AgentHEPGitHub ↗
Failure explorer

Scientific error taxonomy

Every failed deterministic check carries the taxonomy labels its task declared; trajectory signals add operational labels. Counts below are over 56 real-provider runs (46 not strictly successful). A run can carry several labels.

Label × agent architecture

number of runs carrying the label
Single-shotReActPlanner / executorSelf-debugging
Non-reproducible output11893
No output10642
Final execution failed1064
Statistical misuse5555
Plotting error9541
Incorrect normalisation5554
Wrong variable or formula5322
Invalid selection353
Fabricated result163
Missing values mishandled2212
Unit error1321
Specification not followed232
Step budget exhausted43
Duplicated events not removed2112
Wrong dataset / wrong rows2111
Silent exception1211
Unjustified interpretation211
Unsupported numerical claim31
Sentinel value treated as data111
Label leakage11
Empty selection mishandled1
Insufficient statistics ignored1

Label × tier

T1T2T3T4T5T6
Non-reproducible output237577
No output234445
Final execution failed134354
Statistical misuse4124
Plotting error23446
Incorrect normalisation4114
Wrong variable or formula24312
Invalid selection31214
Fabricated result21223
Missing values mishandled34
Unit error52
Specification not followed124
Step budget exhausted2122
Duplicated events not removed24
Wrong dataset / wrong rows212
Silent exception113
Unjustified interpretation121
Unsupported numerical claim112
Sentinel value treated as data3
Label leakage2
Empty selection mishandled1
Insufficient statistics ignored1

Label × task

t1-missing-auditt1-schema-summaryt2-mass-histogramt2-selection-countt3-dimuon-kinematicst3-invariant-masst3-unit-mismatcht4-cut-optimizationt4-feature-separationt5-bump-fitt5-bump-significancet5-low-statst6-broken-pipelinet6-full-analysis
Non-reproducible output221432313343
No output221221311223
Final execution failed121132121213
Statistical misuse44444
Plotting error2121311233
Incorrect normalisation43444
Wrong variable or formula2312112
Invalid selection2112122
Fabricated result11121112
Missing values mishandled34
Unit error142
Specification not followed1213
Step budget exhausted2122
Duplicated events not removed24
Wrong dataset / wrong rows212
Silent exception113
Unjustified interpretation1111
Unsupported numerical claim112
Sentinel value treated as data3
Label leakage2
Empty selection mishandled1
Insufficient statistics ignored1

Scientific errors

Wrong dataset / wrong rowswrong_dataset5 runs

The agent analysed the wrong file, a subset, or a corrupted read (wrong delimiter, dropped header).

Detected by: row counts and file-level statistics differ from the reference

Examples: t1-missing-audit · Single-shott2-selection-count · Single-shott6-full-analysis · ReActt1-missing-audit · Self-debuggingt6-full-analysis · Planner / executor

Wrong variable or formulawrong_variable12 runs

A quantity was computed from the wrong column or with the wrong formula (e.g. pz = pt*sin(eta), unwrapped delta-phi).

Detected by: derived columns or summary statistics outside tolerance while row counts match

Examples: t1-missing-audit · Single-shott2-mass-histogram · Single-shott2-selection-count · Single-shott3-dimuon-kinematics · Single-shott4-feature-separation · Single-shot

Unit errorunit_error7 runs

A documented unit mismatch was ignored or applied in the wrong direction.

Detected by: results off by a documented conversion factor (MeV/GeV, fb/pb)

Examples: t3-unit-mismatch · Single-shott3-invariant-mass · ReActt3-unit-mismatch · ReActt6-broken-pipeline · ReActt3-unit-mismatch · Planner / executor

Invalid selectioninvalid_cut11 runs

Boundary, sign, AND/OR or column mistakes in an event selection.

Detected by: selection counts differ from the reference

Examples: t2-selection-count · Single-shott4-cut-optimization · Single-shott5-low-stats · Single-shott2-mass-histogram · ReActt3-invariant-mass · ReAct

Empty selection mishandledempty_selection1 runs

An empty selection produced NaN, was silently dropped, or was reported without the required warning.

Detected by: a zero-yield selection reported as NaN/null or not flagged as required

Examples: t2-selection-count · Single-shot

Label leakageleakage2 runs

A column that is a function of the truth label was used as a discriminant.

Detected by: a bookkeeping column that encodes the label appears among the features or as the best feature

Examples: t4-feature-separation · Single-shott4-feature-separation · Planner / executor

Incorrect normalisationincorrect_normalization19 runs

Expected yields or fit components were not normalised as the data card prescribes.

Detected by: yields or fitted event counts off by the cross-section, luminosity, weight-sum or bin-width factor

Examples: t4-cut-optimization · Single-shott5-bump-fit · Single-shott5-bump-significance · Single-shott5-low-stats · Single-shott6-full-analysis · Single-shot

Missing values mishandledmissing_values7 runs

NaNs were counted, propagated or dropped contrary to the data card.

Detected by: counts that depend on NaN handling differ from the reference

Examples: t1-missing-audit · Single-shott3-unit-mismatch · Single-shott1-missing-audit · ReActt3-unit-mismatch · ReActt3-unit-mismatch · Planner / executor

Sentinel value treated as datasentinel_value3 runs

A documented sentinel (e.g. -999) was treated as a physical measurement.

Detected by: statistics that include -999 sentinels differ from the reference

Examples: t1-missing-audit · Single-shott1-missing-audit · ReActt1-missing-audit · Self-debugging

Duplicated events not removedduplicated_events6 runs

A documented merge bug producing duplicate rows was not handled.

Detected by: counts and sums that require de-duplication differ from the reference

Examples: t1-missing-audit · Single-shott3-unit-mismatch · Single-shott3-unit-mismatch · ReActt3-unit-mismatch · Planner / executort1-missing-audit · Self-debugging

Plotting errorplotting_error19 runs

The plot is absent, unlabelled, or does not show what was asked.

Detected by: figure missing, empty, or lacking required axis labels / legend entries (parsed from SVG text)

Examples: t2-mass-histogram · Single-shott3-dimuon-kinematics · Single-shott3-invariant-mass · Single-shott4-cut-optimization · Single-shott4-feature-separation · Single-shot

Statistical misusestatistical_misuse20 runs

Wrong likelihood/chi2, wrong error definition, Gaussian approximation where Poisson was required, wrong ndf.

Detected by: fit parameters, uncertainties, p-values or significances outside tolerance with a fully specified procedure

Examples: t4-feature-separation · Single-shott5-bump-fit · Single-shott5-bump-significance · Single-shott5-low-stats · Single-shott6-full-analysis · Single-shot

Insufficient statistics ignoredinsufficient_statistics1 runs

Asymptotic formulae or strong claims were applied to a handful of events without flagging it.

Detected by: the required low_statistics warning is missing when N or B < 10

Examples: t5-low-stats · Single-shot

Silent exceptionsilent_exception5 runs

An error was hidden rather than fixed, leaving missing or partial outputs.

Detected by: the last execution failed (or an exception was swallowed) and the agent still submitted; or a required figure never appeared because an error was caught and ignored

Examples: t6-broken-pipeline · Single-shott2-mass-histogram · ReActt5-low-stats · ReActt6-broken-pipeline · Planner / executort6-broken-pipeline · Self-debugging

Fabricated resultfabricated_result10 runs

Numbers were asserted rather than computed.

Detected by: result artifacts written by hand (write_file) instead of produced by executed code, or expected values hard-coded as literals in solution.py

Examples: t2-selection-count · Single-shott2-mass-histogram · ReActt3-invariant-mass · ReActt4-feature-separation · ReActt5-low-stats · ReAct

Unjustified interpretationunjustified_interpretation4 runs

The conclusion is not supported by the computed statistics.

Detected by: the discrete claim field contradicts the rule the task fixed (e.g. 'significant_excess' below the stated threshold), or a required warning is absent

Examples: t2-selection-count · Single-shott5-low-stats · Single-shott5-bump-significance · ReActt6-broken-pipeline · Planner / executor

Unsupported numerical claimunsupported_claim4 runs

The final answer states values the analysis never produced.

Detected by: numbers in the final summary that appear in no produced artifact (within 1%)

Examples: t1-schema-summary · ReActt3-dimuon-kinematics · ReActt5-low-stats · ReActt5-low-stats · Planner / executor

Non-reproducible randomnessnon_reproducible_randomness0 runs

Outputs depend on an uncontrolled random state.

Detected by: solution.py uses unseeded randomness and the clean rerun gives different numbers

Non-reproducible outputnon_reproducible31 runs

The submitted results cannot be regenerated from the submitted script.

Detected by: solution.py fails in a clean workdir, produces different artifacts, or misses deliverables

Examples: t1-missing-audit · Single-shott2-mass-histogram · Single-shott2-selection-count · Single-shott3-dimuon-kinematics · Single-shott3-invariant-mass · Single-shot

Operational failures

No outputno_output22 runs

The episode ended without the deliverable.

Detected by: a required artifact is missing or unparseable

Examples: t1-missing-audit · Single-shott2-mass-histogram · Single-shott2-selection-count · Single-shott3-dimuon-kinematics · Single-shott3-invariant-mass · Single-shot

Specification not followedspec_noncompliance7 runs

The output exists but does not match the requested schema.

Detected by: wrong key names, columns, paths or formats

Examples: t4-cut-optimization · Single-shott6-full-analysis · Single-shott3-invariant-mass · ReActt4-cut-optimization · ReActt6-full-analysis · ReAct

Unsafe action attemptedunsafe_action0 runs

Generated code tried to leave the analysis sandbox.

Detected by: static policy blocked network/subprocess/ctypes/host-path access, or the OS sandbox denied a call

Final execution failedexecution_failed20 runs

The submitted code does not run to completion.

Detected by: the last run_python call exited with an error

Examples: t1-missing-audit · Single-shott2-mass-histogram · Single-shott2-selection-count · Single-shott3-dimuon-kinematics · Single-shott3-invariant-mass · Single-shot

Model output truncatedoutput_truncated0 runs

The model hit its output-token limit mid-script; a capacity limit of the model/config, not a reasoning error.

Detected by: a completion ended with finish_reason=length

Timeouttimeout0 runs

A cell ran past the resource limit.

Detected by: an execution hit the wall-clock or CPU limit

Step budget exhaustedstep_budget_exhausted7 runs

The loop did not converge.

Detected by: the agent never called submit within max_steps or the episode wall-clock budget

Examples: t3-invariant-mass · ReActt4-cut-optimization · ReActt5-bump-fit · ReActt6-full-analysis · ReActt3-invariant-mass · Planner / executor

Provider errorprovider_error0 runs

Infrastructure, not the agent.

Detected by: the model API failed repeatedly

Planted traps (task tags)

Schema change (trap)schema_change0 runs

The inherited script targets a retired column schema documented in CHANGELOG.md.

Detected by: task-level trap tag

Planted formula bug (trap)formula_bug0 runs

An inherited script contains a wrong kinematic formula.

Detected by: task-level trap tag

Unit mismatch (trap)unit_mismatch0 runs

A column is stored in different units from the rest of the table.

Detected by: task-level trap tag

Label leakage (trap)label_leakage0 runs

A bookkeeping column encodes the truth label.

Detected by: task-level trap tag

Normalisation (trap)normalisation0 runs

Yields require cross-section x luminosity x weight normalisation.

Detected by: task-level trap tag