Datasets:

latkes
/

cybernetic-agents-validation-step2-failure-mapping

kind stringclasses 2 values	system stringclasses 7 values	mode stringclasses 18 values	value float64 -0.09 0.49	present bool 2 classes
rate	AG2/MathChat	1.1	0.3233	null
rate	AG2/MathChat	1.2	0.0134	null
rate	AG2/MathChat	1.3	0.3886	null
rate	AG2/MathChat	1.4	0.0754	null
rate	AG2/MathChat	1.5	0.2965	null
rate	AG2/MathChat	2.1	0.0503	null
rate	AG2/MathChat	2.2	0.2848	null
rate	AG2/MathChat	2.3	0.2161	null
rate	AG2/MathChat	2.4	0.0352	null
rate	AG2/MathChat	2.5	0	null
rate	AG2/MathChat	2.6	0.4941	null
rate	AG2/MathChat	3.1	0.2245	null
rate	AG2/MathChat	3.2	0.1876	null
rate	AG2/MathChat	3.3	0.2144	null
rate	AppWorld	1.1	0.2333	null
rate	AppWorld	1.2	0	null
rate	AppWorld	1.3	0.2	null
rate	AppWorld	1.4	0.0333	null
rate	AppWorld	1.5	0.0667	null
rate	AppWorld	2.1	0.0333	null
rate	AppWorld	2.2	0.0667	null
rate	AppWorld	2.3	0.0667	null
rate	AppWorld	2.4	0.0333	null
rate	AppWorld	2.5	0	null
rate	AppWorld	2.6	0.1333	null
rate	AppWorld	3.1	0.0333	null
rate	AppWorld	3.2	0.1	null
rate	AppWorld	3.3	0.2333	null
rate	ChatDev	1.1	0.2462	null
rate	ChatDev	1.2	0	null
rate	ChatDev	1.3	0.3615	null
rate	ChatDev	1.4	0.0385	null
rate	ChatDev	1.5	0.2923	null
rate	ChatDev	2.1	0.0308	null
rate	ChatDev	2.2	0.2077	null
rate	ChatDev	2.3	0.1462	null
rate	ChatDev	2.4	0.0308	null
rate	ChatDev	2.5	0	null
rate	ChatDev	2.6	0.2846	null
rate	ChatDev	3.1	0.1	null
rate	ChatDev	3.2	0.1692	null
rate	ChatDev	3.3	0.2538	null
rate	HyperAgent	1.1	0.2333	null
rate	HyperAgent	1.2	0	null
rate	HyperAgent	1.3	0.2	null
rate	HyperAgent	1.4	0.0333	null
rate	HyperAgent	1.5	0.0667	null
rate	HyperAgent	2.1	0.0333	null
rate	HyperAgent	2.2	0.0667	null
rate	HyperAgent	2.3	0.0667	null
rate	HyperAgent	2.4	0.0333	null
rate	HyperAgent	2.5	0	null
rate	HyperAgent	2.6	0.1333	null
rate	HyperAgent	3.1	0.0333	null
rate	HyperAgent	3.2	0.1	null
rate	HyperAgent	3.3	0.2333	null
rate	Magentic-One	1.1	0.3282	null
rate	Magentic-One	1.2	0.0103	null
rate	Magentic-One	1.3	0.3385	null
rate	Magentic-One	1.4	0.0462	null
rate	Magentic-One	1.5	0.2615	null
rate	Magentic-One	2.1	0.0308	null
rate	Magentic-One	2.2	0.2462	null
rate	Magentic-One	2.3	0.1744	null
rate	Magentic-One	2.4	0.0256	null
rate	Magentic-One	2.5	0	null
rate	Magentic-One	2.6	0.4154	null
rate	Magentic-One	3.1	0.1692	null
rate	Magentic-One	3.2	0.1641	null
rate	Magentic-One	3.3	0.2154	null
rate	MetaGPT	1.1	0.2478	null
rate	MetaGPT	1.2	0	null
rate	MetaGPT	1.3	0.3826	null
rate	MetaGPT	1.4	0.0391	null
rate	MetaGPT	1.5	0.3217	null
rate	MetaGPT	2.1	0.0304	null
rate	MetaGPT	2.2	0.2261	null
rate	MetaGPT	2.3	0.1565	null
rate	MetaGPT	2.4	0.0304	null
rate	MetaGPT	2.5	0	null
rate	MetaGPT	2.6	0.3043	null
rate	MetaGPT	3.1	0.1087	null
rate	MetaGPT	3.2	0.1783	null
rate	MetaGPT	3.3	0.2565	null
rate	OpenManus	1.1	0.2333	null
rate	OpenManus	1.2	0	null
rate	OpenManus	1.3	0.2	null
rate	OpenManus	1.4	0.0333	null
rate	OpenManus	1.5	0.0667	null
rate	OpenManus	2.1	0.0333	null
rate	OpenManus	2.2	0.0667	null
rate	OpenManus	2.3	0.0667	null
rate	OpenManus	2.4	0.0333	null
rate	OpenManus	2.5	0	null
rate	OpenManus	2.6	0.1333	null
rate	OpenManus	3.1	0.0333	null
rate	OpenManus	3.2	0.1	null
rate	OpenManus	3.3	0.2333	null
elevation	AG2/MathChat	S3*	0.0109	true
elevation	AppWorld	S3*	0.0434	false

End of preview. Expand in Data Studio

cybernetic-agents-validation-step2-failure-mapping

Step 2 (pre-registered, zero compute, plan=experiments/E3-step2-plan.md): do MAST-7 documented failures concentrate on the VSM functions a system lacks? RESULT: Delta_real=-0.01021 (negative — opposite of the paper's prediction), permutation p=0.6023 (threshold 0.05), Delta_SE_rival=-0.00061; verdict='NO analytic teeth beyond chance/SE'. Null as the frozen plan pre-registered (S5 zero cross-system variance) — no support for the paper's analytic-teeth claim. Rows: per-(system,mode) empirical MAST failure rate (kind=rate) + per-(system,VSM-function) elevation with present-flag (kind=elevation).

Dataset Info

Rows: 126
Columns: 5

Columns

Column	Type	Description
kind	Value('string')	'rate' = empirical MAST failure-mode rate for a system; 'elevation' = (in-group − out-group) rate diff for a VSM function
system	Value('string')	MAST-7 system (canonical name)
mode	Value('string')	MAST failure-mode id (kind=rate) or VSM function (kind=elevation)
value	Value('float64')	rate in [0,1] (kind=rate) or elevation diff (kind=elevation)
present	Value('bool')	for kind=elevation: did Step-1 score this VSM function reliably Present (both passes)? null for kind=rate

Generation Parameters

{
  "script_name": "scripts/step2_failure_mapping.py",
  "model": "n/a (statistical analysis of public mcemri/MAD)",
  "description": "Step 2 (pre-registered, zero compute, plan=experiments/E3-step2-plan.md): do MAST-7 documented failures concentrate on the VSM functions a system lacks? RESULT: Delta_real=-0.01021 (negative \u2014 opposite of the paper's prediction), permutation p=0.6023 (threshold 0.05), Delta_SE_rival=-0.00061; verdict='NO analytic teeth beyond chance/SE'. Null as the frozen plan pre-registered (S5 zero cross-system variance) \u2014 no support for the paper's analytic-teeth claim. Rows: per-(system,mode) empirical MAST failure rate (kind=rate) + per-(system,VSM-function) elevation with present-flag (kind=elevation).",
  "experiment_name": "cybernetic-agents-validation",
  "artifact_status": "final",
  "canary": false,
  "hyperparameters": {
    "n_traces": 1242,
    "n_systems": 7,
    "permutations": 10000,
    "seed": 20260516
  },
  "input_datasets": [
    "mcemri/MAD",
    "latkes/cybernetic-agents-validation-step1-characterization-v2-2pass"
  ]
}

Usage

from datasets import load_dataset

dataset = load_dataset("latkes/cybernetic-agents-validation-step2-failure-mapping", split="train")
print(f"Loaded {len(dataset)} rows")

Downloads last month: 26