage at finding binaries, which is worth knowing and is a different claim from the one E195c established.

Results

No result paragraph for this entry was found in the log.

The full record

This entry is written in 2 separate places in the log, shown here in log order.

EXPERIMENTS.md · lines 13076–13080

E208 and E209 — pre-registered before either runs (2026-09-20 21:16)

Both on the pure reward, eight-element space, 200 × 4 × 2 seeds, scored against E201b (fly 136.9 ± 2.1 AUC_Q, 284 ± 22.6 finds, depth −66.8; elitist 145.3 ± 15.5, 246 ± 4.2, depth −73.7). Both sit behind flags that are off by default.

EXPERIMENTS.md · lines 13082–13091

E208 — the operator's element family. FORAGER_MIN_ELEMENTS=5 FORAGER_ELEMENT_FLOOR=0.05 on the full eight elements, elitist + fly. Predictions: (1) every find carries ≥ 5 elements — a property of the projection, so a failure is a bug and not a result (already observed on a 3-round smoke run: element_counts {5: 6} for elitist and {5: 3} for the fly). (2) Finds fall by ≥ 40 % against the unconstrained 284 / 246: five-element compositions really are rarer below the qualifying bar, because the best five-element composition scores −30 meV against the best binary's −87 (E134). (3) The fly's lead over elitist on finds survives the constraint — still ≥ one seed-sd. If (3) fails, the fly's advantage was an advantage at finding binaries, which is worth knowing and is a different claim from the one E195c established.

Related entries

Built with PRISMWebsite and visualizations made using Claude