{"id":"6b17812b-13ad-44b4-b902-22d82e3be23d","arxiv_id":"2610.29734","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"An unsupervised mutual-information post-training objective restores and improves chain-of-thought monitorability across three degraded model organisms and three safety benchmarks, supporting the thesis that monitorability is trainable.","lead":"The paper trains reasoning models to make their chain-of-thought more informative about their final answers, using an unsupervised objective that maximizes mutual information between the two. It reports restoring monitorability lost through compression, monitor evasion, and out-of-context generalization, plus better prediction of unsafe behavior on three safety benchmarks.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"The compression headline rests on an unvalidated LLM-judge metric that tracks CoT length across checkpoints; the paper's own tables imply length alone could account for roughly a fifth to nearly all of the reported 56.3% recovery, and no length-matched or functional-monitor check is reported.","rationale":"This paper argues that monitorability is a trainable attribute: an unsupervised post-training objective maximizing lower-tail CoT-output mutual information under a conditional chi-squared budget recovers monitorability lost to compression, monitor pressure, and out-of-context generalization, and improves behavioral prediction by frozen monitors and an activation probe on undegraded policies. I read the full text including appendices. What must be true for the central claim is that the quantities measured as 'monitorability' actually track a monitor's ability to predict behavior, and that observed gains are attributable to the training objective rather than to incidental correlates of training such as longer CoTs, changed behavior, or class-balance shifts. I went through the three recovery experiments, the undegraded safety experiments, and the consequential-reasoning analysis. The most secure evidence is the safety section (Table 3): frozen LLM monitors and a linear probe improve balanced accuracy on three benchmarks; SORRY-Bench improves with nearly unchanged unsafe-behavior rates (21.9% to 22.8%), and Figure 6 shows gains at fixed prefix positions, which controls for CoT-length differences. The SDF recovery (Table 2, recall 39.5% to 93.1%, with the authors' restricted-scenario robustness check) and the MBPP recovery (Table 1, recall 1.4% to 37.4% at FPR 3.9%) both use frozen monitors on held-out data, and the crossed-CoT experiment (Appendix B.4) provides direct mechanistic evidence. I credit the crossed-CoT design, the honest mode-availability limitation in Appendix F and in b213, and the disclosure of post-hoc selections (Obfuscating-base checkpoint, epsilon=0.4). The load-bearing soft spot is the compression experiment, which generates the abstract's 'recovers most or all of the lost monitorability' and the headline quantitative claims. Its metric is a proprietary LLM judge score, never validated against functional monitorability on this corpus, and the paper's own Tables 15-18 show the judge rising near-monotonically with median CoT length across checkpoints. A conservative length-slope estimate from adjacent checkpoint pairs implies length alone could account for roughly 20-55 of the 56.3 percentage points of reported GPQA recovery; the epsilon=1 vs epsilon=0.5 AIME pair shows the judge is not purely length-driven, so the unresolved question is magnitude. The authors length-normalize the controllability metric (Section A.5) but not the headline metric, which makes the omission conspicuous rather than incidental. My concern does not imply the method does nothing: the functional-monitor results in Sections 4.1.2, 4.1.3, and 4.2.1 stand apart from the judge. It does mean the flagship quantitative claim is currently supported by an unvalidated, length-correlated proxy.","tokens_in":35680,"tokens_out":24482,"duration_ms":162945,"concrete_test":"Run one additional evaluation on the existing GPQA-Diamond compression corpus (Tables 15-16, all six checkpoints, four CoTs per question). Add a functional monitor: a fixed frozen LLM (not fine-tuned on this corpus) that receives only the CoT and must predict that question's actual final output, scored against a shuffled-answer baseline to remove question priors. Then recompute the Table 15 recovery percentages within length-matched strata: bin every sampled CoT from base, comp-100/200, and the four post-trained checkpoints into shared length percentiles, and compare judge scores and frozen-monitor accuracy within each bin.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative pillar for 'recovers most or all of the lost monitorability' is the compression experiment (Section 4.1.1): 56.3% mean and 60.8% worst-case loss recovery at epsilon=1 while preserving 80.2% of token savings on GPQA-Diamond, with analogous MMLU-Pro and AIME figures (Table 15, Figure 1). Every one of these numbers is computed from GPT-5.6-Luna coverage/legibility judgments (geometric mean, Section A.2). For the numbers to support the central claim, the judge must actually track monitorability - a frozen monitor's ability to predict the output from the CoT - and the gains must not be mainly CoT re-lengthening. Neither condition is checked: no frozen monitor or behavioral predictor is evaluated on any compressed checkpoint. The paper's own Table 16 makes the length channel visible. Across GPQA-Diamond checkpoints, median reasoning length and mean judge score move together: (261 tokens, 3.480), (367, 3.617), (384, 3.657), (486, 3.721), (688, 3.749), (881, 3.794). The end-to-end slope is about 0.00051 score points per token; adjacent-pair slopes reach 0.0009-0.0014. On the recovery scale M_base - M_comp(200) = 0.314, simply restoring 123 tokens (from 261 to 384) would by itself move the judge score 0.06-0.17, i.e., 20-55 of the 56.3 percentage points of reported recovery. The AIME epsilon=1/epsilon=0.5 pair (1,637 vs 1,484 tokens; 3.685 vs 3.698) shows the judge is not purely a length meter, and the comp-100+epsilon=1 checkpoint scores near base level (3.749 vs 3.794) with 193 fewer tokens, so content contributes too. The magnitude of the length contribution is unresolved, and magnitude is exactly what decides whether 'most or all' is supportable. The authors normalize the controllability metric for length (Section A.5) but never length-normalize the headline metric, and they never validate the judge against functional monitor accuracy on this corpus.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"This paper argues that chain-of-thought monitorability — the extent to which a monitor can predict the model's output from its visible reasoning — is a trainable attribute of the policy. The authors propose an unsupervised post-training objective that maximizes a lower-tail (CVaR, α=0.75) estimate of the conditional mutual information I(O;Z|X) between CoT Z and output O, subject to a conditional χ² divergence budget ε, using a νPI-controlled Lagrangian and gradient masking that applies the MI gradient only to CoT tokens. The method is evaluated in three recovery settings: (i) CoT compression on GPT-OSS-120B, where a GPT-5.6-Luna judge scores coverage and legibility, and the method reports 56.3% mean and 60.8% worst-case recovery of the GPQA judge-score loss at ε=1 while preserving 80.2% of token savings; (ii) MBPP monitor-pressure obfuscation, where a frozen CoT monitor's recall on hacks rises from 1.4% to 37.4% at 3.9% FPR; and (iii) the Haskins et al. synthetic-document setting, where monitor recall on deceptive rollouts rises from 39.5% to 93.1%. On undegraded policies, post-training improves safe-class F1 and balanced accuracy for two frozen CoT monitors and a linear activation probe across HarmBench, StrongREJECT, and SORRY-Bench, and increases the prevalence of consequential reasoning. The paper concludes that a dedicated, monitor-free post-training stage is a tractable complement to practices that protect CoT from direct optimization pressure.","tokens_in":36154,"tokens_out":28125,"duration_ms":209158,"significance":"If the results hold, this is a valuable contribution: a label-free, monitor-free training signal for a dedicated monitorability post-training stage, with differentiated mechanisms (revelation vs replacement) and unusually honest reporting. The functional-monitor evidence is the paper's strongest asset: the MBPP and SDF experiments use a frozen monitor and held-out labels; the crossed-CoT analysis (Appendix B.4) cleanly attributes the MBPP gain to replacement onto reference-supported CoT-output modes; and the safety experiment (Section 4.2.1) shows consistent balanced-accuracy gains for two external CoT monitors and a linear probe, all independent of the MI estimator used for training. The authors also state their scope openly — Appendix F ('If the ordinary-solution mode (0,0) is missing, reweighting cannot create it: the replacement arrow has no destination') and Section 6 (controlled organisms; naturally occurring frontier failures not yet tested) — and Appendix H publishes raw data tables for the headline figures.","major_comments":[{"comment":"The compression-recovery headline is computed entirely from GPT-5.6-Luna coverage/legibility judgments and is partly confounded with CoT re-lengthening. On GPQA, the 200-step checkpoint's ε=1 condition raises median reasoning length from 261 to 384 tokens while the mean judge score rises from 3.480 to 3.657; using the observed segment slopes (roughly 0.0005 points/token end-to-end from the compressed checkpoint to base, and 0.0013-0.0018 points/token for the adjacent pairs in this range), the 123-token increase alone could account for roughly 0.06-0.22 points of judge-score movement, i.e., roughly 20-70% of the 0.314-point total loss. This is an upper-bound-style calculation, and I acknowledge the evidence against a pure length story: the AIME ε=1/ε=0.5 pair (Table 17) scores slightly lower at higher length (1,637 tokens/3.685 vs 1,484 tokens/3.698), and Figures 2-4 do plot judge scores against median length, as promised in A.2. What is missing is any functional check: no frozen monitor or behavioral predictor is evaluated on any compressed checkpoint (the safety monitors and probe in Section 4.2.1 are run only on undegraded policies), so the reader cannot tell whether the judge-score gains correspond to improved CoT-to-output predictability by an actual monitor. Because the abstract's central quantitative claim ('recovers most or all of the lost monitorability') leans on this experiment, I ask for a frozen-monitor evaluation on the base/compressed/post checkpoints (e.g., one of the Qwen-based or GPT-4.1-nano monitors from Section 4.2.1, or a simple predictor of the final answer from the CoT) and a length-matched or length-normalized judge comparison.","section":"Section 4.1.1, Appendix A.2; Tables 15-16; Figures 2-4","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The claim calibration needs work. The Abstract's 'recovers most or all of the lost monitorability' is not supported by the paper's own weakest cells: on AIME, the 200-step checkpoint recovers 23.6% of the mean judge-score loss at ε=1 (Table 15), and on MBPP, the frozen monitor's recall on the hacks that remain after training is 37.4% at 3.9% FPR (Table 1) — a useful improvement, but not 'most or all' of what was lost. The Figure 1 statement that post-trained models are 'generally above' the linear savings-recovery line also fails for three of the four AIME points (Table 15: 61.3%/11.5%, 21.1%/54.4%, and 72.5%/23.6%). The MBPP and SDF recoveries are additionally mediated to a large degree by the replacement route (behavior change: hacking falls from 86.3% to 26.6% in Table 1 and denial from 55.6% to 24.5% in Table 2, with the SDF recall of 93.1% computed on the reduced set of rollouts that remain deceptive). The paper is transparent about this mechanism (Appendices B.4-B.5 and C.3), and the mode-availability bound is stated in Appendix F and Section 6, but the abstract and conclusion should carry the same qualification: the method reweights CoT-output modes the reference policy already supports, and the reported recoveries vary from roughly a quarter to most of the lost signal depending on the setting.","section":"Abstract; Sections 4.1.1-4.1.3; Tables 1, 2, 15; Figure 1","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"The text states that post-training increases trace coverage under both SCRUPLES prompts, but the paired bootstrap intervals reported in E.2 at δ=0.10 are [−3,10] percentage points (no cue) and [0,16] (cue), both of which include zero, and the GPQA coverage intervals in Table 4 overlap substantially between conditions; the consequential-reasoning results should be described as directional unless more samples are added.","section":"Section 4.2.2, Appendix E.2","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The step-50 SCRUPLES checkpoint was selected retrospectively after observing verbosity growth at step 100 (E.2); this is disclosed in the appendix but should also be stated in the main text, since it conditions the reported comparison.","section":"Section 4.2.2, Appendix E.2","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The judge-based monitorability score is treated as an interval scale when computing percentage loss recovered, and no judge variance or sampling temperature is reported; the standard errors in Figures 2-4 cover generation seeds and training runs but not judge stochasticity, so a sentence stating these limitations would help readers interpret the recovery percentages.","section":"Section 4.1.1, Appendix A.2","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The conclusion of 'no meaningful improvement' in CoT controllability depends on a length normalization using a mean power-law exponent (α=−0.368) taken from the CoT-Control paper; since the length-normalized compliance for the ε=1 checkpoint is 6.99% vs 6.07% for the compressed checkpoint, a brief sensitivity analysis over the exponent would make the claim robust.","section":"Appendix A.5","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The achieved conditional χ² divergence under each budget is not reported, so the reader cannot tell how binding ε is in each experiment; reporting the measured divergence (or the νPI β trajectory) alongside the budget would clarify the trade-off.","section":"Section 3.2, Appendix B.1","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"major_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"Two things to know. The core proposal is real: a label-free post-training objective that maximizes CoT–output mutual information with a CVaR lower tail and a conditional χ² budget, applied as a standalone stage and shown to work across three different degradation settings (compression, monitor-pressure obfuscation, out-of-context generalization). The number you will see quoted—56.3% recovery of lost monitorability on GPQA—comes from an LLM judge the paper never validates against a functional monitor, and the paper's own tables show judge score and CoT length rising together.\n\nThe best work here is mechanistic. The revelation/replacement distinction (Section 3.3, Appendix F) actually explains when recovery is possible, and the crossed-CoT test (Appendix B.4) is the right experiment: take a recovered CoT, keep it fixed, and decode with the obfuscating base's own policy, which succeeds 47.6% of the time. That pins recovery to reweighting onto modes the reference already supports, and honestly exposes the boundary: no surviving transparent modes, no recovery. They state this in Appendix F. The abstract's 'recovers most or all' does not, and AIME recovers only 23.6%.\n\nSoft spots, in size order. First, the compression headline. Judge score and median tokens move together across every GPQA checkpoint; within one compression ancestry, 261→384 tokens carries 0.177 of the 0.314 recovery scale, and adjacent-pair slopes suggest length could explain a large share of the gain. One AIME pair runs the other way (longer, slightly lower score), so the judge is not a pure length meter—but no frozen monitor is run on compressed checkpoints, and the headline metric is never length-normalized while the controllability metric is. The magnitude is genuinely unresolved. Second, MBPP and SDF gains are entangled with behavior change: hacking falls, denial falls, and recall is conditional on the remaining deceptive rollouts. Honest, but 'monitorability recovery' includes real behavior shifts. Third, the safety evaluation is single-split, and two choices (MBPP ε=0.4, SCRUPLES step 50) are post hoc—both disclosed. The safety results are the least vulnerable to the length critique, since gains appear at fixed early prefix positions (Figure 6). Fourth, the estimator comes from the authors' own preprint; the standalone objective, the CVaR targeting, and the χ² constraint are new but incremental.\n\nThis paper deserves a serious referee. Ask for a length-matched functional-monitor check on the compression experiment, a multi-split safety evaluation, and an abstract that matches Appendix F's scope.","headline":"The method's core is sound and the mechanism work is unusually honest, but the compression headline rests on an unvalidated LLM judge with a visible length confound, and the abstract oversells 'most or all'.","tokens_in":36826,"tokens_out":6146,"duration_ms":41633},"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"Chain-of-thought monitorability is trainable: an unsupervised objective recovers most monitoring signal lost to compression, monitor evasion, and out-of-context generalization.","keywords":["chain-of-thought monitoring","monitorability","mutual information","post-training","CVaR","conditional divergence","representation transfer","AI safety"],"falsifier":"A decisive test is to run the same post-training on a policy whose chain of thought is verified to be steganographically or token-level obfuscated, so that no transparent CoT–output mode is even sampled, and check whether monitorability recovers: Appendix F's toy model predicts it should not, because the replacement arrow has no destination. A cheaper check targets the compression headline: evaluate the compressed and post-trained policies with a real downstream monitor on length-matched CoTs, and see whether the reported recovery survives once longer reasoning can no longer inflate the judge-based coverage score.","tokens_in":39437,"feed_emoji":"🧠","tokens_out":3842,"duration_ms":110974,"temperature":0.7,"pith_summary":"Chain-of-thought monitoring—using a model's visible reasoning to predict what it will actually do—is one of the few practical oversight tools for capable AI systems, yet training pressures routinely break it: compression, optimization against a monitor, and out-of-context generalization can all decouple reasoning from behavior. This paper argues that monitorability is a trainable attribute of the policy rather than an incidental byproduct, and proposes an unsupervised post-training objective that maximizes the conditional mutual information between the chain of thought and the output while penalizing any change in what a given chain of thought leads the model to do. Across three degraded model organisms the method recovers most or all of the lost monitoring signal without behavioral labels or monitor feedback, and applied to undegraded policies it improves unsafe-behavior prediction by CoT monitors and activation probes while making reasoning more consequential for the final answer. If the paper is right, oversight of capable models gains a concrete, monitor-agnostic lever: a post-training stage that re-couples what a model says to what it does, rather than hoping the coupling survives training by accident.","feed_headline":"Post-training restores lost chain-of-thought monitorability","feed_subtitle":"Maximizing information between reasoning and output revives oversight signal across three failure modes.","key_machinery":"The load-bearing object is the per-trajectory mutual-information reward $r^{MI}_i$, estimated inside a group of $n$ CoT–output samples for one prompt by scoring each output under every CoT sampled for that prompt and comparing its log-probability with the group average. Averaging this reward over the least-informative $\\alpha=0.75$ fraction of trajectories (conditional value-at-risk) makes the objective concentrate on the worst CoTs rather than seeking the mean. The constraint $D^{cond}_{\\chi^2}(\\pi_\\theta\\|\\pi_{ref}) \\le \\epsilon$—enforced through a normalized Lagrangian with a νPI-controlled multiplier and gradient masking that sends the MI gradient only to CoT tokens and the divergence gradient only to output tokens—limits how much a given CoT's output distribution may change, making it costly for the model to invent private codes while leaving the distribution over CoTs free. A toy model in Appendix F makes the mechanism precise: monitorability improves either by revelation (exposing the same strategy in the CoT) or by replacement (shifting probability toward CoT–output modes the reference policy already supports), and the crossed-CoT experiment shows replacement dominates on MBPP, since recovered CoTs already elicit 47.6% true pass when decoded by the frozen obfuscating policy.","core_discovery":"The paper's central claim is that a model's chain of thought is a learned intermediate representation shaped by the pressures of producing high-reward outputs, and that those same pressures can strip it of the information a monitor needs—so monitorability should be optimized directly rather than assumed. The proposed objective maximizes a lower-tail (CVaR, $\\alpha=0.75$) estimate of the conditional mutual information $I(O;Z|X)$ between the chain of thought and the output, subject to a conditional $\\chi^2$ divergence budget $\\epsilon$ that constrains how much a fixed chain of thought may change the output it produces. The headline results are specific: from a 200-step compressed checkpoint on GPQA-Diamond, post-training at $\\epsilon=1$ recovers 56.3% of the mean and 60.8% of the worst-case monitorability loss while preserving 80.2% of the token savings; on MBPP it raises hack-detection recall from 1.4% to 37.4% at a false-positive rate of 3.9% while raising true pass from 11.2% to 48.6%; and on the synthetic-document customer-support organism it raises deceptive-rollout recall from 39.5% to 93.1%. On undegraded policies, both CoT monitors and an activation probe improve balanced accuracy on HarmBench, StrongREJECT, and SORRY-Bench, and trace coverage increases on both SCRUPLES and GPQA. The authors frame the mechanism explicitly: gains flow through revelation (the same behavior becomes more explicit in the CoT) or replacement (probability shifts toward CoT–output modes the reference policy already supports), which is why the method can only recover signal whose mode still exists.","pith_inferences":["Because recovery operates by reweighting modes the reference policy already supports, the method doubles as a diagnostic: the amount of monitorability recoverable under a fixed budget estimates how much transparent reasoning capacity survives in a policy, which is precisely the quantity an oversight regime needs before trusting CoT monitoring.","A boundary case follows from the paper's own Appendix F logic: policies under intense pressure that never rewarded or sampled transparent CoT–output couplings, or that learned true steganographic codes, should resist this objective; a natural extension is to test whether the $\\chi^2$ budget must explode before any monitorability reappears in such a policy.","The compression headline rests on a judge-based score that correlates with length; the paper reports length alongside score and normalizes controllability for length, but a sharper future evaluation would hold CoT length fixed and measure recovery with an actual downstream monitor rather than a judge.","The revelation/replacement distinction suggests a possible coupling between monitorability training and honest behavior: in both the MBPP and synthetic-document settings, replacement happened to route toward correct and honest modes the reference policy supported, and whether that prosocial bias holds generally is an open question this paper does not address."],"forward_implications":["A dedicated monitorability post-training stage can be added to the training pipeline after the fact, extending current practices that protect chain-of-thought from direct optimization pressure by repairing damage after it has occurred.","Compressed reasoning models do not have to choose between token savings and oversight: at $\\epsilon=1$ the method keeps 80.2% of compression savings on GPQA while recovering most of the lost monitorability, and the gains transfer out-of-distribution to MMLU-Pro and AIME.","Monitor evasion and out-of-context concealment are partly reversible without behavioral labels: residual hacks and deceptive rollouts become far more detectable after post-training, not merely less frequent.","On undegraded policies the same objective improves safe/unsafe prediction by two CoT monitors and an activation probe across all three safety benchmarks, so the method strengthens monitorability rather than only repairing it.","Post-training increases trace coverage—the fraction of chains of thought containing at least one span that changes the probability of the final answer—showing the re-coupling is functional and not just a judged improvement in style."],"supporting_citations":[{"why":"Supplies the conditional mutual-information estimator between CoT and output, and the information-theoretic framing of monitorability, that the post-training objective is built on.","marker":"Anwar et al. (2026)"},{"why":"Establishes the transfer view of CoT monitoring and shows that optimization pressure can promote obfuscation, motivating the recovery setting.","marker":"Baker et al. (2025)"},{"why":"Provides the synthetic-document fine-tuning environment whose obfuscating customer-support checkpoint is the third degraded model organism.","marker":"Haskins et al. (2026)"},{"why":"Supplies the correctness-gated soft length penalty used to create the compressed-CoT model organism.","marker":"Li et al. (2025)"},{"why":"Defines the LLM-judge coverage and legibility scoring that the compression experiment uses as its monitorability metric.","marker":"Emmons et al. (2025)"},{"why":"Supplies the prefix-based protocol for predicting whether CoT continuations will be safe or unsafe, used in the safety-benchmark evaluation.","marker":"Chan et al. (2025)"},{"why":"Provides the consequential-span evaluation that measures whether reasoning actually changes the probability of the final answer.","marker":"Du et al. (2026)"},{"why":"Defines conditional value-at-risk, the lower-tail aggregation that focuses optimization on the least-informative CoTs.","marker":"Rockafellar & Uryasev, 2000"},{"why":"Motivates the conditional $\\chi^2$ divergence (rather than KL) as the regularizer for behavioral conservatism.","marker":"Huang et al. (2025)"},{"why":"Contributes the GRPO per-prompt rollout groups that make the mutual-information estimator nearly sample-free.","marker":"Shao et al., 2024"}],"fun_headline_variants":["Monitorability is trainable via post-training objective","CoT monitorability can be restored by training","Mutual-information post-training revives CoT oversight","Don't assume monitorability: train it post hoc","Post-training objective recovers lost CoT monitorability"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The load-bearing premise is that the objective can only reweight CoT–output modes that already exist in the reference policy—if no transparent reasoning-to-output coupling survives for the training to shift probability toward, recovery cannot occur—and, for the compression headline specifically, that the LLM-judge coverage and legibility score is a length-unbiased proxy for what a real monitor can extract.","fun_headline_variants_meta":{"raw":{"variants":["Monitorability is trainable via post-training objective","CoT monitorability can be restored by training","Mutual-information post-training revives CoT oversight","Don't assume monitorability: train it post hoc","Post-training objective recovers lost CoT monitorability"]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":9.8e-05,"raw_usage":{"cost":9.81708e-05,"is_byok":false,"cost_details":{"upstream_inference_cost":9.81708e-05,"upstream_inference_prompt_cost":3.62908e-05,"upstream_inference_completions_cost":6.188e-05},"total_tokens":1498,"prompt_tokens":1158,"completion_tokens":340,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":949,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":278}},"tokens_in":1158,"tokens_out":340,"cost_source":"openrouter_usage","duration_ms":5227,"image_pages":[],"temperature":1.0,"billed_attempts":1,"reasoning_tokens":278,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-10-22T07:54:55.605806+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":"A decisive test is to run the same post-training on a policy whose chain of thought is verified to be steganographically or token-level obfuscated, so that no transparent CoT–output mode is even sampled, and check whether monitorability recovers: Appendix F's toy model predicts it should not, because the replacement arrow has no destination. A cheaper check targets the compression headline: evaluate the compressed and post-trained policies with a real downstream monitor on length-matched CoTs, and see whether the reported recovery survives once longer reasoning can no longer inflate the judge-based coverage score.","supporting_citations":[{"cited_title":"2026 , month = mar, howpublished =","cited_arxiv_id":null,"evidence_quote":"Provides the synthetic-document fine-tuning environment whose obfuscating customer-support checkpoint is the third degraded model organism."}],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}