SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2603.23738 · 2026

BXRL: Behavior-Explainable Reinforcement Learning

Cameron Allen, Yotam Amitai, Reuth Mirsky, Ram Rachum, Yonatan Nakar

arXiv · PDF · Open in the Atlas

Code that ran

7 of 11 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; 2 honoured the contract we drafted, 2 violated it and 0 have no contract verdict; 3 failures of our instrument; 4 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/2603.23738.json · badge: /api/ran/2603.23738.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
HumanCompatibleAI/HighJax canonical (classifier label)
classifier basis: census
7 ran · 4 no recorded run
FunctionStatusWhere it lives
desired_gap Ran
ran, honoured the contract we drafted
HumanCompatibleAI/HighJax/highjax/idm.py
code served (permissive licence) · get_code("685f9c6ce26d0c9f")
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 ArrayImpl, 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 jax
import jax.numpy as jnp

# Create small JAX arrays for the function parameters
ego_speed = jnp.array(15.0)  # scalar speed in m/s
front_speed = jnp.array(10.0)  # scalar speed in m/s
ego_heading = jnp.array(0.0)  # scalar heading angle in radians
front_heading = jnp.array(0.0)  # scalar heading angle in radians

ARGS = {
    "ego_speed": ego_speed,
    "front_speed": front_speed,
    "ego_heading": ego_heading,
    "front_heading": front_heading,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, 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
lane_local_coordinates Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
HumanCompatibleAI/HighJax/highjax/lanes.py
code served (permissive licence) · get_code("5ac72d3bd951ba21")
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 tuple, ndim 1, 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 jax
import jax.numpy as jnp

# Create lane parameters array with at least 11 elements
# Indices: 0=START_X, 1=START_Y, 7=DIR_X, 8=DIR_Y, 9=DIR_LAT_X, 10=DIR_LAT_Y
lane_params = jnp.array([
    1.0,    # START_X (index 0)
    2.0,    # START_Y (index 1)
    0.0,    # filler (index 2)
    0.0,    # filler (index 3)
    0.0,    # filler (index 4)
    0.0,    # filler (index 5)
    0.0,    # filler (index 6)
    1.0,    # DIR_X (index 7)
    0.0,    # DIR_Y (index 8)
    0.0,    # DIR_LAT_X (index 9)
    1.0,    # DIR_LAT_Y (index 10)
])

world_x = 3.5
world_y = 4.2

ARGS = {
    "lane_params": lane_params,
    "world_x": world_x,
    "world_y": world_y,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
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
lane_position Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
HumanCompatibleAI/HighJax/highjax/lanes.py
code served (permissive licence) · get_code("0fc0a4dc002580b1")
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 ArrayImpl, ndim 1, dtype family float, shape (2,) 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 jax
import jax.numpy as jnp

# lane_params is a JAX array with at least 11 elements (indices 0-10)
# The function accesses: START_X=0, START_Y=1, DIR_X=7, DIR_Y=8, DIR_LAT_X=9, DIR_LAT_Y=10
lane_params = jnp.array([
    1.0,    # START_X (0)
    2.0,    # START_Y (1)
    0.0,    # (2)
    0.0,    # (3)
    0.0,    # (4)
    0.0,    # (5)
    0.0,    # (6)
    1.0,    # DIR_X (7)
    0.0,    # DIR_Y (8)
    0.0,    # DIR_LAT_X (9)
    1.0,    # DIR_LAT_Y (10)
])

longitudinal = 5.0
lateral = 2.5

ARGS = {
    "lane_params": lane_params,
    "longitudinal": longitudinal,
    "lateral": lateral,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 1, dtype family float, shape (2,) — shape (2,) 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
make_lane_params Ran
ran, honoured the contract we drafted
HumanCompatibleAI/HighJax/highjax/lanes.py
code served (permissive licence) · get_code("9606e9edd530e79a")
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 ArrayImpl, ndim 1, dtype family float, shape (11,) 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 jax
import jax.numpy as jnp

ARGS = {
    "start_x": 0.0,
    "start_y": 0.0,
    "end_x": 10.0,
    "end_y": 5.0,
    "width": 4.0,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 1, dtype family float, shape (11,) — shape (11,) 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
make_vehicle_state Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
HumanCompatibleAI/HighJax/highjax/kinematics.py
code served (permissive licence) · get_code("7645af05f2246aa6")
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 ArrayImpl, ndim 1, 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 jax
import jax.numpy as jnp

DEFAULT_TARGET_SPEEDS = jnp.array([20.0, 25.0, 30.0])
LANE_CHANGE_DELAY = 1.0

def speed_to_index(speed: float, target_speeds: jax.Array=DEFAULT_TARGET_SPEEDS) -> jax.Array:
    """Convert speed to index in target_speeds array."""
    x = (speed - target_speeds[0]) / (target_speeds[-1] - target_speeds[0] + 1e-08)
    return jnp.clip(jnp.round(x * (target_speeds.size - 1)), 0, target_speeds.size - 1).astype(jnp.int32)

def make_vehicle_state(x: float, y: float, heading: float, speed: float, target_lane_idx: int, target_speeds: jax.Array=DEFAULT_TARGET_SPEEDS, idm_delta: float=4.0) -> jax.Array:
    """
    Create a vehicle state array.

    Args:
        x, y: Position [m]
        heading: Heading angle [rad]
        speed: Forward speed [m/s]
        target_lane_idx: Index of target lane
        target_speeds: Array of allowed target speeds [m/s]
        idm_delta: IDM acceleration exponent (randomized per vehicle)

    Returns:
        Vehicle state array of shape (VEHICLE_STATE_SIZE,)
    """
    speed_index = speed_to_index(speed, target_speeds)
    target_speed = target_speeds[speed_index]
    lane_change_timer = (x + y) * jnp.pi % LANE_CHANGE_DELAY
    return jnp.array([x, y, heading, speed, target_lane_idx, speed_index, target_speed, lane_change_timer, idm_delta, 0.0])

ARGS = {
    "x": 10.5,
    "y": 15.3,
    "heading": 0.785,
    "speed": 25.0,
    "target_lane_idx": 1,
    "target_speeds": jnp.array([20.0, 25.0, 30.0]),
    "idm_delta": 4.0,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 1, dtype family float — no output shape is stated: observe_signature.py could not observe it (no_result)
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
speed_to_index Ran
ran, violated the contract we drafted
HumanCompatibleAI/HighJax/highjax/kinematics.py
code served (permissive licence) · get_code("8af42708b36ed1a9")
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 ArrayImpl, 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 jax
import jax.numpy as jnp

# Set seed for reproducibility
jax.config.update('jax_default_prng_impl', 'threefry2x32')

# Define arguments for the speed_to_index function
ARGS = {
    "speed": 25.5,
    "target_speeds": jnp.array([20.0, 25.0, 30.0])
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, 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
wrap_to_pi Ran
ran, violated the contract we drafted
HumanCompatibleAI/HighJax/highjax/kinematics.py
code served (permissive licence) · get_code("0844c80d6d0b8c6f")
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 ArrayImpl, ndim 1, dtype family float, shape (4,) 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 jax
import jax.numpy as jnp

# Create test angles - some in different ranges that need wrapping
angle_1d = jnp.array(0.0)  # Single scalar
angle_2d = jnp.array([0.0, 3.14159, 6.28318, -3.14159])  # 1D array
angle_3d = jnp.array([[0.0, 1.57], [3.14159, 6.28318]])  # 2D array

ARGS = {
    "angle": angle_2d
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 1, dtype family float, shape (4,) — shape (4,) 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
find_front_vehicle Not yet run
unverified (harvested, no recorded run)
HumanCompatibleAI/HighJax/highjax/idm.py
code served (permissive licence) · get_code("61b9697bc108bb68")
How it ran

no invocation recorded.

idm_acceleration Not yet run
unverified (harvested, no recorded run)
HumanCompatibleAI/HighJax/highjax/idm.py
code served (permissive licence) · get_code("50df9e643d646c65")
How it ran

no invocation recorded.

step_physics Not yet run
unverified (harvested, no recorded run)
HumanCompatibleAI/HighJax/highjax/stepper.py
code served (permissive licence) · get_code("c4a7a3e7ca4d845f")
How it ran

no invocation recorded.

step_physics_with_sub_states Not yet run
unverified (harvested, no recorded run)
HumanCompatibleAI/HighJax/highjax/stepper.py
code served (permissive licence) · get_code("ce93e07688f425eb")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 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 11 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

A major challenge of Reinforcement Learning is that agents often learn undesired behaviors that seem to defy the reward structure they were given. Explainable Reinforcement Learning (XRL) methods can answer queries such as "explain this specific action", "explain this specific trajectory", and "explain the entire policy". However, XRL lacks a formal definition for behavior as a pattern of actions across many episodes. We provide such a definition, and use it to enable a new query: "Explain this behavior". We present Behavior-Explainable Reinforcement Learning (BXRL), a new problem formulation that treats behaviors as first-class objects. BXRL defines a behavior measure as any function m : Π → R, allowing users to precisely express the pattern of actions that they find interesting and measure how strongly the policy exhibits it. We define contrastive behaviors that reduce the question "why does the agent prefer a to a ′ ?" to "why is m(π) high?" which can be explored with differentiation. We do not implement an explainability method; we instead analyze three existing methods and propose how they could be adapted to explain behavior. We present a port of the HighwayEnv driving environment to JAX, which provides an interface for defining, measuring, and differentiating behaviors with respect to the model parameters.

For agents

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

get_harvested_code_for_paper("2603.23738")
get_code_for_paper("2603.23738")
have("2603.23738")

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

curl https://syntology.ai/api/ran/2603.23738.json

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

[![Syntology run record](https://syntology.ai/api/ran/2603.23738.svg)](https://syntology.ai/paper/2603.23738)

Connect an agent — have() is free.