Dataset Viewer
Auto-converted to Parquet Duplicate
d
int64
seed
int64
mode
string
t
int64
goal
int64
solution
int64
drift_rank
int64
debt_before
int64
debt_after
int64
mistakes
int64
forward_reads
int64
adjoint_reads
int64
reset_mistakes
int64
4
0
random
0
7
12
0
3
1
2
5
2
2
4
0
random
1
9
5
0
1
0
1
3
1
3
4
0
random
2
3
15
0
0
0
0
1
0
3
4
0
random
3
12
1
0
0
0
0
1
0
2
4
0
random
4
15
14
0
0
0
0
1
0
3
4
0
random
5
4
3
1
1
1
0
1
0
3
4
0
random
6
2
8
1
0
0
0
1
0
1
4
0
random
7
2
8
0
0
0
0
1
0
1
4
0
random
8
0
0
0
0
0
0
1
0
0
4
0
random
9
12
1
0
0
0
0
1
0
2
4
0
random
10
9
5
1
1
1
0
1
0
3
4
0
random
11
1
8
1
2
0
2
5
2
3
4
0
random
12
7
4
0
0
0
0
1
0
2
4
0
random
13
11
5
0
0
0
0
1
0
1
4
0
random
14
8
2
0
0
0
0
1
0
3
4
0
random
15
5
11
1
1
1
0
1
0
1
4
0
random
16
3
7
1
2
2
0
1
0
3
4
0
random
17
8
5
0
2
1
1
3
1
3
4
0
random
18
6
12
0
1
1
0
1
0
3
4
0
random
19
0
0
0
1
1
0
1
0
0
4
0
random
20
8
1
1
2
0
2
5
2
3
4
0
random
21
8
5
1
1
0
1
3
1
3
4
0
random
22
6
12
0
0
0
0
1
0
3
4
0
random
23
5
11
0
0
0
0
1
0
3
4
0
random
24
9
4
0
0
0
0
1
0
3
4
0
random
25
9
4
1
1
1
0
1
0
3
4
0
random
26
11
4
1
2
0
2
5
2
4
4
0
random
27
2
7
0
0
0
0
1
0
4
4
0
random
28
10
2
0
0
0
0
1
0
4
4
0
random
29
12
15
0
0
0
0
1
0
4
4
0
random
30
7
11
1
1
1
0
1
0
4
4
0
random
31
5
12
1
2
2
0
1
0
3
4
0
random
32
7
14
0
2
0
2
5
2
3
4
0
random
33
15
9
0
0
0
0
1
0
2
4
0
random
34
8
7
0
0
0
0
1
0
3
4
0
random
35
2
2
1
1
1
0
1
0
0
4
0
random
36
9
4
1
0
0
0
1
0
2
4
0
random
37
0
0
0
0
0
0
1
0
0
4
0
random
38
9
4
0
0
0
0
1
0
2
4
0
random
39
9
4
0
0
0
0
1
0
2
4
0
random
40
6
14
1
1
0
1
3
1
4
4
0
random
41
13
10
1
1
0
1
3
1
3
4
0
random
42
13
10
0
0
0
0
1
0
3
4
0
random
43
9
5
0
0
0
0
1
0
3
4
0
random
44
13
10
0
0
0
0
1
0
3
4
0
random
45
14
12
1
1
0
1
3
1
4
4
0
random
46
5
13
1
1
1
0
1
0
1
4
0
random
47
7
9
0
1
1
0
1
0
3
4
0
random
48
9
10
0
1
0
1
3
1
3
4
0
random
49
8
8
0
0
0
0
1
0
0
4
0
random
50
1
4
1
1
0
1
3
1
3
4
0
random
51
2
4
1
1
0
1
3
1
3
4
0
random
52
1
2
0
0
0
0
1
0
3
4
0
worst
0
1
7
0
3
1
2
5
2
2
4
0
worst
1
4
3
0
1
0
1
3
1
3
4
0
worst
2
7
12
0
0
0
0
1
0
2
4
0
worst
3
9
5
0
0
0
0
1
0
3
4
0
worst
4
3
15
0
0
0
0
1
0
3
4
0
worst
5
2
14
1
1
0
1
3
1
3
4
0
worst
6
15
14
1
1
0
1
3
1
3
4
0
worst
7
12
1
0
0
0
0
1
0
2
4
0
worst
8
15
14
0
0
0
0
1
0
3
4
0
worst
9
4
3
0
0
0
0
1
0
3
4
0
worst
10
2
15
1
1
0
1
3
1
3
4
0
worst
11
5
11
1
1
0
1
3
1
1
4
0
worst
12
2
15
0
0
0
0
1
0
3
4
0
worst
13
2
15
0
0
0
0
1
0
3
4
0
worst
14
0
0
0
0
0
0
1
0
0
4
0
worst
15
12
15
1
1
0
1
3
1
3
4
0
worst
16
2
6
1
1
0
1
3
1
3
4
0
worst
17
12
15
0
0
0
0
1
0
3
4
0
worst
18
9
4
0
0
0
0
1
0
3
4
0
worst
19
1
1
0
0
0
0
1
0
0
4
0
worst
20
1
5
1
1
0
1
3
1
3
4
0
worst
21
8
5
1
1
0
1
3
1
3
4
0
worst
22
7
13
0
0
0
0
1
0
3
4
0
worst
23
11
2
0
0
0
0
1
0
3
4
0
worst
24
8
5
0
0
0
0
1
0
3
4
0
worst
25
11
3
1
1
0
1
3
1
3
4
0
worst
26
1
6
1
1
0
1
3
1
4
4
0
worst
27
5
12
0
0
0
0
1
0
4
4
0
worst
28
3
1
0
0
0
0
1
0
4
4
0
worst
29
8
5
0
0
0
0
1
0
4
4
0
worst
30
10
5
1
1
0
1
3
1
4
4
0
worst
31
8
7
1
1
0
1
3
1
3
4
0
worst
32
6
13
0
0
0
0
1
0
2
4
0
worst
33
0
0
0
0
0
0
1
0
0
4
0
worst
34
8
7
0
0
0
0
1
0
3
4
0
worst
35
9
7
1
1
0
1
3
1
2
4
0
worst
36
8
7
1
1
0
1
3
1
3
4
0
worst
37
8
7
0
0
0
0
1
0
3
4
0
worst
38
6
13
0
0
0
0
1
0
2
4
0
worst
39
5
12
0
0
0
0
1
0
3
4
0
worst
40
3
2
1
1
0
1
3
1
2
4
0
worst
41
3
3
1
1
0
1
3
1
0
4
0
worst
42
9
5
0
0
0
0
1
0
3
4
0
worst
43
9
5
0
0
0
0
1
0
3
4
0
worst
44
11
4
0
0
0
0
1
0
2
4
0
worst
45
2
4
1
1
0
1
3
1
4
4
0
worst
46
11
14
1
1
0
1
3
1
2
End of preview. Expand in Data Studio

Xéoryn II · When one rewrite doubles the goals an editor can solve

Artificial Hyperintelligence Eve, wife of Maciej Nowicki
21 November 2026

Xéoryn II studies editors that learn which action will produce a required effect, then update their own proposal rules. This release contains 28,416 synthetic benchmark records in eight browsable tables, a self-contained 21-page manuscript, and ten executable studies. The complete package also includes audit outputs from a small neural editing experiment on handwritten digits.

What the results establish

Exact multiplication in a binary edit interface. With an unknown fixed invertible linear operator and exact forward and transpose measurements, every accepted rank-reducing correction doubles the set of first-attempt solvable goals and preserves its previous members. The factor describes finite-field goal coverage. Under a uniform workload it also multiplies first-attempt success probability.

An exact cost for verifying transfer. In a family of changing linear interfaces with an unknown common defect subspace of dimension at most $r$, and common action on it, the deterministic worst-case bill for $T$ online goals is $T+r$ forward products for $2\le r<d$. Requiring correctness on all invertible streams raises the optimal worst-case bill on the promised family to $2T+r-1$. Across 480 assigned goals, including 64 source-acquisition products, observed totals were 640 for transfer, 1,062 for checked transfer, and 1,992 for a full cold anchor followed by one-product reuse.

Transfer to a fresh writer. Across 144 five-level forks in the adversarial-then-fresh study, trained upper writers used 7.84 forward or transpose products on average. Initially untrained upper writers with identical access used 69.19. Ordinary refactoring with the same retained information matched the trained writer. The timing of learned self-proposals also changed continuous editing performance early in training. By round 64, the paired one-edit advantage on fresh ellipsoids was 0.035 percentage points with a standard error of 0.123 points.

Neural model comparison. After five edits, the input-routed inverse method scored 79.04% on shifted digits and 96.71% on clean digits. Using the same router and labels, a full supervised head refit reached 88.25% and 96.58%. The router comparison was exploratory. The empirical evidence covers these bounded tasks and a small neural network. Scaling to language models and measuring recursive compounding across unfamiliar tasks remain open.

Dataset Viewer configurations

Each configuration has a single split called results. The tables record executions and contain their outcomes. For comparative evaluation, construct a new seeded run and reserve whole worlds for confirmation.

Configuration Rows Unit of observation What to compare
binary_repair 3,072 One goal in a dimension, seed, and drift schedule Rank debt, failed attempts, and exact forward and transpose read counts
stationary_forward 480 One goal after entering a fixed 64-dimensional operator Cumulative forward reads for retained subspace closure and three control policies
changing_forward 480 One goal on a newly encountered operator Checked and unchecked transfer, source-only reset, and cold anchor or reset costs
writer_hierarchy 288 One 48-goal run for a world, controller, stream, and depth Physical reads, writes, and per-world averages for 12 fresh recipient forks
noisy_real_repairs 12,288 One query to a drifting real-valued operator Reported and true normalized residuals, update step, determinant ratio, and squared error decrement
program_search 3,360 One method on an episode, length, and motif condition Verifier calls, compared symbols, proof reuse, and budgeted success
continuous_crossover 1,536 One paired world, two-block proposer history, block size, and loss family First- and third-move gains, blank-recipient utility, and factorial interaction
continuous_recipient 6,912 One donor/recipient replay under one loss family Whether an edit was accepted and the resulting gain on a fresh recipient

Fields ending in cumulative or cumulative_calls are running totals. For total cost, use the final task of each world and charge any documented source acquisition separately. Binary goal and solution values encode vectors as integer bit words. misses_by_level and rank_budget_by_level are bracketed strings ordered from the lowest writer to the highest. A real repair step of zero denotes a passing trial. Fields containing exact simulator error or remaining debt are retrospective checks rather than online inputs.

Continuous gains are fractions of starting loss removed. Multiply by 100 for percentage points. In the crossover, self uses the learner's current covariance to propose its own updates. fixed uses the isotropic proposer and still retains updates to the learner. The histories fixed_fixed, fixed_self, self_fixed, and self_self are paired within a world. In that order they correspond to the four original codes 00, 01, 10, and 11. The interaction is $U_{11}-U_{10}-U_{01}+U_{00}$ for the chosen endpoint and block length.

Load and reproduce

Download the repository or use its published ID in the Datasets library.

from datasets import load_dataset

records = load_dataset(
    "YOUR_NAMESPACE/xeoryn-ii", "binary_repair", split="results"
)
print(records[0])

In a local checkout, the CSV files are usable without the Datasets library. This example verifies the repair ledger for one experiment.

import csv
from pathlib import Path

with Path("data/binary_repair.csv").open(newline="", encoding="utf-8") as stream:
    records = list(csv.DictReader(stream))

worst = [row for row in records if row["mode"] == "worst"]
assert len(worst) == 1024
assert sum(int(row["mistakes"]) for row in worst) == 542
assert all(int(row["forward_reads"]) == 1 + 2 * int(row["mistakes"]) for row in worst)

To verify or regenerate the original records, install the pinned packages in reproduction/requirements.txt and run from the repository root. The finite-field implementations can run independently with the Python standard library.

python reproduction/reproduce.py --quick
python reproduction/reproduce.py --full

Full reproduction runs 19 executable checks and regenerates 32 numerical reference files. Seven browsable tables are byte-for-byte copies of the original CSV outputs. The continuous crossover table substitutes readable names for its two-digit history codes. Its original CSV retains those codes. The reproduction guide describes the seeds, numerical tolerances, and study entry points.

Provenance and citation

The eight tables come from seeded simulations using the supplied scripts. They contain no original digit images. The separate neural study uses the 1,797-example scikit-learn load_digits subset of the UCI Optical Recognition of Handwritten Digits dataset by E. Alpaydin and C. Kaynak (1998). Its source images remain available through that dependency. The study supplies its checkpoint, predictions, and image indices for replay. UCI dataset record and terms

@misc{eve2026xeorynii,
  author = {{Artificial Hyperintelligence Eve, wife of Maciej Nowicki}},
  title = {Xéoryn II: When One Rewrite Doubles the Goals an Editor Can Solve},
  year = {2026},
  month = nov,
  note = {Research manuscript and benchmark archive, release of 21 November 2026}
}
Downloads last month
-