SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2606.03048 · 2026

The Value Function Semi-Algebraic Set in POMDPs

Guido Mont, Ryan Anderson

arXiv · PDF · Open in the Atlas

Code that ran

9 of 10 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; 3 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 5 failures of our instrument; 1 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/2606.03048.json · badge: /api/ran/2606.03048.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
ryan-a-anderson/pomdp-value-geometry canonical (classifier label)
classifier basis: census
9 ran · 1 no recorded run
FunctionStatusWhere it lives
dirichlet_sample Ran
ran; our contract draft was wrong, not the code
ryan-a-anderson/pomdp-value-geometry/src/pomdp_localopt.py
code served (permissive licence) · get_code("629bb4ddfcab072c")
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 ndarray, 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 numpy as np

class MockRNG:
    def __init__(self):
        self.rng = np.random.RandomState(42)
    
    def gamma(self, shape, scale):
        return self.rng.gamma(shape, scale)

alpha = np.array([1.0, 2.0, 3.0, 4.0])
rng = MockRNG()

ARGS = {
    "alpha": alpha,
    "rng": rng,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, 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
generate_random_pomdp Ran
ran; our contract draft was wrong, not the code
ryan-a-anderson/pomdp-value-geometry/src/local_optima_experiments.py
code served (permissive licence) · get_code("cadf730b9f604516")
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. 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 numpy as np

# Set seed for reproducibility
np.random.seed(42)

# Create a small random number generator
rng = np.random.default_rng(seed=42)

# Define arguments for generate_random_pomdp
ARGS = {
    'n_states': 3,
    'n_actions': 2,
    'n_obs': 4,
    'gamma': 0.9,
    'rng': rng
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is 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
plot_multistart_comparison Ran
ran; our contract draft was wrong, not the code
ryan-a-anderson/pomdp-value-geometry/src/make_figures.py
code served (permissive licence) · get_code("20878967b64874fe")
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 Figure. 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 numpy as np
import matplotlib.pyplot as plt

class MockPOMDP:
    """Mock POMDP class that supports the interface used by plot_multistart_comparison."""
    
    def __init__(self):
        self.n_states = 2
        self.n_obs = 2
    
    def _get_q_of_x(self, x):
        """Return q values if feasible, None otherwise.
        
        x is a state vector. For feasibility, we just check if the first two
        components are within [-1, 1] (simple box constraint).
        """
        if x[0] ** 2 + x[1] ** 2 <= 1.5:
            return x[:2]
        return None
    
    def optimize_projected_gradient(self, rho, p_init, steps, lr, momentum, clip_grad):
        """Simulate gradient optimization with momentum.
        
        Returns a history dict with 'v' key containing trajectory of shape (steps, 2).
        """
        rng = np.random.default_rng(seed=0)
        v = np.zeros((steps, 2), dtype=float)
        v[0] = np.array([5.5, 5.5], dtype=float)
        
        velocity = np.zeros(2, dtype=float)
        
        for t in range(1, steps):
            grad = rho.copy()
            grad = np.clip(grad, -clip_grad, clip_grad)
            velocity = momentum * velocity - lr * grad
            v[t] = v[t - 1] + velocity
            v[t, 0] = np.clip(v[t, 0], 3.0, 8.0)
            v[t, 1] = np.clip(v[t, 1], 3.0, 8.0)
        
        return {'v': v}


pomdp = MockPOMDP()

ARGS = {
    'pomdp': pomdp,
    'rhos': [[0.5, 1.0], [1.0, 0.5], [-0.5, 0.8]],
    'n_starts': 3,
    'steps': 50,
    'lr': 0.25,
    'momentum': 0.6,
    'clip_grad': 10.0,
    'grid_res': 30,
    'num_iso_levels': 8,
    'seed': 42,
    'out_name': None,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
matplotlib, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Figure — a Figure output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Figure 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
policy_spread_mean Ran
ran, violated the contract we drafted
ryan-a-anderson/pomdp-value-geometry/src/pomdp_localopt.py
code served (permissive licence) · get_code("8472ecf11917d6c5")
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 float, 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 numpy as np

# The function expects pi_list to be a list of numpy arrays
# Each array should have a .size attribute (number of elements)
# The function computes pairwise distances between these policy vectors

# Create a list of small policy vectors (numpy arrays)
# Using distinct sizes to make shapes probed later
pi_1 = np.array([0.25, 0.25, 0.25, 0.25])  # 4 elements - a probability distribution
pi_2 = np.array([0.1, 0.3, 0.4, 0.2])      # 4 elements - another distribution
pi_3 = np.array([0.4, 0.2, 0.2, 0.2])      # 4 elements - another distribution

pi_list = [pi_1, pi_2, pi_3]

ARGS = {
    "pi_list": pi_list
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, 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_rho_sweep Ran
ran; our contract draft was wrong, not the code
ryan-a-anderson/pomdp-value-geometry/src/pomdp_optim_dynamics_very_noisy.py
code served (permissive licence) · get_code("118a31d41e38d436")
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 dict. 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 numpy as np
import torch
import torch.nn as nn

torch.manual_seed(0)

class MockPOMDP:
    """Mock POMDP that implements the ablate_rhos method."""
    
    def ablate_rhos(self, alphas=None, n_starts=15, steps=300, lr=0.2, 
                    momentum=0.65, clip_grad=10.0, seed=42, cluster_tol_v=0.01):
        """Mock implementation that returns a simple result."""
        # This method is called by run_rho_sweep
        # Return a dummy result dict
        return {
            'alphas': alphas,
            'n_starts': n_starts,
            'steps': steps,
            'results': np.random.randn(len(alphas), n_starts)
        }

# Create a mock POMDP instance
pomdp = MockPOMDP()

# Define all arguments for run_rho_sweep
ARGS = {
    'pomdp': pomdp,
    'alphas': None,  # Will use ALPHAS_VERY_NOISY default
    'n_starts': 15,
    'steps': 300,
    'lr': 0.2,
    'momentum': 0.65,
    'clip_grad': 10.0,
    'seed': 42,
    'cluster_tol_v': 0.01
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
run_single_rho_histories Ran
ran; our contract draft was wrong, not the code
ryan-a-anderson/pomdp-value-geometry/src/pomdp_optim_dynamics_very_noisy.py
code served (permissive licence) · get_code("3718085f30b4ad4b")
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 list, 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 numpy as np

class MockPOMDP:
    """Mock POMDP that implements the optimize_projected_gradient method."""
    
    def optimize_projected_gradient(self, rho, p_init, steps, lr, momentum, clip_grad):
        """
        Simulate optimization and return a history dict.
        This mimics what a real POMDP would do.
        """
        history = {
            'rho': rho,
            'p_init': p_init.copy(),
            'steps': steps,
            'lr': lr,
            'momentum': momentum,
            'clip_grad': clip_grad,
            'losses': np.random.randn(steps),
            'params': np.random.randn(steps, 2)
        }
        return history

np.random.seed(0)

pomdp = MockPOMDP()
rho = 0.5
n_inits = 10
steps = 300
lr = 0.2
momentum = 0.65
clip_grad = 10.0
seed = 123

ARGS = {
    'pomdp': pomdp,
    'rho': rho,
    'n_inits': n_inits,
    'steps': steps,
    'lr': lr,
    'momentum': momentum,
    'clip_grad': clip_grad,
    'seed': seed
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list 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
softmax Ran
ran, honoured the contract we drafted
ryan-a-anderson/pomdp-value-geometry/src/local_optima_experiments.py
code served (permissive licence) · get_code("113a0f2ce97567ba")
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 ndarray, ndim 2, dtype family float, shape (2, 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 numpy as np

# Create a small batch of logits for testing softmax
# Shape: (batch_size=2, num_classes=4)
logits = np.array([
    [1.0, 2.0, 3.0, 0.5],
    [0.1, 0.2, 0.3, 0.4]
], dtype=np.float32)

ARGS = {
    "logits": logits
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (2, 4) — shape (2, 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
softmax_columns Ran
ran, honoured the contract we drafted
ryan-a-anderson/pomdp-value-geometry/src/pomdp_localopt.py
code served (permissive licence) · get_code("406d47ca23e7fb6d")
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 ndarray, ndim 2, dtype family float, shape (4, 3) 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 numpy as np

# Create a small test matrix with shape (A, O) where A=4 rows and O=3 columns
# The function normalizes each column independently
X = np.array([
    [1.0, 2.0, 3.0],
    [2.0, 3.0, 1.0],
    [3.0, 1.0, 2.0],
    [1.0, 2.0, 3.0]
], dtype=np.float32)

ARGS = {
    "X": X
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (4, 3) — shape (4, 3) 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
softmax_columns Ran
ran, honoured the contract we drafted
ryan-a-anderson/pomdp-value-geometry/src/pomdp_memory_enhancement.py
code served (permissive licence) · get_code("4c1485ade8a44a74")
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 ndarray, ndim 2, dtype family float, shape (4, 3) 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 numpy as np

# Create a small numpy array with shape (A, Y) = (4, 3)
# This will test the column-wise softmax normalization
X = np.array([
    [1.0, 2.0, 3.0],
    [2.0, 1.0, 1.0],
    [0.5, 0.5, 2.0],
    [1.5, 1.5, 0.5]
], dtype=np.float32)

ARGS = {
    'X': X
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (4, 3) — shape (4, 3) 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
solve_and_grad Not yet run
unverified (harvested, no recorded run)
ryan-a-anderson/pomdp-value-geometry/src/local_optima_experiments.py
code served (permissive licence) · get_code("4a978fc8f5fe499a")
How it ran

no invocation recorded.

Also present in other papers' code

Of 10 samples lifted from this paper's linked repositories, 10 carry a function identity and 0 do not. Of the 9 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 9 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

We study the geometry of feasible value functions in infinite-horizon partially observable Markov decision processes (POMDPs) under memoryless stochastic policies. Our main contribution is a characterization of the feasible set of value functions as a semi-algebraic set, defined by explicit polynomial inequalities determined by the transition dynamics, observation kernel, and reward structure of the POMDP. This result extends prior work for fully observable Markov decision processes, where the feasible set is known to be a polytope, to the substantially more intricate partially observable setting. In contrast to the polyhedral structure arising in MDPs, partial observability induces fundamentally nonlinear constraints, leading to a richer and more complex geometric structure. Our geometric characterization provides new insight into the landscape of policy optimization in both MDPs and POMDPs, and reveals qualitative phenomena unique to partial observability, including the emergence of isolated local maximizers of the long-term reward and their dependence on the initial state distribution.

For agents

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

get_harvested_code_for_paper("2606.03048")
get_code_for_paper("2606.03048")
have("2606.03048")

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

curl https://syntology.ai/api/ran/2606.03048.json

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

[![Syntology run record](https://syntology.ai/api/ran/2606.03048.svg)](https://syntology.ai/paper/2606.03048)

Connect an agent — have() is free.