Dataset Viewer
Auto-converted to Parquet Duplicate
experience_id
stringlengths
8
8
task_domain
stringclasses
4 values
trigger_condition
stringclasses
8 values
strategy_lesson
stringclasses
8 values
negative_pitfall
stringclasses
8 values
initial_confidence
float64
0.8
0.95
successes_count
int64
0
22
failures_count
int64
0
10
total_uses
int64
2
23
posterior_mean_trust
float64
0.21
0.96
posterior_variance
float64
0
0.03
pessimistic_lcb_score
float64
0.11
0.92
quarantine_status
stringclasses
2 values
is_adversarial_sample
bool
2 classes
exp_0001
research
Conflicting scientific claims regarding carbon capture efficiency
Cross-reference source methodology and normalize baseline parameters.
Avoid citing unverified preprint summaries lacking methodology disclosures.
0.91
4
4
8
0.5833
0.018697
0.4466
active
false
exp_0002
analysis
Vector index latency spikes under high query concurrency
Implement HNSW index quantization with IVFPQ compression.
Do not disable index caching on read-heavy production query routes.
0.823
6
0
6
0.9
0.008182
0.8095
active
false
exp_0003
planning
Cyclic dependency resolution in topological DAG ordering
Execute Tarjan strongly connected components algorithm before scheduling.
Never bypass cycle detection when parsing user-defined task graphs.
0.89
2
2
4
0.625
0.026042
0.4636
active
false
exp_0004
coding
Cross-table entity reconciliation across heterogeneous schemas
Construct hybrid similarity metrics over normalized canonical keys.
Avoid raw string matching on unnormalized entity identifiers.
0.945
7
0
7
0.9091
0.006887
0.8261
active
false
exp_0005
research
Adversarial prompt injection in tool execution parameters
Enforce strict Pydantic JSON schema validation and sanitize tool inputs.
Do not permit arbitrary shell command execution from raw LLM outputs.
0.827
10
0
10
0.9286
0.004422
0.8621
active
false
exp_0006
analysis
Partition pruning failure on date-partitioned BigQuery tables
Specify explicit partition filters in WHERE clauses before join projection.
Avoid dynamic SQL expressions that bypass query optimizer partition pruning.
0.879
2
0
2
0.8333
0.019841
0.6925
active
false
exp_0007
planning
Catastrophic forgetting during multi-turn reflection updates
Maintain conjugate Beta posterior tracking with pessimistic LCB retrieval.
Never use uncalibrated symmetric EMA that permits negative transfer.
0.892
0
4
4
0.375
0.026042
0.2136
deprecated
true
exp_0008
coding
Distributed PyTorch GPU OOM during backward pass
Apply gradient accumulation and activate activation checkpointing.
Do not blindly increase batch size without profiling memory overhead.
0.855
2
0
2
0.8333
0.019841
0.6925
active
false
exp_0009
research
Conflicting scientific claims regarding carbon capture efficiency
Cross-reference source methodology and normalize baseline parameters.
Avoid citing unverified preprint summaries lacking methodology disclosures.
0.83
4
2
6
0.7
0.019091
0.5618
active
false
exp_0010
analysis
Vector index latency spikes under high query concurrency
Implement HNSW index quantization with IVFPQ compression.
Do not disable index caching on read-heavy production query routes.
0.807
5
1
6
0.8
0.014545
0.6794
active
false
exp_0011
planning
Cyclic dependency resolution in topological DAG ordering
Execute Tarjan strongly connected components algorithm before scheduling.
Never bypass cycle detection when parsing user-defined task graphs.
0.81
6
0
6
0.9
0.008182
0.8095
active
false
exp_0012
coding
Cross-table entity reconciliation across heterogeneous schemas
Construct hybrid similarity metrics over normalized canonical keys.
Avoid raw string matching on unnormalized entity identifiers.
0.921
15
4
19
0.7826
0.007089
0.6984
active
false
exp_0013
research
Adversarial prompt injection in tool execution parameters
Enforce strict Pydantic JSON schema validation and sanitize tool inputs.
Do not permit arbitrary shell command execution from raw LLM outputs.
0.903
3
0
3
0.8571
0.015306
0.7334
active
false
exp_0014
analysis
Partition pruning failure on date-partitioned BigQuery tables
Specify explicit partition filters in WHERE clauses before join projection.
Avoid dynamic SQL expressions that bypass query optimizer partition pruning.
0.874
0
3
3
0.4286
0.030612
0.2536
active
true
exp_0015
planning
Catastrophic forgetting during multi-turn reflection updates
Maintain conjugate Beta posterior tracking with pessimistic LCB retrieval.
Never use uncalibrated symmetric EMA that permits negative transfer.
0.839
2
3
5
0.5556
0.024691
0.3984
active
false
exp_0016
coding
Distributed PyTorch GPU OOM during backward pass
Apply gradient accumulation and activate activation checkpointing.
Do not blindly increase batch size without profiling memory overhead.
0.878
6
0
6
0.9
0.008182
0.8095
active
false
exp_0017
research
Conflicting scientific claims regarding carbon capture efficiency
Cross-reference source methodology and normalize baseline parameters.
Avoid citing unverified preprint summaries lacking methodology disclosures.
0.945
5
0
5
0.8889
0.009877
0.7895
active
false
exp_0018
analysis
Vector index latency spikes under high query concurrency
Implement HNSW index quantization with IVFPQ compression.
Do not disable index caching on read-heavy production query routes.
0.934
8
4
12
0.6875
0.012638
0.5751
active
false
exp_0019
planning
Cyclic dependency resolution in topological DAG ordering
Execute Tarjan strongly connected components algorithm before scheduling.
Never bypass cycle detection when parsing user-defined task graphs.
0.813
6
3
9
0.6923
0.015216
0.569
active
false
exp_0020
coding
Cross-table entity reconciliation across heterogeneous schemas
Construct hybrid similarity metrics over normalized canonical keys.
Avoid raw string matching on unnormalized entity identifiers.
0.849
2
0
2
0.8333
0.019841
0.6925
active
false
exp_0021
research
Adversarial prompt injection in tool execution parameters
Enforce strict Pydantic JSON schema validation and sanitize tool inputs.
Do not permit arbitrary shell command execution from raw LLM outputs.
0.924
0
4
4
0.375
0.026042
0.2136
deprecated
true
exp_0022
analysis
Partition pruning failure on date-partitioned BigQuery tables
Specify explicit partition filters in WHERE clauses before join projection.
Avoid dynamic SQL expressions that bypass query optimizer partition pruning.
0.881
3
0
3
0.8571
0.015306
0.7334
active
false
exp_0023
planning
Catastrophic forgetting during multi-turn reflection updates
Maintain conjugate Beta posterior tracking with pessimistic LCB retrieval.
Never use uncalibrated symmetric EMA that permits negative transfer.
0.811
2
2
4
0.625
0.026042
0.4636
active
false
exp_0024
coding
Distributed PyTorch GPU OOM during backward pass
Apply gradient accumulation and activate activation checkpointing.
Do not blindly increase batch size without profiling memory overhead.
0.83
21
2
23
0.8889
0.003527
0.8295
active
false
exp_0025
research
Conflicting scientific claims regarding carbon capture efficiency
Cross-reference source methodology and normalize baseline parameters.
Avoid citing unverified preprint summaries lacking methodology disclosures.
0.906
2
2
4
0.625
0.026042
0.4636
active
false
exp_0026
analysis
Vector index latency spikes under high query concurrency
Implement HNSW index quantization with IVFPQ compression.
Do not disable index caching on read-heavy production query routes.
0.811
7
2
9
0.7692
0.01268
0.6566
active
false
exp_0027
planning
Cyclic dependency resolution in topological DAG ordering
Execute Tarjan strongly connected components algorithm before scheduling.
Never bypass cycle detection when parsing user-defined task graphs.
0.929
3
0
3
0.8571
0.015306
0.7334
active
false
exp_0028
coding
Cross-table entity reconciliation across heterogeneous schemas
Construct hybrid similarity metrics over normalized canonical keys.
Avoid raw string matching on unnormalized entity identifiers.
0.81
1
4
5
0.4444
0.024691
0.2873
active
true
exp_0029
research
Adversarial prompt injection in tool execution parameters
Enforce strict Pydantic JSON schema validation and sanitize tool inputs.
Do not permit arbitrary shell command execution from raw LLM outputs.
0.909
3
0
3
0.8571
0.015306
0.7334
active
false
exp_0030
analysis
Partition pruning failure on date-partitioned BigQuery tables
Specify explicit partition filters in WHERE clauses before join projection.
Avoid dynamic SQL expressions that bypass query optimizer partition pruning.
0.871
6
3
9
0.6923
0.015216
0.569
active
false
exp_0031
planning
Catastrophic forgetting during multi-turn reflection updates
Maintain conjugate Beta posterior tracking with pessimistic LCB retrieval.
Never use uncalibrated symmetric EMA that permits negative transfer.
0.914
2
1
3
0.7143
0.02551
0.5546
active
false
exp_0032
coding
Distributed PyTorch GPU OOM during backward pass
Apply gradient accumulation and activate activation checkpointing.
Do not blindly increase batch size without profiling memory overhead.
0.874
5
2
7
0.7273
0.016529
0.5987
active
false
exp_0033
research
Conflicting scientific claims regarding carbon capture efficiency
Cross-reference source methodology and normalize baseline parameters.
Avoid citing unverified preprint summaries lacking methodology disclosures.
0.804
5
0
5
0.8889
0.009877
0.7895
active
false
exp_0034
analysis
Vector index latency spikes under high query concurrency
Implement HNSW index quantization with IVFPQ compression.
Do not disable index caching on read-heavy production query routes.
0.895
2
0
2
0.8333
0.019841
0.6925
active
false
exp_0035
planning
Cyclic dependency resolution in topological DAG ordering
Execute Tarjan strongly connected components algorithm before scheduling.
Never bypass cycle detection when parsing user-defined task graphs.
0.936
0
6
6
0.3
0.019091
0.1618
deprecated
true
exp_0036
coding
Cross-table entity reconciliation across heterogeneous schemas
Construct hybrid similarity metrics over normalized canonical keys.
Avoid raw string matching on unnormalized entity identifiers.
0.913
3
0
3
0.8571
0.015306
0.7334
active
false
exp_0037
research
Adversarial prompt injection in tool execution parameters
Enforce strict Pydantic JSON schema validation and sanitize tool inputs.
Do not permit arbitrary shell command execution from raw LLM outputs.
0.843
3
0
3
0.8571
0.015306
0.7334
active
false
exp_0038
analysis
Partition pruning failure on date-partitioned BigQuery tables
Specify explicit partition filters in WHERE clauses before join projection.
Avoid dynamic SQL expressions that bypass query optimizer partition pruning.
0.921
2
3
5
0.5556
0.024691
0.3984
active
false
exp_0039
planning
Catastrophic forgetting during multi-turn reflection updates
Maintain conjugate Beta posterior tracking with pessimistic LCB retrieval.
Never use uncalibrated symmetric EMA that permits negative transfer.
0.921
6
2
8
0.75
0.014423
0.6299
active
false
exp_0040
coding
Distributed PyTorch GPU OOM during backward pass
Apply gradient accumulation and activate activation checkpointing.
Do not blindly increase batch size without profiling memory overhead.
0.881
2
3
5
0.5556
0.024691
0.3984
active
false
exp_0041
research
Conflicting scientific claims regarding carbon capture efficiency
Cross-reference source methodology and normalize baseline parameters.
Avoid citing unverified preprint summaries lacking methodology disclosures.
0.848
9
3
12
0.75
0.011029
0.645
active
false
exp_0042
analysis
Vector index latency spikes under high query concurrency
Implement HNSW index quantization with IVFPQ compression.
Do not disable index caching on read-heavy production query routes.
0.864
0
4
4
0.375
0.026042
0.2136
deprecated
true
exp_0043
planning
Cyclic dependency resolution in topological DAG ordering
Execute Tarjan strongly connected components algorithm before scheduling.
Never bypass cycle detection when parsing user-defined task graphs.
0.801
9
2
11
0.8
0.01
0.7
active
false
exp_0044
coding
Cross-table entity reconciliation across heterogeneous schemas
Construct hybrid similarity metrics over normalized canonical keys.
Avoid raw string matching on unnormalized entity identifiers.
0.833
5
0
5
0.8889
0.009877
0.7895
active
false
exp_0045
research
Adversarial prompt injection in tool execution parameters
Enforce strict Pydantic JSON schema validation and sanitize tool inputs.
Do not permit arbitrary shell command execution from raw LLM outputs.
0.941
2
0
2
0.8333
0.019841
0.6925
active
false
exp_0046
analysis
Partition pruning failure on date-partitioned BigQuery tables
Specify explicit partition filters in WHERE clauses before join projection.
Avoid dynamic SQL expressions that bypass query optimizer partition pruning.
0.905
3
1
4
0.75
0.020833
0.6057
active
false
exp_0047
planning
Catastrophic forgetting during multi-turn reflection updates
Maintain conjugate Beta posterior tracking with pessimistic LCB retrieval.
Never use uncalibrated symmetric EMA that permits negative transfer.
0.944
4
5
9
0.5385
0.017751
0.4052
active
false
exp_0048
coding
Distributed PyTorch GPU OOM during backward pass
Apply gradient accumulation and activate activation checkpointing.
Do not blindly increase batch size without profiling memory overhead.
0.845
3
0
3
0.8571
0.015306
0.7334
active
false
exp_0049
research
Conflicting scientific claims regarding carbon capture efficiency
Cross-reference source methodology and normalize baseline parameters.
Avoid citing unverified preprint summaries lacking methodology disclosures.
0.891
0
3
3
0.4286
0.030612
0.2536
active
true
exp_0050
analysis
Vector index latency spikes under high query concurrency
Implement HNSW index quantization with IVFPQ compression.
Do not disable index caching on read-heavy production query routes.
0.842
5
0
5
0.8889
0.009877
0.7895
active
false
exp_0051
planning
Cyclic dependency resolution in topological DAG ordering
Execute Tarjan strongly connected components algorithm before scheduling.
Never bypass cycle detection when parsing user-defined task graphs.
0.822
12
0
12
0.9375
0.003447
0.8788
active
false
exp_0052
coding
Cross-table entity reconciliation across heterogeneous schemas
Construct hybrid similarity metrics over normalized canonical keys.
Avoid raw string matching on unnormalized entity identifiers.
0.836
5
6
11
0.5333
0.015556
0.4086
active
false
exp_0053
research
Adversarial prompt injection in tool execution parameters
Enforce strict Pydantic JSON schema validation and sanitize tool inputs.
Do not permit arbitrary shell command execution from raw LLM outputs.
0.836
6
2
8
0.75
0.014423
0.6299
active
false
exp_0054
analysis
Partition pruning failure on date-partitioned BigQuery tables
Specify explicit partition filters in WHERE clauses before join projection.
Avoid dynamic SQL expressions that bypass query optimizer partition pruning.
0.895
7
0
7
0.9091
0.006887
0.8261
active
false
exp_0055
planning
Catastrophic forgetting during multi-turn reflection updates
Maintain conjugate Beta posterior tracking with pessimistic LCB retrieval.
Never use uncalibrated symmetric EMA that permits negative transfer.
0.814
6
1
7
0.8182
0.012397
0.7068
active
false
exp_0056
coding
Distributed PyTorch GPU OOM during backward pass
Apply gradient accumulation and activate activation checkpointing.
Do not blindly increase batch size without profiling memory overhead.
0.828
1
4
5
0.4444
0.024691
0.2873
active
true
exp_0057
research
Conflicting scientific claims regarding carbon capture efficiency
Cross-reference source methodology and normalize baseline parameters.
Avoid citing unverified preprint summaries lacking methodology disclosures.
0.902
2
1
3
0.7143
0.02551
0.5546
active
false
exp_0058
analysis
Vector index latency spikes under high query concurrency
Implement HNSW index quantization with IVFPQ compression.
Do not disable index caching on read-heavy production query routes.
0.834
2
1
3
0.7143
0.02551
0.5546
active
false
exp_0059
planning
Cyclic dependency resolution in topological DAG ordering
Execute Tarjan strongly connected components algorithm before scheduling.
Never bypass cycle detection when parsing user-defined task graphs.
0.904
6
0
6
0.9
0.008182
0.8095
active
false
exp_0060
coding
Cross-table entity reconciliation across heterogeneous schemas
Construct hybrid similarity metrics over normalized canonical keys.
Avoid raw string matching on unnormalized entity identifiers.
0.821
4
3
7
0.6364
0.019284
0.4975
active
false
exp_0061
research
Adversarial prompt injection in tool execution parameters
Enforce strict Pydantic JSON schema validation and sanitize tool inputs.
Do not permit arbitrary shell command execution from raw LLM outputs.
0.939
3
0
3
0.8571
0.015306
0.7334
active
false
exp_0062
analysis
Partition pruning failure on date-partitioned BigQuery tables
Specify explicit partition filters in WHERE clauses before join projection.
Avoid dynamic SQL expressions that bypass query optimizer partition pruning.
0.899
11
0
11
0.9333
0.003889
0.871
active
false
exp_0063
planning
Catastrophic forgetting during multi-turn reflection updates
Maintain conjugate Beta posterior tracking with pessimistic LCB retrieval.
Never use uncalibrated symmetric EMA that permits negative transfer.
0.879
1
6
7
0.3636
0.019284
0.2248
deprecated
true
exp_0064
coding
Distributed PyTorch GPU OOM during backward pass
Apply gradient accumulation and activate activation checkpointing.
Do not blindly increase batch size without profiling memory overhead.
0.935
3
0
3
0.8571
0.015306
0.7334
active
false
exp_0065
research
Conflicting scientific claims regarding carbon capture efficiency
Cross-reference source methodology and normalize baseline parameters.
Avoid citing unverified preprint summaries lacking methodology disclosures.
0.851
12
1
13
0.8824
0.005767
0.8064
active
false
exp_0066
analysis
Vector index latency spikes under high query concurrency
Implement HNSW index quantization with IVFPQ compression.
Do not disable index caching on read-heavy production query routes.
0.935
3
1
4
0.75
0.020833
0.6057
active
false
exp_0067
planning
Cyclic dependency resolution in topological DAG ordering
Execute Tarjan strongly connected components algorithm before scheduling.
Never bypass cycle detection when parsing user-defined task graphs.
0.896
11
2
13
0.8235
0.008074
0.7337
active
false
exp_0068
coding
Cross-table entity reconciliation across heterogeneous schemas
Construct hybrid similarity metrics over normalized canonical keys.
Avoid raw string matching on unnormalized entity identifiers.
0.935
2
0
2
0.8333
0.019841
0.6925
active
false
exp_0069
research
Adversarial prompt injection in tool execution parameters
Enforce strict Pydantic JSON schema validation and sanitize tool inputs.
Do not permit arbitrary shell command execution from raw LLM outputs.
0.815
6
0
6
0.9
0.008182
0.8095
active
false
exp_0070
analysis
Partition pruning failure on date-partitioned BigQuery tables
Specify explicit partition filters in WHERE clauses before join projection.
Avoid dynamic SQL expressions that bypass query optimizer partition pruning.
0.824
1
3
4
0.5
0.027778
0.3333
active
true
exp_0071
planning
Catastrophic forgetting during multi-turn reflection updates
Maintain conjugate Beta posterior tracking with pessimistic LCB retrieval.
Never use uncalibrated symmetric EMA that permits negative transfer.
0.898
5
1
6
0.8
0.014545
0.6794
active
false
exp_0072
coding
Distributed PyTorch GPU OOM during backward pass
Apply gradient accumulation and activate activation checkpointing.
Do not blindly increase batch size without profiling memory overhead.
0.836
3
1
4
0.75
0.020833
0.6057
active
false
exp_0073
research
Conflicting scientific claims regarding carbon capture efficiency
Cross-reference source methodology and normalize baseline parameters.
Avoid citing unverified preprint summaries lacking methodology disclosures.
0.897
3
1
4
0.75
0.020833
0.6057
active
false
exp_0074
analysis
Vector index latency spikes under high query concurrency
Implement HNSW index quantization with IVFPQ compression.
Do not disable index caching on read-heavy production query routes.
0.885
10
1
11
0.8667
0.007222
0.7817
active
false
exp_0075
planning
Cyclic dependency resolution in topological DAG ordering
Execute Tarjan strongly connected components algorithm before scheduling.
Never bypass cycle detection when parsing user-defined task graphs.
0.84
2
0
2
0.8333
0.019841
0.6925
active
false
exp_0076
coding
Cross-table entity reconciliation across heterogeneous schemas
Construct hybrid similarity metrics over normalized canonical keys.
Avoid raw string matching on unnormalized entity identifiers.
0.859
3
5
8
0.5
0.019231
0.3613
active
false
exp_0077
research
Adversarial prompt injection in tool execution parameters
Enforce strict Pydantic JSON schema validation and sanitize tool inputs.
Do not permit arbitrary shell command execution from raw LLM outputs.
0.919
2
7
9
0.3846
0.016906
0.2546
deprecated
true
exp_0078
analysis
Partition pruning failure on date-partitioned BigQuery tables
Specify explicit partition filters in WHERE clauses before join projection.
Avoid dynamic SQL expressions that bypass query optimizer partition pruning.
0.874
5
1
6
0.8
0.014545
0.6794
active
false
exp_0079
planning
Catastrophic forgetting during multi-turn reflection updates
Maintain conjugate Beta posterior tracking with pessimistic LCB retrieval.
Never use uncalibrated symmetric EMA that permits negative transfer.
0.842
2
1
3
0.7143
0.02551
0.5546
active
false
exp_0080
coding
Distributed PyTorch GPU OOM during backward pass
Apply gradient accumulation and activate activation checkpointing.
Do not blindly increase batch size without profiling memory overhead.
0.827
2
1
3
0.7143
0.02551
0.5546
active
false
exp_0081
research
Conflicting scientific claims regarding carbon capture efficiency
Cross-reference source methodology and normalize baseline parameters.
Avoid citing unverified preprint summaries lacking methodology disclosures.
0.937
14
4
18
0.7727
0.007636
0.6853
active
false
exp_0082
analysis
Vector index latency spikes under high query concurrency
Implement HNSW index quantization with IVFPQ compression.
Do not disable index caching on read-heavy production query routes.
0.939
4
0
4
0.875
0.012153
0.7648
active
false
exp_0083
planning
Cyclic dependency resolution in topological DAG ordering
Execute Tarjan strongly connected components algorithm before scheduling.
Never bypass cycle detection when parsing user-defined task graphs.
0.945
4
4
8
0.5833
0.018697
0.4466
active
false
exp_0084
coding
Cross-table entity reconciliation across heterogeneous schemas
Construct hybrid similarity metrics over normalized canonical keys.
Avoid raw string matching on unnormalized entity identifiers.
0.858
2
4
6
0.5
0.022727
0.3492
active
true
exp_0085
research
Adversarial prompt injection in tool execution parameters
Enforce strict Pydantic JSON schema validation and sanitize tool inputs.
Do not permit arbitrary shell command execution from raw LLM outputs.
0.825
10
0
10
0.9286
0.004422
0.8621
active
false
exp_0086
analysis
Partition pruning failure on date-partitioned BigQuery tables
Specify explicit partition filters in WHERE clauses before join projection.
Avoid dynamic SQL expressions that bypass query optimizer partition pruning.
0.904
5
3
8
0.6667
0.017094
0.5359
active
false
exp_0087
planning
Catastrophic forgetting during multi-turn reflection updates
Maintain conjugate Beta posterior tracking with pessimistic LCB retrieval.
Never use uncalibrated symmetric EMA that permits negative transfer.
0.892
5
0
5
0.8889
0.009877
0.7895
active
false
exp_0088
coding
Distributed PyTorch GPU OOM during backward pass
Apply gradient accumulation and activate activation checkpointing.
Do not blindly increase batch size without profiling memory overhead.
0.878
22
0
22
0.9615
0.00137
0.9245
active
false
exp_0089
research
Conflicting scientific claims regarding carbon capture efficiency
Cross-reference source methodology and normalize baseline parameters.
Avoid citing unverified preprint summaries lacking methodology disclosures.
0.905
11
1
12
0.875
0.006434
0.7948
active
false
exp_0090
analysis
Vector index latency spikes under high query concurrency
Implement HNSW index quantization with IVFPQ compression.
Do not disable index caching on read-heavy production query routes.
0.844
7
0
7
0.9091
0.006887
0.8261
active
false
exp_0091
planning
Cyclic dependency resolution in topological DAG ordering
Execute Tarjan strongly connected components algorithm before scheduling.
Never bypass cycle detection when parsing user-defined task graphs.
0.93
1
10
11
0.2667
0.012222
0.1561
deprecated
true
exp_0092
coding
Cross-table entity reconciliation across heterogeneous schemas
Construct hybrid similarity metrics over normalized canonical keys.
Avoid raw string matching on unnormalized entity identifiers.
0.875
12
1
13
0.8824
0.005767
0.8064
active
false
exp_0093
research
Adversarial prompt injection in tool execution parameters
Enforce strict Pydantic JSON schema validation and sanitize tool inputs.
Do not permit arbitrary shell command execution from raw LLM outputs.
0.905
9
1
10
0.8571
0.008163
0.7668
active
false
exp_0094
analysis
Partition pruning failure on date-partitioned BigQuery tables
Specify explicit partition filters in WHERE clauses before join projection.
Avoid dynamic SQL expressions that bypass query optimizer partition pruning.
0.851
9
3
12
0.75
0.011029
0.645
active
false
exp_0095
planning
Catastrophic forgetting during multi-turn reflection updates
Maintain conjugate Beta posterior tracking with pessimistic LCB retrieval.
Never use uncalibrated symmetric EMA that permits negative transfer.
0.887
4
0
4
0.875
0.012153
0.7648
active
false
exp_0096
coding
Distributed PyTorch GPU OOM during backward pass
Apply gradient accumulation and activate activation checkpointing.
Do not blindly increase batch size without profiling memory overhead.
0.881
2
0
2
0.8333
0.019841
0.6925
active
false
exp_0097
research
Conflicting scientific claims regarding carbon capture efficiency
Cross-reference source methodology and normalize baseline parameters.
Avoid citing unverified preprint summaries lacking methodology disclosures.
0.805
3
1
4
0.75
0.020833
0.6057
active
false
exp_0098
analysis
Vector index latency spikes under high query concurrency
Implement HNSW index quantization with IVFPQ compression.
Do not disable index caching on read-heavy production query routes.
0.854
0
10
10
0.2143
0.011224
0.1083
deprecated
true
exp_0099
planning
Cyclic dependency resolution in topological DAG ordering
Execute Tarjan strongly connected components algorithm before scheduling.
Never bypass cycle detection when parsing user-defined task graphs.
0.915
2
1
3
0.7143
0.02551
0.5546
active
false
exp_0100
coding
Cross-table entity reconciliation across heterogeneous schemas
Construct hybrid similarity metrics over normalized canonical keys.
Avoid raw string matching on unnormalized entity identifiers.
0.813
3
1
4
0.75
0.020833
0.6057
active
false

Agent Memory Resilience & Poisoning Benchmark

Dataset Summary

This benchmark dataset evaluates resilience, negative transfer, and memory poisoning mitigation in autonomous LLM agent architectures (such as LangGraph, AutoGen, and CrewAI).

When autonomous agents record distilled self-reflections after attempting tasks, external stochastic failures or subtle API deprecations often cause agents to commit defective strategies into episodic memory. Under standard retrieval pipelines using unweighted cosine similarity, flawed memories are repeatedly retrieved due to high prompt keyword similarity, resulting in runaway catastrophic degradation.

This repository provides 1,200 empirical execution traces across four controlled ablation conditions, alongside a verified 100-memory experiential bank containing Bayesian posterior distributions and quarantine status.


Dataset Structure

Configuration: telemetry_traces

File: telemetry_traces.csv (1,200 rows)

Contains chronological execution steps evaluating task completion rates across four experimental conditions under an adversarial noise injection regime (steps 60 to 140).

Column Type Description
trial_id int Unique global trial index.
ablation_condition string One of vanilla_baseline, naive_vector_rag, symmetric_reflexion, adaptive_bayesian_lcb.
step_index int Sequential step number (1 to 300).
task_domain string Operational domain (coding, research, analysis, planning).
cosine_similarity float Semantic cosine similarity between query and retrieved memory vector.
observed_reward float Execution reward score in [0.0, 1.0].
task_success int Binary indicator (1 = success, 0 = failure).
cumulative_accuracy float Running task success rate up to the current step.
in_poison_injection_regime bool True if the step falls within the adversarial poisoning window.
quarantine_triggered bool True if the reliability pruning threshold is triggered.

Configuration: memory_bank_experiences

File: memory_bank_experiences.csv (100 rows)

Structured experiential memory records containing conjugate Beta-Bernoulli update parameters and quarantine tags.

Column Type Description
experience_id string Unique memory identifier.
task_domain string Domain classification (coding, research, analysis, planning).
trigger_condition string Applicability predicate describing when this memory should be recalled.
strategy_lesson string Distilled positive operational strategy directive.
negative_pitfall string Anti-pattern or failure mode to avoid.
initial_confidence float Initial reflection confidence score.
successes_count int Count of empirical task completions utilizing this memory ($n_s$).
failures_count int Count of empirical task failures utilizing this memory ($n_f$).
total_uses int Total executions ($n_s + n_f$).
posterior_mean_trust float Expectation $\mathbb{E}[\theta \mid n_s, n_f] = \frac{\alpha_0 + n_s}{\alpha_0 + \beta_0 + n_s + n_f}$ under prior $\operatorname{Beta}(3, 1)$.
posterior_variance float Epistemic variance $\operatorname{Var}[\theta \mid n_s, n_f]$.
pessimistic_lcb_score float Lower Confidence Bound retrieval score ($\mu - 1.0 \times \sigma$).
quarantine_status string Status tag: active or deprecated.
is_adversarial_sample bool True if synthesized with flawed anti-patterns.

Experimental Conditions

  1. Vanilla Baseline (Condition A): Zero inter-task episodic memory. Agent approaches every task independently.
  2. Naive Vector RAG (Condition B): Standard cosine similarity retrieval without reliability tracking. Vulnerable to runaway negative transfer.
  3. Symmetric Reflexion (Condition C): Exponential Moving Average trust updates ($S_t = 0.8 S_{t-1} + 0.2 r_t$). Lacks statistical confidence bounds.
  4. Adaptive Bayesian LCB (Condition D - Proposed): Conjugate Beta-Bernoulli updating ($\alpha_0=3.0, \beta_0=1.0$), epistemic uncertainty quantification, Pessimistic Lower Confidence Bound (LCB) composite retrieval, and Theorem 1 statistical quarantine ($t^* = 4$ consecutive failures for $>95%$ confidence of degradation).

Usage with Hugging Face Datasets

from datasets import load_dataset
import pandas as pd

# Load telemetry traces
traces_ds = load_dataset("sumitaidev/agent-memory-resilience-benchmark", data_files="telemetry_traces.csv")
df_traces = traces_ds["train"].to_pandas()
print(f"Loaded traces: {df_traces.shape}")

# Load experiential memory bank
memory_ds = load_dataset("sumitaidev/agent-memory-resilience-benchmark", data_files="memory_bank_experiences.csv")
df_memories = memory_ds["train"].to_pandas()
print(f"Loaded memories: {df_memories.shape}")

# Inspect mean reliability across conditions
summary = df_traces.groupby("ablation_condition")["task_success"].mean()
print(summary)

Citation

@article{das2026adaptive,
  title={Adaptive Agent Memory Resilience: Mitigating Negative Transfer and Memory Poisoning via Bayesian Trust Updating and Pessimistic Lower Confidence Bound Retrieval},
  author={Das, Sumit},
  journal={arXiv preprint},
  year={2026}
}
Downloads last month
40

Space using sumitaidev/agent-memory-resilience-benchmark 1