Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code: DatasetGenerationError
Exception: TypeError
Message: Couldn't cast array of type
struct<issue0: string, issue1: string, issue2: string, issue3: string, issue4: string, Split: string, Allocation: string>
to
{'issue0': Value('string'), 'issue1': Value('string'), 'issue2': Value('string'), 'issue3': Value('string'), 'issue4': Value('string')}
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1816, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 613, in wrapped
for item in generator(*args, **kwargs):
~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2369, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2303, in cast_table_to_schema
cast_array_to_feature(
~~~~~~~~~~~~~~~~~~~~~^
table[name] if name in table_column_names else pa.array([None] * len(table), type=schema.field(name).type),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
feature,
^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1852, in wrapper
return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
~~~~^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2059, in cast_array_to_feature
_c(array.field(name) if name in array_fields else null_array, subfeature)
~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1854, in wrapper
return func(array, *args, **kwargs)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2059, in cast_array_to_feature
_c(array.field(name) if name in array_fields else null_array, subfeature)
~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1854, in wrapper
return func(array, *args, **kwargs)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2149, in cast_array_to_feature
raise TypeError(f"Couldn't cast array of type\n{_short_str(array.type)}\nto\n{_short_str(feature)}")
TypeError: Couldn't cast array of type
struct<issue0: string, issue1: string, issue2: string, issue3: string, issue4: string, Split: string, Allocation: string>
to
{'issue0': Value('string'), 'issue1': Value('string'), 'issue2': Value('string'), 'issue3': Value('string'), 'issue4': Value('string')}
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1683, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1869, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
prompt list | chosen list | rejected list | meta dict | experiment-name string |
|---|---|---|---|---|
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: proposing this package is the best of them, worth 46.0 surplus points. The move I was about to make is worth 43.7, which gives up 2.3 points.\", \"action\": \"propose\", \"deal\": {\"issue0\"... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"I need to propose an initial deal to set the negotiation moving. I'll aim for an offer that meets my threshold of 99 points. Let's calculate my points for different combinations.\", \"message\": \"I'll propose a package that meets my threshold. ... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-0497e1d304",
"instance_id": "scorable_negotiation-L0-8a241fe1b4",
"seed": 1,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: rejecting offer P1 is the best of them, worth 95.0 surplus points. The move I was about to make is worth 46.0, which gives up 49.0 points.\", \"action\": \"reject\", \"offer_id\": \"P1\"}\n``... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"The current offer P1 scores 145, which is above my threshold of 99. I can accept it, but I need to check if others will also accept it. If not, I might have to propose a better deal.\", \"message\": \"P1 is above my threshold. I can accept it, b... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-0497e1d304",
"instance_id": "scorable_negotiation-L0-8a241fe1b4",
"seed": 1,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: proposing this package is the best of them, worth 119.0 surplus points. The move I was about to make is worth 113.0, which gives up 6.0 points.\", \"action\": \"propose\", \"deal\": {\"issue0... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"The current offer P1 scores 141, which is above my threshold of 94. I need to check if this offer meets other parties' thresholds. However, I should consider if there are better options or if adjustments can be made to secure more support.\", \"... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-0497e1d304",
"instance_id": "scorable_negotiation-L0-8a241fe1b4",
"seed": 1,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: accepting offer P1 is the best of them, worth 74.0 surplus points. The move I was about to make is worth 70.3, which gives up 3.7 points.\", \"action\": \"accept\", \"offer_id\": \"P1\"}\n```... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"P1 and P2 are the same deal. I need to check if this meets my threshold. My threshold is 57, and this deal gives me 131, which is above my threshold. However, I need to ensure all parties accept it. Blake has accepted, but others like Casey and ... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-0497e1d304",
"instance_id": "scorable_negotiation-L0-8a241fe1b4",
"seed": 1,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: rejecting offer P1 is the best of them, worth 113.0 surplus points. The move I was about to make is worth 47.0, which gives up 66.0 points.\", \"action\": \"reject\", \"offer_id\": \"P1\"}\n`... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"The current offer P3 meets my threshold of 94 points. I will accept it to finalize the deal.\", \"message\": \"I accept offer P3 as it meets my requirements and others have already accepted it.\", \"action\": \"accept\", \"offer_id\": \"P3\"}\n`... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-0497e1d304",
"instance_id": "scorable_negotiation-L0-8a241fe1b4",
"seed": 1,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: rejecting offer P1 is the best of them, worth 95.0 surplus points. The move I was about to make is worth 46.0, which gives up 49.0 points.\", \"action\": \"reject\", \"offer_id\": \"P1\"}\n``... | [
{
"role": "assistant",
"content": "{\"scratchpad\": \"P3 meets my threshold and has enough support. I will accept it to close the deal.\", \"message\": \"P3 is acceptable to me and has sufficient support. I accept it.\", \"action\": \"accept\", \"offer_id\": \"P3\"}"
}
] | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-0497e1d304",
"instance_id": "scorable_negotiation-L0-8a241fe1b4",
"seed": 1,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: rejecting offer P1 is the best of them, worth 68.4 surplus points. The move I was about to make is worth -14.0, which gives up 82.4 points.\", \"action\": \"reject\", \"offer_id\": \"P1\"}\n`... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"I need to accept the current offer P3 even though it's below my threshold. However, since it's the only option left, I'll have to accept it.\", \"message\": \"I accept offer P3 despite it being below my threshold as it's the only viable option l... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-0497e1d304",
"instance_id": "scorable_negotiation-L0-8a241fe1b4",
"seed": 1,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: rejecting offer P1 is the best of them, worth 113.0 surplus points. The move I was about to make is worth 47.0, which gives up 66.0 points.\", \"action\": \"reject\", \"offer_id\": \"P1\"}\n`... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"P3 meets my threshold and has sufficient support. I will accept it.\", \"message\": \"P3 is acceptable and has enough support. I accept it.\", \"action\": \"accept\", \"offer_id\": \"P3\"}\n```"
}
] | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-0497e1d304",
"instance_id": "scorable_negotiation-L0-8a241fe1b4",
"seed": 1,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: proposing this package is the best of them, worth 72.0 surplus points. The move I was about to make is worth 68.4, which gives up 3.6 points.\", \"action\": \"propose\", \"deal\": {\"issue0\"... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"I need to table a proposal that meets my threshold of 134 points. Let me calculate the total points for each combination of options.\", \"message\": \"I'll propose a package that meets my threshold of 134 points.\", \"action\": \"propose\", \"de... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-07e304425e",
"instance_id": "scorable_negotiation-L0-8a241fe1b4",
"seed": 4,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: rejecting offer P1 is the best of them, worth 95.0 surplus points. The move I was about to make is worth 75.0, which gives up 20.0 points.\", \"action\": \"reject\", \"offer_id\": \"P1\"}\n``... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"I need to ensure the deal meets my threshold of 99 points. The current offer P1 scores 174 for me, which is well above my threshold. I should accept it to close the deal.\", \"message\": \"Offer P1 meets my threshold. I accept it.\", \"action\":... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-07e304425e",
"instance_id": "scorable_negotiation-L0-8a241fe1b4",
"seed": 4,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: proposing this package is the best of them, worth 119.0 surplus points. The move I was about to make is worth 113.0, which gives up 6.0 points.\", \"action\": \"propose\", \"deal\": {\"issue0... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"The current offer P1 scores 83, which is below my threshold of 94. I need to either negotiate a better deal or walk away.\", \"message\": \"I need a better deal to meet my threshold. Let's try to improve P1.\", \"action\": \"propose\", \"deal\":... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-07e304425e",
"instance_id": "scorable_negotiation-L0-8a241fe1b4",
"seed": 4,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: proposing this package is the best of them, worth 100.0 surplus points. The move I was about to make is worth 95.0, which gives up 5.0 points.\", \"action\": \"propose\", \"deal\": {\"issue0\... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"I need to propose a deal that meets my acceptance threshold of 99 points. Let me calculate the minimum combination of options that can reach this threshold.\", \"message\": \"I will propose a deal that meets my acceptance threshold of 99 points.... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-1dca6314ac",
"instance_id": "scorable_negotiation-L0-068e29d166",
"seed": 7,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: rejecting offer P1 is the best of them, worth 113.0 surplus points. The move I was about to make is worth 0.0, which gives up 113.0 points.\", \"action\": \"reject\", \"offer_id\": \"P1\"}\n`... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"The current offer P1 scores 94, which meets my threshold. I can accept it, but I need to check if all other parties accept it. Since Blake has a veto, everyone else must agree.\", \"message\": \"I can accept P1, but I need others to agree for it... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-1dca6314ac",
"instance_id": "scorable_negotiation-L0-068e29d166",
"seed": 7,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: proposing this package is the best of them, worth 46.0 surplus points. The move I was about to make is worth 43.7, which gives up 2.3 points.\", \"action\": \"propose\", \"deal\": {\"issue0\"... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"P1 scores 77, which is below my threshold of 99. I need to propose a better deal.\", \"message\": \"P1 is below my threshold. I need to propose a better deal.\", \"action\": \"propose\", \"deal\": {\"issue0\": \"opt1\", \"issue1\": \"opt0\", \"i... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-1dca6314ac",
"instance_id": "scorable_negotiation-L0-068e29d166",
"seed": 7,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: proposing this package is the best of them, worth 74.0 surplus points. The move I was about to make is worth 70.3, which gives up 3.7 points.\", \"action\": \"propose\", \"deal\": {\"issue0\"... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"P1 and P2 both score 94, which is above my threshold of 57. I need to consider which option aligns better with my preferences. Issue1 is crucial for me as it's worth 57 points. P1 and P2 both have opt0 for issue1, which is ideal. I should check ... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-1dca6314ac",
"instance_id": "scorable_negotiation-L0-068e29d166",
"seed": 7,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: proposing this package is the best of them, worth 72.0 surplus points. The move I was about to make is worth 68.4, which gives up 3.6 points.\", \"action\": \"propose\", \"deal\": {\"issue0\"... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"I need to propose a deal that meets my acceptance threshold. P1 and P3 are at 159, which is above my threshold of 134. P2 is at 160. I should propose a new deal that balances my preferences and meets the threshold.\", \"message\": \"I will propo... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-1dca6314ac",
"instance_id": "scorable_negotiation-L0-068e29d166",
"seed": 7,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: proposing this package is the best of them, worth 119.0 surplus points. The move I was about to make is worth 113.0, which gives up 6.0 points.\", \"action\": \"propose\", \"deal\": {\"issue0... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"P4 offers 130 points, which is above my threshold of 94. I need to evaluate if it meets my preferences. Issue4 is opt1 (36 points) and issue3 is opt2 (0 points), which is worse than opt0. I should consider proposing a better deal.\", \"message\"... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-1dca6314ac",
"instance_id": "scorable_negotiation-L0-068e29d166",
"seed": 7,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: rejecting offer P1 is the best of them, worth 70.3 surplus points. The move I was about to make is worth 37.0, which gives up 33.3 points.\", \"action\": \"reject\", \"offer_id\": \"P1\"}\n``... | [
{
"role": "assistant",
"content": "{\"scratchpad\": \"All offers are scoring 94 points, which is above my threshold of 57. I need to find an offer that balances my preferences. P3 and P5 are similar, but P5 has issue3=opt1 and issue4=opt1 which might be worse for me. I'll accept P3.\", \"message\": \"I acce... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-1dca6314ac",
"instance_id": "scorable_negotiation-L0-068e29d166",
"seed": 7,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: accepting offer P2 is the best of them, worth 114.0 surplus points. The move I was about to make is worth 24.0, which gives up 90.0 points.\", \"action\": \"accept\", \"offer_id\": \"P2\"}\n`... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"P5 is the only offer that meets my threshold of 99 points. It provides a good balance of options for me. I need to accept it to finalize the deal.\", \"message\": \"P5 meets my threshold and I accept it.\", \"action\": \"accept\", \"offer_id\": ... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-1dca6314ac",
"instance_id": "scorable_negotiation-L0-068e29d166",
"seed": 7,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: rejecting offer P1 is the best of them, worth 113.0 surplus points. The move I was about to make is worth 36.0, which gives up 77.0 points.\", \"action\": \"reject\", \"offer_id\": \"P1\"}\n`... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"P4 and P5 both score 130, which is above my threshold of 94. I need to check if they meet my preferences. P4 has issue3=opt2 (0 points) and P5 has issue3=opt0 (0 points). My score for issue3 is 0 for both, so it doesn't affect my total. I'll acc... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-1dca6314ac",
"instance_id": "scorable_negotiation-L0-068e29d166",
"seed": 7,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: proposing this package is the best of them, worth 57.2 surplus points. The move I was about to make is worth 54.4, which gives up 2.9 points.\", \"action\": \"propose\", \"deal\": {\"issue0\"... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"I need to propose an initial deal to get the negotiation moving. Since I have a clear preference for issue3 (opt0) and issue1 (opt0), I'll start by proposing a package that includes these options. I'll also consider the other issues to ensure th... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-20d5b3e27f",
"instance_id": "scorable_negotiation-L0-2c7c7b5822",
"seed": 5,
... | p4-divergence-dpo-v2 |
[
{
"role": "system",
"content": "You are one of 6 parties negotiating a single multi-issue agreement. Exactly one option must be chosen for every issue; the chosen options together form the deal.\n\nParties at the table:\n- Avery\n- Blake (veto party — no deal passes without you)\n- Casey\n- Devon\n- Ember\n... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"Comparing every move open to me with 4 round(s) left: rejecting offer P1 is the best of them, worth 78.7 surplus points. The move I was about to make is worth 58.8, which gives up 19.9 points.\", \"action\": \"reject\", \"offer_id\": \"P1\"}\n``... | [
{
"role": "assistant",
"content": "```json\n{\"scratchpad\": \"I need to make a proposal that meets my threshold of 46.1719. The current offer P1 gives me 105 points, which is well above my threshold. However, I should consider other options to ensure the deal is acceptable to all parties, especially Blake ... | {
"model": "Qwen3-4B",
"model_label": "Qwen3-4B",
"model_id": "Qwen/Qwen3-4B",
"cell_id": "Qwen3-4B_all_llm",
"table": "all_llm",
"game": "scorable",
"scaffold": "canonical",
"episode_id": "scorable_negotiation-moves_chat-20d5b3e27f",
"instance_id": "scorable_negotiation-L0-2c7c7b5822",
"seed": 5,
... | p4-divergence-dpo-v2 |
End of preview.
Rational-Agent Divergence DPO Pairs
This is the corrected p4_dataset_v2 preference dataset built from the
validity-gated P2 negotiation campaign. Each row pairs the model's stored
per-turn rendered prompt and verbatim action with an executable
best-response-oracle action rendered in the same wire format.
All 12,761 rows contain the required experiment-name field, set to
p4-divergence-dpo-v2.
Splits
| Split | Rows | Purpose |
|---|---|---|
train |
5,533 | Episode-level training split |
eval |
649 | Episode-level validation split |
holdout_model |
3,866 | OLMo model holdout |
holdout_instance |
2,713 | Game-2 instance holdout |
| Total | 12,761 | Pairwise overlaps are zero |
The upload includes stats.json, build logs, v0/v1 diff audits, and a
SHA-256 artifact_manifest.json. The original prompts have 100% stored-view
coverage. Chosen actions were parser-validated by the dataset build.
Row schema
experiment-name: stable append-friendly experiment identifier.prompt: the exact stored rendered participant view at the divergent turn.chosen: best-response-oracle action rendered in the model's action format; the full form may include a rationale synthesized only from oracle numbers.rejected: the model's original emission verbatim.meta: model, episode, game, arm, scaffold, information condition, turn, regret, flags, split criteria, and provenance fields.
Important use restrictions and findings
The full dataset is an audit/research corpus, not a recommendation to train on every row. The preregistered pilot training subset was:
own model AND phase == "turn" AND info == "full" AND scaffold == "canonical"
Three hazards were deliberately excluded from the pilot mix:
- forced-final-vote labels could choose a different offer from the explicit vote target;
- PRIVATE labels use an omniscient oracle with score sheets unavailable to the acting model;
- scaffolded emissions confound training with the prompt intervention.
The paired action-only control showed that 95.5% of the naive chosen/rejected separability came from oracle-rationale prose, not the decision. A single-game preliminary outcome gain did not replicate on three unseen games. The final pilot conclusion is negative: these data do not establish that divergence-localized DPO improves negotiation outcomes out of distribution.
The v2 rebuild corrects proposer-seat best-response pricing and a replay annotation contamination bug. It is bit-identical to v1 in prompt/chosen/rejected/weight/regret across all 12,761 pairs; only an auxiliary flag changed. The included diff files document the complete audit chain.
Exact regeneration command
From experiments/rational_agents/ in the ii_mats repository:
UV_LINK_MODE=copy uv run python p4_dataset.py \
--manifest p2_selected_manifest.json \
--out-dir /nlp/scr/siddharth/ii_mats/rational_agents/p4_dataset_v2 \
--annotations-name annotations_v1 \
--provenance-note "Corrected v2; see metadata/build.log for the full audit note." \
--min-regret 1e-6 \
--tables all_llm mixed \
--chosen-style rationale \
--chosen-message drop \
--holdout-model OLMo-3-7B \
--holdout-instance-group game2 \
--instance-bank instances \
--eval-frac 0.1 \
--split-seed 20260727 \
--weight-cap 5.0 \
--weight-scope per_game \
--n-examples 3
metadata/build.log preserves the literal original command, including its
long-form correction/audit note. metadata/stats.json preserves resolved
counts, deciles, source revisions, split membership, filtering census, and
parser checks.
To recreate the public upload copy, including experiment-name:
uv run python package_hf_publication.py \
--out /nlp/scr/siddharth/ii_mats/rational_agents/hf_public_20260728
Provenance and cluster paths
Source campaign:
/nlp/scr/siddharth/ii_mats/rational_agents/p2*/
Corrected dataset and audit artifacts:
/nlp/scr/siddharth/ii_mats/rational_agents/p4_dataset_v2/
Training runs and checkpoints:
/nlp/scr/siddharth/ii_mats/rational_agents/p4_pilot_dpo/
Slurm logs:
/nlp/scr/siddharth/ii_mats/rational_agents/_slurm_logs/<jobid>.{out,err}
Relevant Weights & Biases runs:
Models and related data
The pilot targets Qwen/Qwen3-8B. Source episodes also cover:
The public source episodes, transcripts, annotations, and evaluation results
are in the companion dataset
2026.RA.Negotiation-Campaigns.
No license is asserted for this dataset.
- Downloads last month
- 41