SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2605.01347 · 2026

MAD-OPD: Breaking the Ceiling in On-Policy Distillation via Multi-Agent Debate

Jun Wang, Qilong Zhang, Qianglong Chen, Jianze Wang, Ying Liu, Yu Cao, Yong Xie, Hua Yang, Jinlong Chen, Xuchun Hu, Single Teacher

arXiv · PDF · Open in the Atlas

Code that ran

5 of 7 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 3 failures of our instrument; 2 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2605.01347.json · badge: /api/ran/2605.01347.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
chiefovoavicii/MAD-OPD canonical (classifier label)
classifier basis: census
3 ran · 2 no recorded run
sierra-research/tau2-bench canonical (classifier label)
classifier basis: census
2 ran · 0 no recorded run
FunctionStatusWhere it lives
audio_bytes_to_string Ran
ran; our contract draft was wrong, not the code
sierra-research/tau2-bench/src/tau2/data_model/audio.py
code served (permissive licence) · get_code("0e1894243144defb")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import base64

# Create a simple bytes object representing audio data
audio_data = b'\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f'

ARGS = {
    'data': audio_data
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
audio_string_to_bytes Ran
ran; our contract draft was wrong, not the code
sierra-research/tau2-bench/src/tau2/data_model/audio.py
code served (permissive licence) · get_code("9bc9c67906873029")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned bytes. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import base64

# Create a valid base64-encoded string representing audio bytes
audio_bytes = b"audio_data_sample"
encoded_string = base64.b64encode(audio_bytes).decode('ascii')

ARGS = {
    "data": encoded_string
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bytes — a bytes output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a bytes output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
generalized_jsd_loss_single Ran
ran; our contract draft was wrong, not the code
chiefovoavicii/MAD-OPD/mad_opd/trainers/mad_opd_core.py
code served (permissive licence) · get_code("c754b06031b4c93b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn.functional as F

torch.manual_seed(0)

# Simple case 1: Full logits (no top-k sidecar)
student_logits_1 = torch.randn(2, 3, 8)  # [batch=2, seq_len=3, vocab_size=8]
teacher_logits_1 = torch.randn(2, 3, 8)  # [batch=2, seq_len=3, vocab_size=8]
labels_1 = torch.tensor([[1, 2, -100], [0, 1, 2]])  # -100 means ignore

ARGS = {
    "student_logits": student_logits_1,
    "teacher_logits": teacher_logits_1,
    "labels": labels_1,
    "beta": 0.5,
    "temperature": 1.0,
    "chunk_size": 2,
    "top_k": None,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
run Ran
ran, honoured the contract we drafted
chiefovoavicii/MAD-OPD/eval/mbpp_plus_eval.py
code served (permissive licence) · get_code("ff018fdd1576b054")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import subprocess

# The function run() takes:
# - cmd: list[str] - a list of command strings
# - env: dict | None - optional environment variables dict
# - cwd: str | None - optional working directory string

# Looking at the function body:
# - cmd is passed to subprocess.run() as the first positional argument
# - env defaults to os.environ.copy() if None
# - cwd is passed to subprocess.run()
# - The function just needs valid types, not necessarily working commands

ARGS = {
    "cmd": ["echo", "hello"],
    "env": None,
    "cwd": None,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
wait_for_vllm Ran
ran, violated the contract we drafted
chiefovoavicii/MAD-OPD/eval/bfcl_eval.py
code served (permissive licence) · get_code("97e7067f2d857daf")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned bool, ndim 0, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import time

# The function wait_for_vllm takes:
# - url: str - a URL string for the vLLM server
# - timeout: int - timeout in seconds (default 600)
#
# The function uses these parameters by:
# - url: concatenates it with '/v1/models' and passes to urllib.request.urlopen
# - timeout: adds it to time.monotonic() to set a deadline
#
# Since this is a network function that would try to connect to a real server,
# we just need to provide valid argument types. The function will fail gracefully
# when it can't connect, which is expected in a test environment.

ARGS = {
    "url": "http://localhost:8000",
    "timeout": 30
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float — no output shape is stated: observe_signature.py could not observe it (call:TimeoutError)
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
multi_teacher_jsd_loss Not yet run
unverified (harvested, no recorded run)
chiefovoavicii/MAD-OPD/mad_opd/trainers/mad_opd_core.py
code served (permissive licence) · get_code("ca7a3c7e38226c29")
How it ran

no invocation recorded.

wait_for_vllm Not yet run
unverified (harvested, no recorded run)
chiefovoavicii/MAD-OPD/eval/tau2_bench_eval.py
code served (permissive licence) · get_code("0e1c52741a32644c")
How it ran

no invocation recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 7 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

On-policy distillation (OPD) trains a student on its own trajectories under tokenlevel teacher supervision, but existing methods are capped by a single-teacher capability ceiling: when the teacher errs, the student inherits the error. OPD also remains largely unexplored in agentic tasks, where per-step errors compound across long trajectories and destabilize training. We propose MAD-OPD (Multi-Agent Debate-driven On-Policy Distillation), which breaks this ceiling by recasting the distillation teacher as a deliberative collective of teachers that debate over the student's on-policy state; the debate produces an emergent collective intelligence that supplies token-level supervision, with each teacher's contribution weighted by its post-debate confidence. To extend OPD to agentic tasks, we also introduce On-Policy Agentic Distillation (OPAD), which adds step-level sampling to stabilize training under multi-step error compounding. We additionally derive a task-adaptive divergence principle, selecting JSD (Jensen-Shannon divergence) for agentic stability and reverse KL (Kullback-Leibler) divergence for code generation, and verify it both theoretically and empirically. Across six teacher-student configurations (Qwen3 and Qwen3.5; 1.7B-14B students, 8B-32B teachers) and five agentic and code benchmarks, MAD-OPD ranks first across all six configurations; on the 14B+8B→4B setting it lifts the agentic average by +2.4% and the code average by +3.7% over the stronger single-teacher OPD. Code is available at: https://github.com/chiefovoavicii/MAD-OPD.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2605.01347")
get_code_for_paper("2605.01347")
have("2605.01347")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2605.01347.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2605.01347.svg)](https://syntology.ai/paper/2605.01347)

Connect an agent — have() is free.