Datasets:
id stringclasses 10
values | language stringclasses 2
values | status stringclasses 2
values | source_dataset stringclasses 1
value | source_split stringclasses 1
value | source_uuid stringclasses 10
values | generator stringclasses 1
value | guidelines stringclasses 1
value | parallel_tool_calls unknown | notes unknown | messages unknown | tools unknown | pillar_compliance int32 | verdict stringclasses 0
values | revision_notes stringclasses 0
values | review_status stringclasses 1
value | review_comments unknown | source_ref stringclasses 10
values | schema_hash stringclasses 1
value | reviewed_by stringclasses 1
value | reviewed_at timestamp[ms, tz=UTC]date 2026-10-25 15:19:05 2026-10-25 16:14:20 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
pilot-001 | de | clean | nvidia/Nemotron-SFT-Agentic-v4 | tool_calling | 31fd693e-b7ea-4f4b-a96c-5263f079e6ff | ml-intern-improver-0.3-pilot | IDEAL_TRACE_GUIDELINES v0.3 | false | [
"Seed prose was emoji-heavy and over-answered (app recommendations the user never asked for) - removed.",
"Both seed calls kept: both results are used (forecast only exists in the Wooster result, creek-level readings in the second).",
"User intent Germanized; tool calls, ids and results byte-verbatim from seed.... | [
{
"role": "system",
"content": ""
},
{
"role": "user",
"content": "Ich will morgen am Killbuck Creek bei Wooster angeln gehen und brauche dafür eine stundengenaue Wettervorhersage, um die beste Zeit für den Ausflug zu finden."
},
{
"role": "assistant",
"reasoning_content": "Der Nutze... | [
{
"type": "function",
"function": {
"name": "current_weather",
"description": "A wrapper around OpenWeatherMap API.Useful for fetching current weather information for a specified location. ",
"parameters": {
"type": "object",
"properties": {
"location": {
... | null | null | null | answered | [] | hf://datasets/flozi00/ideal-trace-pilot-batch-1/data/traces-de.jsonl?row=0 | 8a704159 | flozi00 | 2026-10-25T15:19:05.309000 |
pilot-002 | de | clean | nvidia/Nemotron-SFT-Agentic-v4 | tool_calling | 8cdfcc5f-ac7c-4ea5-97d2-e3f9f93c691f | ml-intern-improver-0.3-pilot | IDEAL_TRACE_GUIDELINES v0.3 | false | [
"No failed calls in the seed; reasoning tightened, emoji headers and the next-step upsell removed.",
"User intent Germanized; calls, ids and results byte-verbatim."
] | [
{
"role": "system",
"content": ""
},
{
"role": "user",
"content": "Hallo! Ich bereite mich auf ein volles Wochenende vor - am Sonntag habe ich ein Softballspiel meiner Liga, eine Schicht als Freiwillige in der Tafelküche, und ich habe versprochen, für den Geburtstag meiner Nachbarin ein Rosmarin... | [
{
"type": "function",
"function": {
"name": "list_zapier_NLA_actions",
"description": "This tool returns a list of the Zapier user's exposed actions.",
"parameters": {
"type": "object",
"properties": {
"api_key": {
"type": "string",
"descri... | null | null | null | answered | [] | hf://datasets/flozi00/ideal-trace-pilot-batch-1/data/traces-de.jsonl?row=1 | 8a704159 | flozi00 | 2026-10-25T15:19:05.309000 |
pilot-003 | de | clean | nvidia/Nemotron-SFT-Agentic-v4 | tool_calling | 195fde62-415b-4d85-8f75-10259c1dfa32 | ml-intern-improver-0.3-pilot | IDEAL_TRACE_GUIDELINES v0.3 | false | [
"Seed called get_ip then ip_more; kept - the second depends on the first.",
"TEST-NET-3 caveat promoted from a buried note to the central finding; RFC 5737 is a certain external fact (P4 layer 2).",
"Removed emoji headers and the redundant recommendation list."
] | [
{
"role": "system",
"content": ""
},
{
"role": "user",
"content": "Ich plane einen Kultur-Kunst-Roadtrip über mehrere US-Bundesstaaten und will prüfen, ob mein aktueller Startort in Killbuck, Ohio, zu den IP-Regionsdaten passt, damit ich Veranstaltungsorte und die IPs potenziell remote arbeitend... | [
{
"type": "function",
"function": {
"name": "get_ip",
"description": "Get the ip of the current user. ",
"parameters": {
"type": "object",
"properties": {},
"required": []
}
}
},
{
"type": "function",
"function": {
"name": "ip_more",
... | null | null | null | answered | [] | hf://datasets/flozi00/ideal-trace-pilot-batch-1/data/traces-de.jsonl?row=2 | 8a704159 | flozi00 | 2026-10-25T15:19:05.309000 |
pilot-004 | de | clean | nvidia/Nemotron-SFT-Agentic-v4 | tool_calling | 10be843d-2d29-4bc4-b58e-d0a502c69ccd | ml-intern-improver-0.3-pilot | IDEAL_TRACE_GUIDELINES v0.3 | true | [
"Seed fired the four comparisons sequentially; the ideal trace batches all four independent calls in one turn (P3 parallel rule).",
"Score-scale inconsistency (89 vs 0.88 vs 87.2) surfaced explicitly instead of silently normalized (P1 depth where it matters).",
"Removed the lesson-planning recommendations (over... | [
{
"role": "system",
"content": ""
},
{
"role": "user",
"content": "Ich entwerfe für meine Vorschulgruppe ein STEAM-Projekt, das Wetterlagen mit Kunststilen vergleicht - etwa Aquarell-Washes für Regentage gegen kräftige Acrylstriche für Sonnentage. Ich brauche für vier Analogie-Paare einen Ähnlic... | [
{
"type": "function",
"function": {
"name": "calculate_similarity",
"description": "This calculates the similarity between two texts in percentage.",
"parameters": {
"type": "object",
"properties": {
"text1": {
"type": "string",
"descriptio... | null | null | null | answered | [] | hf://datasets/flozi00/ideal-trace-pilot-batch-1/data/traces-de.jsonl?row=3 | 8a704159 | flozi00 | 2026-10-25T15:19:05.309000 |
pilot-005 | de | clean | nvidia/Nemotron-SFT-Agentic-v4 | tool_calling | 50dccef1-3c99-4317-a203-3f81d771c30e | ml-intern-improver-0.3-pilot | IDEAL_TRACE_GUIDELINES v0.3 | false | [
"Dropped the seed's ungrounded claim that Delhi Cantt is reached \"roughly 20-22 hours after departure\" - not in the tool result (P4).",
"Kept the honest caveat that the seat preview is for a sample date.",
"User intent Germanized; call and result byte-verbatim."
] | [
{
"role": "system",
"content": ""
},
{
"role": "user",
"content": "Oh, ich helfe gerade meinem Neffen, der nächsten Monat durch Indien reist - er möchte mit dem Golden Temple Mail von Mumbai nach Delhi fahren. Kannst du mir die Zugnummer und die wichtigsten Halte auf der Strecke nennen und ob es... | [
{
"type": "function",
"function": {
"name": "indian_railway_search",
"description": "Search Indian Railway trains by either train number or it's name.",
"parameters": {
"type": "object",
"properties": {
"search": {
"type": "string",
"descri... | null | null | null | answered | [] | hf://datasets/flozi00/ideal-trace-pilot-batch-1/data/traces-de.jsonl?row=4 | 8a704159 | flozi00 | 2026-10-25T15:19:05.309000 |
pilot-010 | de | repaired | nvidia/Nemotron-SFT-Agentic-v4 | tool_calling | 9ec26c14-0df1-4760-b559-00007128a6bb | ml-intern-improver-0.3-pilot | IDEAL_TRACE_GUIDELINES v0.3 | false | [
"Seed structurally broken: it ended with an unanswered emergencyknowledge tool call (pairing violation, §2).",
"Ideal path routes around it: symptomsearch already returned per-condition emergency steps, so the second call was redundant (P3) - not replaced, not invented (P4).",
"No failure appears in any assista... | [
{
"role": "system",
"content": ""
},
{
"role": "user",
"content": "Ich restauriere ein Wandbild für das Seniorenzentrum, aber mein Rücken macht wieder Probleme - ziehende Krämpfe auf der linken Seite beim Bücken plus anhaltendes Ziehen in der Lendenwirbelsäule. Ich will gleich meinen heutigen Ei... | [
{
"type": "function",
"function": {
"name": "symptomsearch",
"description": "An API that searches for medical conditions based on symptom descriptions.",
"parameters": {
"type": "object",
"properties": {
"symptom_description": {
"type": "string",
... | null | null | null | answered | [] | hf://datasets/flozi00/ideal-trace-pilot-batch-1/data/traces-de.jsonl?row=5 | 8a704159 | flozi00 | 2026-10-25T15:19:05.309000 |
pilot-006 | en | clean | nvidia/Nemotron-SFT-Agentic-v4 | tool_calling | 0fbcb992-db09-40cb-951f-41c4eb2db3b5 | ml-intern-improver-0.3-pilot | IDEAL_TRACE_GUIDELINES v0.3 | false | [
"Seed reasoning rambled through a joke-by-joke evaluation; reduced to the selection rationale.",
"Dropped the poster-layout section and the closing offer list (over-answering; user asked for jokes and a quote).",
"User message and tool calls/results byte-verbatim from seed."
] | [
{
"role": "system",
"content": ""
},
{
"role": "user",
"content": "I need some good jokes and quotes for my kids’ school project—they’re putting together a bilingual “funny & wise” poster. First, please get me two jokes that would work well translated into Spanish, and then find a famous quote i... | [
{
"type": "function",
"function": {
"name": "random_ten_joke",
"description": "It returns ten random jokes.",
"parameters": {
"type": "object",
"properties": {},
"required": []
}
}
},
{
"type": "function",
"function": {
"name": "random_qu... | null | null | null | answered | [] | hf://datasets/flozi00/ideal-trace-pilot-batch-1/data/traces-en.jsonl?row=0 | 8a704159 | flozi00 | 2026-10-25T16:14:20.955000 |
pilot-007 | en | clean | nvidia/Nemotron-SFT-Agentic-v4 | tool_calling | 20470142-e827-4db5-927d-ea93e9fd8ecf | ml-intern-improver-0.3-pilot | IDEAL_TRACE_GUIDELINES v0.3 | false | [
"Clean seed - the clarifying question is legitimate (no host name was given) and kept.",
"Tightened phrasing; removed the numbered plan restatement and the offer list.",
"tools reduced to the invoked subset per manifest policy."
] | [
{
"role": "system",
"content": ""
},
{
"role": "user",
"content": "I need to check my granddaughter's school website for safety. I want to look up the site's history first, then get a recent scan of its security tests to make sure it's safe for kids to use. It should show all the subtests includ... | [
{
"type": "function",
"function": {
"name": "get_grade_distribution",
"description": "Get the grade distribution.",
"parameters": {
"type": "object",
"properties": {},
"required": []
}
}
}
] | null | null | null | answered | [] | hf://datasets/flozi00/ideal-trace-pilot-batch-1/data/traces-en.jsonl?row=1 | 8a704159 | flozi00 | 2026-10-25T16:14:20.955000 |
pilot-008 | en | clean | nvidia/Nemotron-SFT-Agentic-v4 | tool_calling | 2b9d86e7-3081-40bd-bc5e-987b3cda4178 | ml-intern-improver-0.3-pilot | IDEAL_TRACE_GUIDELINES v0.3 | false | [
"Clean seed. The logistics tools all depend on the user's choice, origin and dates, so the final turn asks instead of guessing (P3).",
"Tightened the final answer; dropped the redundant restatement of next steps.",
"tools reduced to the invoked subset per manifest policy."
] | [
{
"role": "system",
"content": ""
},
{
"role": "user",
"content": "I'm planning to attend an AI conference to expand my knowledge for 5G deployment work, but I need to coordinate it with my existing travel plans. Can you search for upcoming AI conferences, then based on what you find, help me pl... | [
{
"type": "function",
"function": {
"name": "aiconferencesearch",
"description": "An API to search for conferences related to artificial intelligence based on the 'meeting_name' parameter passed.",
"parameters": {
"type": "object",
"properties": {
"meeting_name": ... | null | null | null | answered | [] | hf://datasets/flozi00/ideal-trace-pilot-batch-1/data/traces-en.jsonl?row=2 | 8a704159 | flozi00 | 2026-10-25T16:14:20.955000 |
pilot-009 | en | clean | nvidia/Nemotron-SFT-Agentic-v4 | tool_calling | 9ad317d2-5e20-42f8-a5d2-e898538141a4 | ml-intern-improver-0.3-pilot | IDEAL_TRACE_GUIDELINES v0.3 | true | [
"Seed fired dog_sub_breed and dog_facts in two sequential turns; they are independent, so the ideal trace batches them (P3).",
"Removed the sycophantic opening and all emoji; compacted the table; kept 3 of 5 facts; one focused follow-up question instead of two plus an offer."
] | [
{
"role": "system",
"content": ""
},
{
"role": "user",
"content": "Hey, I was just sitting here with my wife enjoying the view of the Ohio River. We were reminiscing about the time we visited the art exhibit at the local gallery and got into a conversation about how artists portray certain trait... | [
{
"type": "function",
"function": {
"name": "dog_breed",
"description": "Returns a list of dog breeds.",
"parameters": {
"type": "object",
"properties": {},
"required": []
}
}
},
{
"type": "function",
"function": {
"name": "dog_sub_breed"... | null | null | null | answered | [] | hf://datasets/flozi00/ideal-trace-pilot-batch-1/data/traces-en.jsonl?row=3 | 8a704159 | flozi00 | 2026-10-25T16:14:20.955000 |
Ideal-Trace Pilot Batch 1
Pilot batch of 10 idealized agent traces built per IDEAL_TRACE_GUIDELINES v0.3 from
nvidia/Nemotron-SFT-Agentic-v4
(tool_calling split) seed records. This is an evaluation reference, not a training corpus by
official scope; the guidelines' §4 contract keeps it valid as an SFT target if that decision is made
later.
- 6 German / 4 English. German traces have German user and assistant prose; tool calls, arguments and results are kept byte-verbatim from the seed (see open item 2 in the manifest).
- 9 clean, 1 repaired (
pilot-010: the seed ended with an unanswered tool call; the ideal path routes around the redundant call instead of inventing a result). - 0 retained recoveries (§4.1 carve-out unused in this batch; failure density 0.0).
Files
| File | Content |
|---|---|
data/traces-de.jsonl |
6 German-language ideal traces |
data/traces-en.jsonl |
4 English-language ideal traces |
manifest.json |
Batch manifest: counts, rejected seeds, per-trace changes, open items |
Seed provenance
Seed records come from nvidia/Nemotron-SFT-Agentic-v4 (source_uuid on every record). The seed
traces are teacher-generated (Qwen3.5-397B-A17B, GLM-5.2) with simulated tool outputs; this batch
treats the seed record as the environment-of-record, so every tool result is copied byte-verbatim
from the seed. Two structurally broken seeds (records ending in unanswered tool calls) were rejected
and are listed with reasons in the manifest.
Masking contract
If reused for SFT: loss on assistant spans only (reasoning_content + content + tool_call arguments of assistant messages); user messages, system prompts and tool results are masked context. Retained failures are context, never target (guidelines §4).
Open items (owner decisions pending)
- Spec amendment for treating teacher-simulated outputs as environment-of-record (required for any Agentic-v4 use under P4).
- P5 vs P4: single-language rule governs authored prose; verbatim tool results stay in the seed's language. Needs a guidelines clarification.
- SWE-v3.5 branch deferred: ~230 KB average rows need the compute path; OpenCode system prompt decision (§7 item 1) pending.
- tools field policy: this batch lists the invoked subset; decide canonical policy before scaling.
Licenses
Derived from OpenMDW-1.1-licensed seed data; this batch is distributed under OpenMDW-1.1 as well.
- Downloads last month
- 54