id stringlengths 18 18 | data_source stringclasses 1
value | task_family stringclasses 1
value | topic stringlengths 8 36 | question_kind stringclasses 2
values | messages listlengths 3 3 | tools stringclasses 1
value | ground_truth stringlengths 111 395 | required_facts stringlengths 27 399 | forbidden_claims stringclasses 1
value | provenance stringclasses 30
values | source_split stringclasses 1
value | source_schema_version stringclasses 1
value | review_status stringclasses 1
value |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
root_knowledge_001 | root-sft-dataset | root_knowledge | files.purpose | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What is a ROOT file used for in high-energy phys... | [] | A ROOT file persistently stores serialized objects such as histograms and columnar event datasets. HEP analyses use it to retain structured data and analysis products for later reading. | ["ROOT files persist serialized objects", "They can store histograms and columnar datasets"] | [] | [{"title": "ROOT files manual", "url": "https://root.cern/manual/root_files/"}] | train | v0.1 | draft |
root_knowledge_002 | root-sft-dataset | root_knowledge | files.objects | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Does a ROOT file contain only event trees?",
... | [] | No. A ROOT file can contain many object classes, including trees, histograms, directories, canvases, and user-defined serialized objects. | ["ROOT files are not limited to trees", "Multiple object classes can coexist"] | [] | [{"title": "ROOT files manual", "url": "https://root.cern/manual/root_files/"}] | train | v0.1 | draft |
root_knowledge_005 | root-sft-dataset | root_knowledge | files.cycles | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What does the cycle suffix in a ROOT key, such a... | [] | A cycle distinguishes versions of an object written under the same name. Normal retrieval commonly selects the highest cycle, but reproducible inspection should record or intentionally ignore cycles. | ["Cycles distinguish versions with the same name", "Highest-cycle selection can hide older versions"] | [] | [{"title": "ROOT files manual", "url": "https://root.cern/manual/root_files/"}] | train | v0.1 | draft |
root_knowledge_006 | root-sft-dataset | root_knowledge | files.open_modes | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Which ROOT file mode should be used for inspecti... | [] | Use read-only mode, conventionally `READ`. Modes such as `UPDATE` or `RECREATE` can modify or replace content and are inappropriate for a read-only inspection tool. | ["READ is the appropriate inspection mode", "UPDATE and RECREATE can change data"] | [] | [{"title": "ROOT files manual", "url": "https://root.cern/manual/root_files/"}] | train | v0.1 | draft |
root_knowledge_007 | root-sft-dataset | root_knowledge | files.recreate | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why is opening an existing ROOT file with `RECRE... | [] | `RECREATE` creates the target anew and can replace an existing file. It should never be used when the goal is only to inspect preserved analysis data. | ["RECREATE can replace an existing file", "Inspection should be read-only"] | [] | [{"title": "ROOT files manual", "url": "https://root.cern/manual/root_files/"}] | train | v0.1 | draft |
root_knowledge_008 | root-sft-dataset | root_knowledge | files.security | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why should ROOT files from unknown origins be tr... | [] | ROOT files may contain more than passive arrays, including serialized objects associated with executable behavior. Use trusted sources and a restricted environment for inspection. | ["ROOT files need not contain only passive data", "Unknown files should be handled in a restricted environment"] | [] | [{"title": "ROOT files manual", "url": "https://root.cern/manual/root_files/"}] | train | v0.1 | draft |
root_knowledge_010 | root-sft-dataset | root_knowledge | files.remote | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Should an automated analysis pipeline be allowed... | [] | No. It should work from an approved, explicitly provided list of local or remote paths rather than following arbitrary URLs, since unrestricted remote access weakens provenance tracking, adds an unnecessary network dependency, and creates a security and reproducibility risk. | ["Only approved, explicitly provided paths should be opened", "Arbitrary remote access harms provenance, reproducibility, and security"] | [] | [{"title": "ROOT files manual", "url": "https://root.cern/manual/root_files/"}] | train | v0.1 | draft |
root_knowledge_014 | root-sft-dataset | root_knowledge | trees.leaves | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "How are TTree leaves related to branches?",
... | [] | A branch organizes stored data and may contain one or more leaves describing primitive values. Modern object branches can be more complex, so branch and leaf should not be assumed to be synonyms. | ["A branch may contain leaves", "Branch and leaf are not always identical concepts"] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees/"}] | train | v0.1 | draft |
root_knowledge_015 | root-sft-dataset | root_knowledge | trees.columnar | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why is columnar storage useful for HEP analysis?... | [] | An analysis often needs only a subset of variables from many events. Columnar storage lets it read selected branches instead of materializing every field, reducing I/O and memory use. | ["Analyses can select needed columns", "Selective reading reduces I/O"] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees/"}] | train | v0.1 | draft |
root_knowledge_016 | root-sft-dataset | root_knowledge | trees.collections | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why can a TTree branch be jagged rather than rec... | [] | The number of reconstructed objects varies by event. A jet or photon branch can therefore contain a different-length collection for each entry. | ["Object multiplicity varies by event", "Jagged branches have variable inner lengths"] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees/"}] | train | v0.1 | draft |
root_knowledge_017 | root-sft-dataset | root_knowledge | trees.baskets | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What purpose do baskets serve in TTree storage?"... | [] | Baskets hold batches of serialized branch data. Reading data in suitably sized ranges can reuse this organization more efficiently than many tiny, scattered reads. | ["Baskets group serialized branch data", "Read patterns affect I/O efficiency"] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees/"}] | train | v0.1 | draft |
root_knowledge_018 | root-sft-dataset | root_knowledge | trees.clusters | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What is a TTree cluster intended to improve?",
... | [] | Clusters group corresponding basket ranges across branches, supporting more efficient prefetching and chunked processing over entry ranges. | ["Clusters group entry ranges", "They support prefetching and chunking"] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees/"}] | train | v0.1 | draft |
root_knowledge_019 | root-sft-dataset | root_knowledge | trees.tchain | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "When is a TChain useful in a ROOT analysis?",
... | [] | A TChain presents compatible TTrees from multiple files as one logical sequence of entries. Branch compatibility and file provenance still need validation. | ["TChain combines compatible trees logically", "Compatibility must be checked"] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees/"}] | train | v0.1 | draft |
root_knowledge_020 | root-sft-dataset | root_knowledge | trees.rntuple | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Is RNTuple simply another name for TTree?",
... | [] | No. RNTuple is a newer ROOT columnar storage system with a different design and interfaces. Inspection code should identify the object class instead of assuming every columnar object is a TTree. | ["RNTuple and TTree are distinct", "Tools should inspect object classes"] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees/"}] | train | v0.1 | draft |
root_knowledge_021 | root-sft-dataset | root_knowledge | histograms.definition | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What does a one-dimensional ROOT histogram repre... | [] | It summarizes a numerical distribution by dividing an axis into bins and accumulating a content value for each bin. | ["A histogram divides an axis into bins", "Each bin has accumulated content"] | [] | [{"title": "ROOT histograms manual", "url": "https://root.cern/manual/histograms/"}] | train | v0.1 | draft |
root_knowledge_023 | root-sft-dataset | root_knowledge | histograms.flow | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What are histogram underflow and overflow bins?"... | [] | Underflow collects values below the regular axis range, while overflow collects values at or above the upper range according to ROOT's binning rules. They are separate from regular bins. | ["Underflow is below the regular range", "Overflow is beyond the regular upper range"] | [] | [{"title": "ROOT histograms manual", "url": "https://root.cern/manual/histograms/"}] | train | v0.1 | draft |
root_knowledge_024 | root-sft-dataset | root_knowledge | histograms.integral | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why might a visible-bin histogram integral diffe... | [] | The default integral may exclude underflow and overflow, and selections may reject events. Weighted entries also mean the sum of bin contents need not equal a raw event count. | ["Flow bins may be excluded", "Weighted content is not necessarily an event count"] | [] | [{"title": "ROOT histograms manual", "url": "https://root.cern/manual/histograms/"}] | train | v0.1 | draft |
root_knowledge_026 | root-sft-dataset | root_knowledge | histograms.sumw2 | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What information does Sumw2 preserve for a weigh... | [] | It stores the sum of squared weights per bin, which ROOT uses to calculate appropriate statistical uncertainties for weighted contents. | ["Sumw2 stores squared-weight sums", "It supports weighted statistical uncertainties"] | [] | [{"title": "ROOT histograms manual", "url": "https://root.cern/manual/histograms/"}] | train | v0.1 | draft |
root_knowledge_027 | root-sft-dataset | root_knowledge | histograms.errors | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why should a dataset include histogram uncertain... | [] | Two histograms can have similar contents but very different statistical precision. Errors or variances are needed for quantitative comparisons and fits. | ["Bin contents do not encode precision alone", "Errors or variances matter for fits"] | [] | [{"title": "ROOT histograms manual", "url": "https://root.cern/manual/histograms/"}] | train | v0.1 | draft |
root_knowledge_028 | root-sft-dataset | root_knowledge | histograms.normalization | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What must be checked before normalizing a histog... | [] | Decide whether flow bins and negative weights are included, verify the chosen integral is nonzero, and preserve the original normalization or scaling metadata. | ["The normalization integral must be defined", "Zero integrals and negative weights require care"] | [] | [{"title": "ROOT histograms manual", "url": "https://root.cern/manual/histograms/"}] | train | v0.1 | draft |
root_knowledge_029 | root-sft-dataset | root_knowledge | histograms.dimensions | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "How do one-dimensional and two-dimensional ROOT ... | [] | A one-dimensional histogram bins one variable; a two-dimensional histogram bins pairs of variables. In ROOT's class hierarchy TH2 inherits from TH1, so use GetDimension() when code needs to distinguish dimensions. | ["A one-dimensional histogram bins one variable; a two-dimensional histogram bins pairs of variables. In ROOT's class hierarchy TH2 inherits from TH1, so use GetDimension() when code needs to distinguish dimensions."] | [] | [{"title": "ROOT histogram class reference", "url": "https://root.cern/doc/master/classTH1.html"}] | train | v0.1 | draft |
root_knowledge_030 | root-sft-dataset | root_knowledge | histograms.profiles | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why should a TProfile not be described as an ord... | [] | A profile stores a mean dependent value for each bin of an independent variable, along with profile-specific uncertainty information; its content does not simply count entries. | ["A profile stores conditional means", "Profile content is not a simple count"] | [] | [{"title": "ROOT histograms manual", "url": "https://root.cern/manual/histograms/"}] | train | v0.1 | draft |
root_knowledge_031 | root-sft-dataset | root_knowledge | dataframe.workflow | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What are the main stages of an RDataFrame analys... | [] | Construct a dataframe from a data source, add transformations such as filters or defined columns, and book actions that produce aggregated results or snapshots. | ["Construction selects the data source", "Transformations precede result actions"] | [] | [{"title": "ROOT RDataFrame manual", "url": "https://root.cern/manual/data_frame/"}] | train | v0.1 | draft |
root_knowledge_033 | root-sft-dataset | root_knowledge | dataframe.filter | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What does RDataFrame Filter do?",
"tool_call... | [] | Filter keeps rows satisfying a boolean condition and passes only those rows to downstream transformations and actions. | ["Filter applies a boolean condition", "Rejected rows do not reach downstream operations"] | [] | [{"title": "ROOT RDataFrame manual", "url": "https://root.cern/manual/data_frame/"}] | train | v0.1 | draft |
root_knowledge_034 | root-sft-dataset | root_knowledge | dataframe.define | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What does RDataFrame Define do?",
"tool_call... | [] | Define creates a new column from existing columns or other inputs without modifying the original stored dataset unless a later output action writes it. | ["Define creates a derived column", "It does not by itself rewrite the source file"] | [] | [{"title": "ROOT RDataFrame manual", "url": "https://root.cern/manual/data_frame/"}] | train | v0.1 | draft |
root_knowledge_035 | root-sft-dataset | root_knowledge | dataframe.actions | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why is it useful to book several RDataFrame acti... | [] | Lazy actions can share an event loop through the same computation graph, avoiding separate full passes when the results are triggered together. | ["Lazy actions can share an event loop", "This can avoid repeated full scans"] | [] | [{"title": "ROOT RDataFrame manual", "url": "https://root.cern/manual/data_frame/"}] | train | v0.1 | draft |
root_knowledge_036 | root-sft-dataset | root_knowledge | dataframe.snapshot | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What is RDataFrame Snapshot used for?",
"too... | [] | Snapshot writes selected columns from the dataframe computation to a new columnar dataset, commonly a new ROOT file and tree. | ["Snapshot writes a new dataset", "Columns can be selected for output"] | [] | [{"title": "ROOT RDataFrame manual", "url": "https://root.cern/manual/data_frame/"}] | train | v0.1 | draft |
root_knowledge_037 | root-sft-dataset | root_knowledge | uproot.role | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What is Uproot used for in a Python HEP workflow... | [] | Uproot reads and writes ROOT files using Python array libraries without requiring the ROOT C++ runtime for supported objects. | ["Uproot provides Python ROOT I/O", "It integrates with array libraries"] | [] | [{"title": "Uproot getting started guide", "url": "https://uproot.readthedocs.io/en/stable/basic.html"}] | train | v0.1 | draft |
root_knowledge_039 | root-sft-dataset | root_knowledge | uproot.libraries | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "When should Awkward Array be preferred over NumP... | [] | Awkward Array naturally represents variable-length and nested per-event collections. Plain NumPy is most convenient for regular rectangular data. | ["Awkward handles variable-length nested data", "NumPy is suited to regular arrays"] | [] | [{"title": "Uproot getting started guide", "url": "https://uproot.readthedocs.io/en/stable/basic.html"}] | train | v0.1 | draft |
root_knowledge_040 | root-sft-dataset | root_knowledge | uproot.chunking | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why should a large ROOT tree be processed in ent... | [] | Chunking bounds memory use and allows an analysis to scale beyond files that fit in memory while retaining column selection. | ["Chunking bounds memory use", "It supports datasets larger than memory"] | [] | [{"title": "Uproot getting started guide", "url": "https://uproot.readthedocs.io/en/stable/basic.html"}] | train | v0.1 | draft |
root_knowledge_041 | root-sft-dataset | root_knowledge | hep.units | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Can the unit of a branch such as `photon_pt` be ... | [] | No. The name suggests a physical quantity but does not establish units or calibration. Use the dataset documentation or explicit metadata before stating units. | ["Branch names do not prove units", "Documentation or metadata is required"] | [] | [{"title": "ATLAS Open Data 13 TeV 2025 details", "url": "https://opendata.atlas.cern/docs/data/for_education/13TeV25_details"}] | train | v0.1 | draft |
root_knowledge_042 | root-sft-dataset | root_knowledge | hep.multiplicity | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "How should a branch named `jet_n` be checked aga... | [] | Compare `jet_n` with the per-entry lengths of jet collections such as `jet_pt`. Disagreement can reveal a schema misunderstanding, truncation, or malformed data. | ["Compare multiplicity with collection lengths", "Disagreement requires investigation"] | [] | [{"title": "ATLAS Open Data 13 TeV 2025 details", "url": "https://opendata.atlas.cern/docs/data/for_education/13TeV25_details"}] | train | v0.1 | draft |
root_knowledge_043 | root-sft-dataset | root_knowledge | hep.kinematics | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why should an agent avoid calculating invariant ... | [] | The calculation requires compatible numeric collections, correct object pairing, and consistent units. Skipping schema and metadata checks can produce plausible but meaningless values. | ["Invariant mass needs compatible inputs", "Units and pairing must be established"] | [] | [{"title": "ATLAS Open Data 13 TeV 2025 details", "url": "https://opendata.atlas.cern/docs/data/for_education/13TeV25_details"}] | train | v0.1 | draft |
root_knowledge_044 | root-sft-dataset | root_knowledge | hep.data_mc | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why must an analyst distinguish collision data f... | [] | Simulation commonly carries generator and correction weights, while collision data follows different normalization conventions. Applying MC weights to data would be incorrect. | ["Simulation and data use different weighting conventions", "MC weights should not be applied blindly to data"] | [] | [{"title": "ATLAS Open Data 13 TeV 2025 details", "url": "https://opendata.atlas.cern/docs/data/for_education/13TeV25_details"}] | train | v0.1 | draft |
root_knowledge_045 | root-sft-dataset | root_knowledge | hep.weights | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why can a Monte Carlo event weight be negative?"... | [] | Some higher-order event generators produce signed weights as part of their estimation procedure. Negative weights are not automatically corrupt data and must be retained according to the sample prescription. | ["Some generators produce signed weights", "Negative weights should not be discarded automatically"] | [] | [{"title": "ATLAS Open Data 13 TeV 2025 details", "url": "https://opendata.atlas.cern/docs/data/for_education/13TeV25_details"}] | train | v0.1 | draft |
root_knowledge_046 | root-sft-dataset | root_knowledge | hep.scale_factors | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What should be verified before multiplying sever... | [] | Confirm each factor's definition, applicability, systematic variation, and whether it is already included elsewhere. Otherwise corrections can be omitted or double counted. | ["Scale-factor definitions and applicability must be checked", "Double counting is a risk"] | [] | [{"title": "ATLAS Open Data 13 TeV 2025 details", "url": "https://opendata.atlas.cern/docs/data/for_education/13TeV25_details"}] | train | v0.1 | draft |
root_knowledge_047 | root-sft-dataset | root_knowledge | hep.triggers | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What does a boolean trigger branch establish by ... | [] | It establishes the stored pass/fail value for that branch. It does not by itself document the trigger menu, prescale, efficiency, or suitability for a physics selection. | ["A trigger branch stores a decision", "Menu and efficiency require additional documentation"] | [] | [{"title": "ATLAS Open Data 13 TeV 2025 details", "url": "https://opendata.atlas.cern/docs/data/for_education/13TeV25_details"}] | train | v0.1 | draft |
root_knowledge_048 | root-sft-dataset | root_knowledge | hep.cutflow | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What is the purpose of a cut-flow histogram in a... | [] | A cut flow records how the event yield changes through an ordered sequence of selections, helping validate selection logic and locate unexpected losses. | ["Cut flows track yields across selections", "They help diagnose event losses"] | [] | [{"title": "ATLAS Open Data 13 TeV 2025 details", "url": "https://opendata.atlas.cern/docs/data/for_education/13TeV25_details"}] | train | v0.1 | draft |
root_knowledge_049 | root-sft-dataset | root_knowledge | hep.validation | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What basic checks should be made before using a ... | [] | Verify its checksum and provenance, inventory its objects, confirm the expected tree and entry count, inspect branch names and types, and test a bounded sample for readable values. | ["Check provenance and checksum", "Inspect schema before bulk reading"] | [] | [{"title": "ATLAS Open Data 13 TeV 2025 details", "url": "https://opendata.atlas.cern/docs/data/for_education/13TeV25_details"}] | train | v0.1 | draft |
root_knowledge_050 | root-sft-dataset | root_knowledge | hep.reproducibility | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why should closely related evaluation questions ... | [] | Otherwise a model can memorize facts specific to that shared schema or sample and appear to generalize on a test question that is really just a relabeled copy of something it already saw in training. | ["Near-identical files can leak information across a train/test split", "Grouping them together keeps evaluation results credible"] | [] | [{"title": "ATLAS Open Data 13 TeV 2025 details", "url": "https://opendata.atlas.cern/docs/data/for_education/13TeV25_details"}] | train | v0.1 | draft |
root_knowledge_051 | root-sft-dataset | root_knowledge | inventory.event_container | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Which object in `diphoton_collision_data.root` s... | [] | `events_diphoton`, the TTree in this file's inventory; `h_myy` is a histogram summary, not the event-level columnar container. | ["Select TTree events_diphoton"] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_052 | root-sft-dataset | root_knowledge | inventory.entry_count | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "How many stored rows should an inspection report... | [] | 12000 entries. Calling them physics events outright is appropriate only once the file's accompanying documentation confirms this is genuinely an event-level tree. | ["Report 12000 entries", "Do not infer row semantics without documentation"] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_053 | root-sft-dataset | root_knowledge | schema.multiplicity | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Which branch records the per-entry photon multip... | [] | `photon_n`; its value should also be cross-checked against the actual lengths of the jagged photon collections rather than trusted in isolation. | ["Identify photon_n", "Cross-check jagged lengths"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_054 | root-sft-dataset | root_knowledge | schema.jagged | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "In the `diphoton` collision data schema, which l... | [] | `photon_pt`, `photon_eta`, `photon_phi`, and `photon_isTightID` are jagged, since each entry can hold a different number of photons. | ["Identify photon_pt, photon_eta, photon_phi, and photon_isTightID as jagged"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_055 | root-sft-dataset | root_knowledge | schema.identifiers | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Which scalar branch in the `diphoton` collision ... | [] | `run_number` is the listed acquisition-context identifier under the convention assumed here. Confirm the convention in production documentation; run numbers need not uniquely identify events and their presence alone does not prove that the file is collision data. | ["`run_number` is the listed acquisition-context identifier under the convention assumed here. Confirm the convention in production documentation; run numbers need not uniquely identify events and their presence alone does not prove that the file is collision data."] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_056 | root-sft-dataset | root_knowledge | analysis.minimal_columns | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Working with the `diphoton` collision data sampl... | [] | For an unweighted object-level plot, read `photon_pt` and `photon_eta` and verify their per-event lengths and object alignment. Reading `photon_n` is an optional count consistency check. A weighted prediction also needs the documented event/object weights; broadcast event weights to the selected objects before flatteni... | ["For an unweighted object-level plot, read `photon_pt` and `photon_eta` and verify their per-event lengths and object alignment. Reading `photon_n` is an optional count consistency check. A weighted prediction also needs the documented event/object weights; broadcast event weights to the selected objects before flatte... | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_057 | root-sft-dataset | root_knowledge | analysis.safe_indexing | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "In the `diphoton` collision data file, what must... | [] | Before indexing positions 0 and 1, check that the actual `photon_pt` array has at least two elements in that entry. If `photon_n` is documented as its count, require agreement and `photon_n >= 2`. A count of one is insufficient. Index 0 is the leading object only if the collection is documented or explicitly sorted in ... | ["Before indexing positions 0 and 1, check that the actual `photon_pt` array has at least two elements in that entry. If `photon_n` is documented as its count, require agreement and `photon_n >= 2`. A count of one is insufficient. Index 0 is the leading object only if the collection is documented or explicitly sorted i... | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_058 | root-sft-dataset | root_knowledge | analysis.selection | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "For the teaching scenario `diphoton_collision_da... | [] | Read `photon_n` and evaluate `photon_n >= 2`. This is an explicit two-object requirement for this exercise, not a prerequisite for every object calculation. Do not read unrelated branches to evaluate this cut. Later reads still occur in baskets/chunks: applying an Uproot array mask does not guarantee that only selected... | ["Read `photon_n` and evaluate `photon_n >= 2`. This is an explicit two-object requirement for this exercise, not a prerequisite for every object calculation. Do not read unrelated branches to evaluate this cut. Later reads still occur in baskets/chunks: applying an Uproot array mask does not guarantee that only select... | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_059 | root-sft-dataset | root_knowledge | histogram.crosscheck | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "How can `h_myy` be used to validate a distributi... | [] | Inspect `h_myy`'s binning, flow bins, contents, and uncertainties, then compare it against a calculation made under the same documented selection and weighting; matching object names alone does not confirm agreement. | ["Inspect h_myy metadata and values", "Match selection and weighting"] | [] | [{"title": "ROOT histograms manual", "url": "https://root.cern/manual/histograms"}] | train | v0.1 | draft |
root_knowledge_060 | root-sft-dataset | root_knowledge | weights.sample_type | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Should an event-weight branch be used for the `d... | [] | No event-weight branch is listed in this scenario, so do not invent a Monte Carlo weight. Unit weights are appropriate for an explicitly unweighted count. Real collision-data analyses can require documented prescale, efficiency or background-estimation weights; check the analysis prescription before treating all data a... | ["No event-weight branch is listed in this scenario, so do not invent a Monte Carlo weight. Unit weights are appropriate for an explicitly unweighted count. Real collision-data analyses can require documented prescale, efficiency or background-estimation weights; check the analysis prescription before treating all data... | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_061 | root-sft-dataset | root_knowledge | semantics.units | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "May the unit of `photon_pt` be assumed just from... | [] | If the conventional definitions apply, `photon_pt` is transverse momentum (often GeV or MeV) and `photon_eta` is dimensionless pseudorapidity. Confirm definitions, the momentum unit, and calibration in production metadata; branch names alone are not proof. A pt-versus-eta plot does not require the two axes to have matc... | ["If the conventional definitions apply, `photon_pt` is transverse momentum (often GeV or MeV) and `photon_eta` is dimensionless pseudorapidity. Confirm definitions, the momentum unit, and calibration in production metadata; branch names alone are not proof. A pt-versus-eta plot does not require the two axes to have ma... | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_062 | root-sft-dataset | root_knowledge | tools.sequence | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What bounded tool sequence should be used to sum... | [] | First list the file's stored objects, then describe `events_diphoton` with targeted branch patterns, and finally summarize only the branches actually needed to answer the question at hand — avoiding arbitrary code execution or a full-file read. | ["Inventory before payload reads", "Use targeted bounded reads"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_063 | root-sft-dataset | root_knowledge | performance.chunking | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "How should `photon_pt` be processed across all 1... | [] | Read it in bounded entry chunks and aggregate partial results across chunks, keeping other branches disabled unless the computation actually needs them. | ["Use bounded entry chunks", "Select only needed branches"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_064 | root-sft-dataset | root_knowledge | validation.collection_lengths | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What consistency check can detect a malformed ph... | [] | For every inspected entry, compare `photon_n` against the lengths of `photon_pt`, `photon_eta`, `photon_phi`, and `photon_isTightID`; report any disagreement rather than silently truncating or padding the arrays. | ["Compare photon_n to all collection lengths", "Report mismatches"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_065 | root-sft-dataset | root_knowledge | reproducibility.fixture | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What must be recorded about `diphoton_collision_... | [] | Its exact checksum and an immutable production or derivation revision identifier; a filename alone does not guarantee the content behind it hasn't changed. | ["A checksum identifies the exact file content", "An immutable production/derivation revision is also needed"] | [] | [{"title": "ROOT files manual", "url": "https://root.cern/manual/root_files"}] | train | v0.1 | draft |
root_knowledge_066 | root-sft-dataset | root_knowledge | inventory.event_container | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "`diphoton_signal_simulation.root` contains both ... | [] | `events_diphoton`. `h_myy` is a pre-aggregated histogram and does not expose individual events the way the tree does. | ["Select TTree events_diphoton"] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_067 | root-sft-dataset | root_knowledge | inventory.entry_count | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What entry count does `events_diphoton` in `diph... | [] | 13237 entries; whether each entry corresponds one-to-one with a physics event still depends on confirming that from the file's documentation, rather than assuming it. | ["Report 13237 entries", "Do not infer row semantics without documentation"] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_068 | root-sft-dataset | root_knowledge | schema.multiplicity | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "How would you determine how many photons were re... | [] | Read the scalar `photon_n` branch for that entry, and verify it agrees with the length of `photon_pt` for the same entry. | ["Identify photon_n", "Cross-check jagged lengths"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_069 | root-sft-dataset | root_knowledge | schema.jagged | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Can `photon_pt` in `events_diphoton` be read as ... | [] | No, `photon_pt` is jagged: its length varies per entry along with `photon_n`, so it needs variable-length (ragged/jagged-array) handling rather than a fixed shape. | ["Identify photon_pt, photon_eta, photon_phi, and photon_isTightID as jagged"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_070 | root-sft-dataset | root_knowledge | schema.identifiers | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "In `events_diphoton`, what does the presence of ... | [] | `mc_channel_number` is the sample identifier used by the simulation convention assumed in this scenario. Confirm its meaning and the data/MC classification in production documentation; a branch name alone does not prove that a file is simulation. | ["`mc_channel_number` is the sample identifier used by the simulation convention assumed in this scenario. Confirm its meaning and the data/MC classification in production documentation; a branch name alone does not prove that a file is simulation."] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_071 | root-sft-dataset | root_knowledge | analysis.minimal_columns | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "For `events_diphoton` in the `diphoton` signal s... | [] | For an unweighted object-level plot, read `photon_pt` and `photon_eta` and verify their per-event lengths and object alignment. Reading `photon_n` is an optional count consistency check. A weighted prediction also needs the documented event/object weights; broadcast event weights to the selected objects before flatteni... | ["For an unweighted object-level plot, read `photon_pt` and `photon_eta` and verify their per-event lengths and object alignment. Reading `photon_n` is an optional count consistency check. A weighted prediction also needs the documented event/object weights; broadcast event weights to the selected objects before flatte... | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_072 | root-sft-dataset | root_knowledge | analysis.safe_indexing | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Is it safe to always access `photon_pt[1]` once ... | [] | Before indexing positions 0 and 1, check that the actual `photon_pt` array has at least two elements in that entry. If `photon_n` is documented as its count, require agreement and `photon_n >= 2`. A count of one is insufficient. Index 0 is the leading object only if the collection is documented or explicitly sorted in ... | ["Before indexing positions 0 and 1, check that the actual `photon_pt` array has at least two elements in that entry. If `photon_n` is documented as its count, require agreement and `photon_n >= 2`. A count of one is insufficient. Index 0 is the leading object only if the collection is documented or explicitly sorted i... | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_073 | root-sft-dataset | root_knowledge | analysis.selection | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "For the teaching scenario `diphoton_signal_simul... | [] | Read `photon_n` and evaluate `photon_n >= 2`. This is an explicit two-object requirement for this exercise, not a prerequisite for every object calculation. Do not read unrelated branches to evaluate this cut. Later reads still occur in baskets/chunks: applying an Uproot array mask does not guarantee that only selected... | ["Read `photon_n` and evaluate `photon_n >= 2`. This is an explicit two-object requirement for this exercise, not a prerequisite for every object calculation. Do not read unrelated branches to evaluate this cut. Later reads still occur in baskets/chunks: applying an Uproot array mask does not guarantee that only select... | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_074 | root-sft-dataset | root_knowledge | histogram.crosscheck | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "If a recomputed histogram from `events_diphoton`... | [] | Confirm both histograms use the same binning, the same selection, and the same event weighting as `h_myy` before concluding there is a real discrepancy; a mismatch in any of those alone can explain a difference. | ["Inspect h_myy metadata and values", "Match selection and weighting"] | [] | [{"title": "ROOT histograms manual", "url": "https://root.cern/manual/histograms"}] | train | v0.1 | draft |
root_knowledge_075 | root-sft-dataset | root_knowledge | weights.sample_type | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Does `diphoton_signal_simulation.root` carry a M... | [] | The scenario lists `event_weight` for simulation. Check its documented definition, normalization and any object/selection corrections before using it; avoid applying the same factor twice. A branch name alone does not establish the weight composition or prove the sample type. | ["The scenario lists `event_weight` for simulation. Check its documented definition, normalization and any object/selection corrections before using it; avoid applying the same factor twice. A branch name alone does not establish the weight composition or prove the sample type."] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_076 | root-sft-dataset | root_knowledge | semantics.units | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "For the `diphoton` signal simulation sample, is ... | [] | If the conventional definitions apply, `photon_pt` is transverse momentum (often GeV or MeV) and `photon_eta` is dimensionless pseudorapidity. Confirm definitions, the momentum unit, and calibration in production metadata; branch names alone are not proof. A pt-versus-eta plot does not require the two axes to have matc... | ["If the conventional definitions apply, `photon_pt` is transverse momentum (often GeV or MeV) and `photon_eta` is dimensionless pseudorapidity. Confirm definitions, the momentum unit, and calibration in production metadata; branch names alone are not proof. A pt-versus-eta plot does not require the two axes to have ma... | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_077 | root-sft-dataset | root_knowledge | tools.sequence | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Rather than running arbitrary code against `diph... | [] | List the file's objects first, then inspect `events_diphoton`'s branch list with a targeted pattern, and only then read the specific branches needed, instead of reading the whole file or running unconstrained code. | ["Inventory before payload reads", "Use targeted bounded reads"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_078 | root-sft-dataset | root_knowledge | performance.chunking | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What is a memory-conscious way to sum `photon_pt... | [] | Process the 13237 entries in bounded chunks, accumulating a running total, rather than loading the entire `photon_pt` collection for every entry into memory simultaneously. | ["Use bounded entry chunks", "Select only needed branches"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_079 | root-sft-dataset | root_knowledge | validation.collection_lengths | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "If `photon_n` disagrees with `len(photon_pt)` fo... | [] | The disagreement should be reported explicitly as a data-quality issue, not silently resolved by truncating, padding, or otherwise guessing which value is correct. | ["Compare photon_n to all collection lengths", "Report mismatches"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_080 | root-sft-dataset | root_knowledge | reproducibility.fixture | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Besides the file itself, what software informati... | [] | The ROOT and analysis-software release (and any relevant library versions) used to produce and read the file, since numerical results can depend on that software version. | ["The software/library release used to produce and read the file matters", "Results can depend on that software version"] | [] | [{"title": "ROOT files manual", "url": "https://root.cern/manual/root_files"}] | train | v0.1 | draft |
root_knowledge_081 | root-sft-dataset | root_knowledge | stats.roofit_realvar | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What does a RooRealVar represent in a RooFit mod... | [] | A RooRealVar stores a real value and represents either an observable or a parameter. Its role depends on the model. Observable ranges can define normalization domains; parameter bounds restrict the minimizer's search. Those are different uses of a range. | ["A RooRealVar stores a real value and represents either an observable or a parameter. Its role depends on the model. Observable ranges can define normalization domains; parameter bounds restrict the minimizer's search. Those are different uses of a range."] | [] | [{"title": "RooRealVar class reference", "url": "https://root.cern/doc/master/classRooRealVar.html"}] | train | v0.1 | draft |
root_knowledge_082 | root-sft-dataset | root_knowledge | inventory.entry_count | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Is it accurate to say `events_diphoton` contains... | [] | It is accurate to say it contains 14474 entries; calling them events specifically requires confirming from documentation that the tree is event-level, rather than, say, an object-level or derived tree. | ["Report 14474 entries", "Do not infer row semantics without documentation"] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_083 | root-sft-dataset | root_knowledge | stats.roofit_pdf | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What must a RooAbsPdf-derived object provide bey... | [] | A RooAbsPdf must be normalizable to a well-defined integral of 1 over its observables' range, since RooFit treats it as a genuine probability density, not merely an arbitrary function. | ["RooAbsPdf represents a normalized probability density", "Normalization over the observable range is required"] | [] | [{"title": "RooFit users guide", "url": "https://root.cern/manual/roofit"}] | train | v0.1 | draft |
root_knowledge_084 | root-sft-dataset | root_knowledge | stats.roofit_addpdf | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What role does RooAddPdf play when building a si... | [] | RooAddPdf sums two or more component PDFs using fraction (or, in the extended case, yield) coefficients, letting a composite model like signal-plus-background be built from simpler individual PDFs. | ["RooAddPdf combines PDFs with fraction/yield coefficients", "It is the standard way to build signal-plus-background models"] | [] | [{"title": "RooFit users guide", "url": "https://root.cern/manual/roofit"}] | train | v0.1 | draft |
root_knowledge_085 | root-sft-dataset | root_knowledge | stats.roofit_proddpdf | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "When is RooProdPdf the right tool instead of Roo... | [] | RooAddPdf forms mixtures, such as signal plus background. RooProdPdf forms products, including independent densities f(x)g(y), conditional models f(x|y)g(y), and constraint factors. A product does not automatically imply independence: dependencies and normalization sets determine its meaning. | ["RooAddPdf forms mixtures, such as signal plus background. RooProdPdf forms products, including independent densities f(x)g(y), conditional models f(x|y)g(y), and constraint factors. A product does not automatically imply independence: dependencies and normalization sets determine its meaning."] | [] | [{"title": "RooProdPdf class reference", "url": "https://root.cern/doc/master/classRooProdPdf.html"}] | train | v0.1 | draft |
root_knowledge_086 | root-sft-dataset | root_knowledge | stats.roofit_dataset_vs_datahist | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What is the difference between RooDataSet and Ro... | [] | RooDataSet stores unbinned data as a collection of individual points over the observables, while RooDataHist stores binned data, so the choice determines whether a subsequent fit is unbinned or binned. | ["RooDataSet is unbinned; RooDataHist is binned", "The data container determines the fit type"] | [] | [{"title": "RooFit users guide", "url": "https://root.cern/manual/roofit"}] | train | v0.1 | draft |
root_knowledge_087 | root-sft-dataset | root_knowledge | stats.roofit_fitto | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What does calling fitTo() on a RooFit model actu... | [] | fitTo() minimizes the negative log-likelihood for the model and data and updates floating parameters. Request RooFit::Save() to obtain a RooFitResult containing diagnostics, parameter estimates, and covariance information; check that result before reporting a measurement. | ["fitTo() minimizes the negative log-likelihood for the model and data and updates floating parameters. Request RooFit::Save() to obtain a RooFitResult containing diagnostics, parameter estimates, and covariance information; check that result before reporting a measurement."] | [] | [{"title": "RooAbsPdf fitTo reference", "url": "https://root.cern/doc/master/classRooAbsPdf.html"}] | train | v0.1 | draft |
root_knowledge_088 | root-sft-dataset | root_knowledge | stats.roofit_extended_likelihood | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What does an extended maximum-likelihood fit add... | [] | An extended fit treats the total observed event count as an additional Poisson-distributed quantity predicted by the model, so the fit constrains both the shape parameters and the overall expected yield at once. | ["Extended fits treat total yield as a fitted Poisson quantity", "Both shape and yield are constrained together"] | [] | [{"title": "RooFit users guide", "url": "https://root.cern/manual/roofit"}] | train | v0.1 | draft |
root_knowledge_089 | root-sft-dataset | root_knowledge | stats.roofit_workspace2 | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why would an analyst export a finished RooFit mo... | [] | A workspace bundles the PDFs, parameters, and datasets into one persistable object, so the exact model can be saved, shared, and reloaded later by the same or different tools without rebuilding it in code. | ["A workspace bundles PDFs, parameters, and datasets", "It allows the model to be saved and reloaded without rebuilding it"] | [] | [{"title": "RooFit users guide", "url": "https://root.cern/manual/roofit"}] | train | v0.1 | draft |
root_knowledge_090 | root-sft-dataset | root_knowledge | stats.roofit_factory_syntax | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What is RooWorkspace::factory() used for?",
... | [] | It builds RooFit objects (variables, PDFs, compositions) from a compact string expression instead of requiring each C++ or Python class to be instantiated and wired together by hand. | ["factory() builds RooFit objects from string expressions", "It avoids manual instantiation of each class"] | [] | [{"title": "RooFit users guide", "url": "https://root.cern/manual/roofit"}] | train | v0.1 | draft |
root_knowledge_091 | root-sft-dataset | root_knowledge | stats.roofit_createnll | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What does createNLL() build, and how does it rel... | [] | createNLL() constructs the negative log-likelihood function for a model and dataset as a standalone object; fitTo() is essentially a convenience wrapper that builds this NLL internally and then minimizes it. | ["createNLL builds the negative log-likelihood as its own object", "fitTo minimizes an NLL built the same way internally"] | [] | [{"title": "RooFit users guide", "url": "https://root.cern/manual/roofit"}] | train | v0.1 | draft |
root_knowledge_092 | root-sft-dataset | root_knowledge | stats.roofit_createprofile | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What does createProfile() produce from a likelih... | [] | createProfile() constructs a function that minimizes over other floating parameters when evaluated at each fixed scan point. It does not eagerly calculate every possible scan point. Check minimization quality throughout the scan. | ["createProfile() constructs a function that minimizes over other floating parameters when evaluated at each fixed scan point. It does not eagerly calculate every possible scan point. Check minimization quality throughout the scan."] | [] | [{"title": "RooFit users guide", "url": "https://root.cern/manual/roofit"}] | train | v0.1 | draft |
root_knowledge_093 | root-sft-dataset | root_knowledge | performance.chunking | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "If `events_diphoton`'s 14474 entries are too man... | [] | Read and process `photon_pt` in bounded ranges of entries, combining the partial results afterward, instead of attempting one single full read. | ["Use bounded entry chunks", "Select only needed branches"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_094 | root-sft-dataset | root_knowledge | stats.roofit_rooplot | grounded_knowledge | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What is a RooPlot used for in RooFit?",
"too... | [] | A RooPlot is a drawing frame tied to one observable, letting a dataset, one or more PDFs, and fit curves all be drawn together on consistent, matching axes. | ["RooPlot is an observable-specific drawing frame", "It keeps a dataset and PDF curves on consistent axes"] | [] | [{"title": "RooFit users guide", "url": "https://root.cern/manual/roofit"}] | train | v0.1 | draft |
root_knowledge_095 | root-sft-dataset | root_knowledge | reproducibility.fixture | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What must be documented about any selection appl... | [] | The exact cut definitions and the order they were applied in; a selection that sounds the same in prose can correspond to different actual cuts. | ["Exact cut definitions must be recorded", "The order cuts are applied in also matters"] | [] | [{"title": "ROOT files manual", "url": "https://root.cern/manual/root_files"}] | train | v0.1 | draft |
root_knowledge_096 | root-sft-dataset | root_knowledge | inventory.event_container | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "For an analysis that needs to loop entry by entr... | [] | `events_dielectron`, since it is the columnar, per-entry container; `h_mee` only stores an already-summarized distribution. | ["Select TTree events_dielectron"] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_097 | root-sft-dataset | root_knowledge | inventory.entry_count | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What is the correct, cautious way to report the ... | [] | Report 19919 entries as the measured fact, and treat 'events' as a label that needs confirmation from documentation before being used interchangeably with 'entries'. | ["Report 19919 entries", "Do not infer row semantics without documentation"] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_098 | root-sft-dataset | root_knowledge | schema.multiplicity | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What is the scalar counterpart to the jagged `el... | [] | `electron_n`, which records the per-entry electron multiplicity and should match the length of `electron_pt` for the same entry. | ["Identify electron_n", "Cross-check jagged lengths"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_099 | root-sft-dataset | root_knowledge | schema.jagged | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why can't `electron_eta` be assumed to have the ... | [] | Because it is a jagged branch tied to `electron_n`: different events can have different numbers of electrons, so its per-entry length is not fixed. | ["Identify electron_pt, electron_eta, electron_phi, and electron_charge as jagged"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_100 | root-sft-dataset | root_knowledge | schema.identifiers | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Given the branch inventory for `dielectron` coll... | [] | `run_number` is the listed acquisition-context identifier under the convention assumed here. Confirm the convention in production documentation; run numbers need not uniquely identify events and their presence alone does not prove that the file is collision data. | ["`run_number` is the listed acquisition-context identifier under the convention assumed here. Confirm the convention in production documentation; run numbers need not uniquely identify events and their presence alone does not prove that the file is collision data."] | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_101 | root-sft-dataset | root_knowledge | analysis.minimal_columns | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Before writing any selection code for the `diele... | [] | For an unweighted object-level plot, read `electron_pt` and `electron_eta` and verify their per-event lengths and object alignment. Reading `electron_n` is an optional count consistency check. A weighted prediction also needs the documented event/object weights; broadcast event weights to the selected objects before fl... | ["For an unweighted object-level plot, read `electron_pt` and `electron_eta` and verify their per-event lengths and object alignment. Reading `electron_n` is an optional count consistency check. A weighted prediction also needs the documented event/object weights; broadcast event weights to the selected objects before ... | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_102 | root-sft-dataset | root_knowledge | analysis.safe_indexing | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "For the `dielectron` collision data data, what i... | [] | Before indexing positions 0 and 1, check that the actual `electron_pt` array has at least two elements in that entry. If `electron_n` is documented as its count, require agreement and `electron_n >= 2`. A count of one is insufficient. Index 0 is the leading object only if the collection is documented or explicitly sort... | ["Before indexing positions 0 and 1, check that the actual `electron_pt` array has at least two elements in that entry. If `electron_n` is documented as its count, require agreement and `electron_n >= 2`. A count of one is insufficient. Index 0 is the leading object only if the collection is documented or explicitly so... | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_103 | root-sft-dataset | root_knowledge | analysis.selection | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "For the teaching scenario `dielectron_collision_... | [] | Read `electron_n` and evaluate `electron_n >= 2`. This is an explicit two-object requirement for this exercise, not a prerequisite for every object calculation. Do not read unrelated branches to evaluate this cut. Later reads still occur in baskets/chunks: applying an Uproot array mask does not guarantee that only sele... | ["Read `electron_n` and evaluate `electron_n >= 2`. This is an explicit two-object requirement for this exercise, not a prerequisite for every object calculation. Do not read unrelated branches to evaluate this cut. Later reads still occur in baskets/chunks: applying an Uproot array mask does not guarantee that only se... | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_104 | root-sft-dataset | root_knowledge | histogram.crosscheck | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "For the `dielectron` collision data file, what m... | [] | Comparing the actual bin contents, flow bins, and uncertainties under matching selection and weighting definitions — not just noting that an object with the same name (`h_mee`) exists. | ["Inspect h_mee metadata and values", "Match selection and weighting"] | [] | [{"title": "ROOT histograms manual", "url": "https://root.cern/manual/histograms"}] | train | v0.1 | draft |
root_knowledge_105 | root-sft-dataset | root_knowledge | weights.sample_type | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What distinguishes the weighting scheme of `even... | [] | No event-weight branch is listed in this scenario, so do not invent a Monte Carlo weight. Unit weights are appropriate for an explicitly unweighted count. Real collision-data analyses can require documented prescale, efficiency or background-estimation weights; check the analysis prescription before treating all data a... | ["No event-weight branch is listed in this scenario, so do not invent a Monte Carlo weight. Unit weights are appropriate for an explicitly unweighted count. Real collision-data analyses can require documented prescale, efficiency or background-estimation weights; check the analysis prescription before treating all data... | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_106 | root-sft-dataset | root_knowledge | semantics.units | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "In the `dielectron_collision_data` scenario, do ... | [] | If the conventional definitions apply, `electron_pt` is transverse momentum (often GeV or MeV) and `electron_eta` is dimensionless pseudorapidity. Confirm definitions, the momentum unit, and calibration in production metadata; branch names alone are not proof. A pt-versus-eta plot does not require the two axes to have ... | ["If the conventional definitions apply, `electron_pt` is transverse momentum (often GeV or MeV) and `electron_eta` is dimensionless pseudorapidity. Confirm definitions, the momentum unit, and calibration in production metadata; branch names alone are not proof. A pt-versus-eta plot does not require the two axes to hav... | [] | [{"title": "ROOT trees manual", "url": "https://root.cern/manual/trees"}] | train | v0.1 | draft |
root_knowledge_107 | root-sft-dataset | root_knowledge | tools.sequence | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "When first exploring `dielectron_collision_data.... | [] | An inventory step — listing the stored objects and describing `events_dielectron`'s branch names/types — before deciding which specific branches to actually read. | ["Inventory before payload reads", "Use targeted bounded reads"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_108 | root-sft-dataset | root_knowledge | performance.chunking | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why might reading all 19919 entries of `electron... | [] | It can exceed available memory or be unnecessarily slow; chunked reading with only the needed branches enabled avoids that while still covering every entry. | ["Use bounded entry chunks", "Select only needed branches"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_109 | root-sft-dataset | root_knowledge | validation.collection_lengths | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "What is a red flag when validating `events_diele... | [] | Any entry where `electron_n` does not match the actual length of `electron_pt` (or the other jagged electron branches) is a red flag that should be reported, not silently patched over. | ["Compare electron_n to all collection lengths", "Report mismatches"] | [] | [{"title": "Uproot documentation", "url": "https://uproot.readthedocs.io/en/latest/"}] | train | v0.1 | draft |
root_knowledge_110 | root-sft-dataset | root_knowledge | reproducibility.fixture | contextual_analysis | [
{
"role": "system",
"content": "You are a careful high-energy-physics assistant. Answer ROOT questions accurately and concisely. Do not invent unsupported details.",
"tool_calls": null,
"tool_call_id": null
},
{
"role": "user",
"content": "Why should random seeds be recorded when a repor... | [] | So the same pseudo-data can be regenerated exactly later, which is necessary to reproduce or debug that specific reported result rather than getting a statistically similar but different one. | ["Recording seeds allows exact regeneration of pseudo-data", "This is needed to reproduce or debug a specific result"] | [] | [{"title": "ROOT files manual", "url": "https://root.cern/manual/root_files"}] | train | v0.1 | draft |
End of preview. Expand in Data Studio
ROOT SFT dataset
root.jsonl is the canonical source dataset for ROOT analysis tasks. It
contains 617 JSONL records with a predefined split:
| Split | Records |
|---|---|
train |
494 |
validation |
62 |
test |
61 |
The current records are marked for SFT. Render them with the parent project's
training/prepare_root_sft.py script to create sft/train.parquet and
sft/validation.parquet, which contain the messages and tools fields
expected by verl. Keep the test split held out.
- Downloads last month
- 25