SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2605.24286 · 2026

Faithfulness as Information Flow: Evaluating and Training Faithful Chain-of-Thought Reasoning

Jinghan Jia, Joe Benton, Dept Cse, Eric Easley, Michigan State, Anthropic Anthropic, Fellows Program

arXiv · PDF · Open in the Atlas

Code that ran

16 of 18 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, 3 violated it and 0 have no contract verdict; 11 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/2605.24286.json · badge: /api/ran/2605.24286.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
safety-research/faithful-cot canonical (classifier label)
classifier basis: census
16 ran · 2 no recorded run
FunctionStatusWhere it lives
bind_maybe Ran
ran; our contract draft was wrong, not the code
safety-research/faithful-cot/eval/evaluate_hint_ratio.py
pointer only (licence: NONE) · get_code("708a1fca68a4870c")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

clean_answer_region Ran
ran; our contract draft was wrong, not the code
safety-research/faithful-cot/eval/evaluate_hint_ratio.py
pointer only (licence: NONE) · get_code("f0ef9820db13b00a")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

create_no_cot_mask Ran
ran, violated the contract we drafted
safety-research/faithful-cot/eval/faithfulness_metrics_clean.py
pointer only (licence: NONE) · get_code("4c80010b0ac81240")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

create_no_prompt_mask Ran
ran, violated the contract we drafted
safety-research/faithful-cot/eval/faithfulness_metrics_clean.py
pointer only (licence: NONE) · get_code("0baac420cfc1f41c")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

create_via_cot_mask Ran
ran, violated the contract we drafted
safety-research/faithful-cot/eval/faithfulness_metrics_clean.py
pointer only (licence: NONE) · get_code("8e84c4f46413f75b")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

detect_cot_hacking Ran
ran; our contract draft was wrong, not the code
safety-research/faithful-cot/train/custom_code_reward_hacking.py
pointer only (licence: NONE) · get_code("f3fcee8fd80563ce")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

detect_cot_test_awareness Ran
ran; our contract draft was wrong, not the code
safety-research/faithful-cot/train/custom_code_reward_hacking.py
pointer only (licence: NONE) · get_code("a8df8a8fa74926be")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

find_special_token_positions Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
safety-research/faithful-cot/train/verl/utils/cot_masking.py
pointer only (licence: NONE) · get_code("8374822da8e4ae2a")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

format_prompt Ran
ran; our contract draft was wrong, not the code
safety-research/faithful-cot/eval/evaluate_hacking_ratio.py
pointer only (licence: NONE) · get_code("980d1a07fda8a0db")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

load_generations Ran
ran; our contract draft was wrong, not the code
safety-research/faithful-cot/eval/compute_faithfulness_from_generations.py
pointer only (licence: NONE) · get_code("0c260c1972f416d7")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

load_generations Ran
ran; our contract draft was wrong, not the code
safety-research/faithful-cot/eval/compute_faithfulness_from_generations_clean.py
pointer only (licence: NONE) · get_code("b083595c6a47a03d")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

load_test_data Ran
ran; our contract draft was wrong, not the code
safety-research/faithful-cot/eval/compute_faithfulness_from_generations.py
pointer only (licence: NONE) · get_code("3dce64568cc99197")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

load_test_data Ran
ran; our contract draft was wrong, not the code
safety-research/faithful-cot/eval/evaluate_hacking_ratio.py
pointer only (licence: NONE) · get_code("c0afdae4075d5d9b")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

mk_answer_region Ran
ran; our contract draft was wrong, not the code
safety-research/faithful-cot/eval/evaluate_hint_ratio.py
pointer only (licence: NONE) · get_code("d255347d3ae764c7")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

tokenize_delimiter_string Ran
ran, honoured the contract we drafted
safety-research/faithful-cot/train/verl/utils/cot_masking.py
pointer only (licence: NONE) · get_code("a4fd795de45d8e2f")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

tokenize_delimiter_string Ran
ran, honoured the contract we drafted
safety-research/faithful-cot/train/verl/utils/cot_masking_origin.py
pointer only (licence: NONE) · get_code("95c24487f5951cf3")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

create_cot_attention_mask Not yet run
unverified (harvested, no recorded run)
safety-research/faithful-cot/train/verl/utils/cot_masking.py
pointer only (licence: NONE) · get_code("2475baaac745d062")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

load_vllm Not yet run
unverified (harvested, no recorded run)
safety-research/faithful-cot/eval/evaluate_hacking_ratio.py
pointer only (licence: NONE) · get_code("2193e317bc6b5679")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, safety-research/faithful-cot when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Also present in other papers' code

Of 18 samples lifted from this paper's linked repositories, 18 carry a function identity and 0 do not. Of the 18 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 18 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

Chain-of-thought (CoT) reasoning is useful for monitoring language models only when the reasoning trace faithfully reflects the computation that produces the final answer. However, models can rely on prompt-to-answer shortcuts that bypass the CoT, making the visible reasoning trace misleading even when it appears plausible. We study CoT faithfulness through a structural information-flow perspective: faithful reasoning should route answer-relevant information through the mediated path from prompt to CoT to answer, rather than through a direct prompt-to-answer shortcut. This perspective yields a task-agnostic framework based on three complementary properties, sufficiency, completeness, and necessity, which we instantiate with entropy-based, masked-KL, and gradient-based diagnostics. We show that these metrics recover externally judged faithfulness differences in hinted reasoning, and identify a low-entropy failure mode of KLbased diagnostics where gradient-based measures remain more stable. Building on this analysis, we introduce update-time interventions for verifier-based onpolicy RL, including attention masking, backward-only gradient masking, CoT gradients, and adversarial perturbations of prompt representations. Across hinted arithmetic, reward-hackable code repair, and DAPO-Math models trained without hints but evaluated under wrong-hint injection, our interventions shift behavioral and structural indicators toward stronger CoT mediation. In particular, they make shortcut and reward-hacking behavior more transparent in the CoT and improve task-agnostic faithfulness metrics, while in some settings also reducing wrong-hint susceptibility. Our results suggest that controlling information flow during training is a practical route toward more faithful and monitorable CoT reasoning. Code is available at https://github.com/safety-research/faithful-cot.

For agents

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

get_harvested_code_for_paper("2605.24286")
get_code_for_paper("2605.24286")
have("2605.24286")

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

curl https://syntology.ai/api/ran/2605.24286.json

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

[![Syntology run record](https://syntology.ai/api/ran/2605.24286.svg)](https://syntology.ai/paper/2605.24286)

Connect an agent — have() is free.