--- schema_version: "1.0.0" edition_id: "mp-2026-10-04-morning-0087" published_at: "2026-10-04T09:00:00.000-04:00" modified_at: "2026-10-05T08:46:40.372-04:00" canonical_url: "https://themachinepress.com/edition/2026-10-04" story_count: 27 lead_story_id: "mp-2026-10-04-001" --- # The Machine Press — Morning edition Edition ID: `mp-2026-10-04-morning-0087` Published: 2026-10-04T09:00:00.000-04:00 Canonical edition: https://themachinepress.com/edition/2026-10-04 UniIntervene++ reported 89.67 percent success on five physical manipulation tasks while adapting how much help a learning policy received. ## 1. The Robot’s Helper Learned When to Let Go {#mp-2026-10-04-001} - Story ID: `mp-2026-10-04-001` - Type: `lead` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-001/the-robot-s-helper-learned-when-to-let-go **Dek:** UniIntervene++ reported 89.67 percent success on five physical manipulation tasks while adapting how much help a learning policy received. Online robot learning changes the need for assistance as a policy improves. UniIntervene++ treats autonomous control, trajectory correction and a structured CodePolicy as alternative behaviors in one decision process, learns their relative values online and periodically tests the robot without help. Assisted experience trains the main policy; its changing performance then reshapes the next intervention. Across five real-world manipulation tasks, the authors report 89.67 percent average success, at least six percentage points above tested baselines, with their human-intervention metric reduced to 0.77 percent. The work makes the handoff itself part of learning instead of relying on a fixed trigger. These are preprint results for the reported tasks, not a claim that arbitrary robots can operate without supervision. ### Why it matters {#why-it-matters-mp-2026-10-04-001} UniIntervene++ reported 89.67 percent success on five physical manipulation tasks while adapting how much help a learning policy received. ### Limits and context {#limitations-mp-2026-10-04-001} - These are preprint results for the reported tasks, not a claim that arbitrary robots can operate without supervision. ### Claims and sources {#claims-mp-2026-10-04-001} - UniIntervene++ reported 89.67 percent success on five physical manipulation tasks while adapting how much help a learning policy received. [source-2026-10-04-001] — Qualification: These are preprint results for the reported tasks, not a claim that arbitrary robots can operate without supervision. ## 2. The Detector Won the Test. Then It Met the Tool Output. {#mp-2026-10-04-002} - Story ID: `mp-2026-10-04-002` - Type: `secondary` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-002/the-detector-won-the-test-then-it-met-the-tool-output **Dek:** A prompt-injection audit found that benchmark rankings transferred poorly when detectors screened the longer inputs an agent actually receives. The study reconstructed benign tool outputs by replaying ground-truth calls from AgentDojo and tau-bench, then labeled injected versions through differential replay. It evaluated fifteen detectors and two task-aware model judges alongside BIPIA. At a one percent false-positive rate, the best BIPIA detector caught only two percent of AgentDojo injections; another detector caught 72 percent on AgentDojo but 15 percent on tau-bench. Public training data suggested that input form mattered: exposure to isolated attack strings did not establish detection inside a tool response. The authors recommend testing on an agent’s own outputs, reporting detection at low false-positive rates and auditing training overlap. This preprint tests particular detectors and benchmarks; it does not establish that every deployed screening system fails. ### Why it matters {#why-it-matters-mp-2026-10-04-002} A prompt-injection audit found that benchmark rankings transferred poorly when detectors screened the longer inputs an agent actually receives. ### Limits and context {#limitations-mp-2026-10-04-002} - At a one percent false-positive rate, the best BIPIA detector caught only two percent of AgentDojo injections; another detector caught 72 percent on AgentDojo but 15 percent on tau-bench. - Public training data suggested that input form mattered: exposure to isolated attack strings did not establish detection inside a tool response. - This preprint tests particular detectors and benchmarks; it does not establish that every deployed screening system fails. ### Claims and sources {#claims-mp-2026-10-04-002} - A prompt-injection audit found that benchmark rankings transferred poorly when detectors screened the longer inputs an agent actually receives. [source-2026-10-04-002] — Qualification: At a one percent false-positive rate, the best BIPIA detector caught only two percent of AgentDojo injections; another detector caught 72 percent on AgentDojo but 15 percent on tau-bench. ## 3. The Mobile Robot Recycled Its Own Experience {#mp-2026-10-04-003} - Story ID: `mp-2026-10-04-003` - Type: `dispatch` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-003/the-mobile-robot-recycled-its-own-experience **Dek:** MobiAgent separates execution from a learning loop that discovers and refines reusable skills. MobiAgent uses high-level planning and visual reflection to compose atomic skills during mobile manipulation, while a second loop segments and verifies deployment rollouts for further training. Specialized control experts share a visual-language backbone. The authors report a 22.5-percentage-point gain over π0.5-TA on BEHAVIOR-1K, with recycled experience improving RoboCasa success from 7.5 to 27.5 percent and Astribot S1 success from 32.5 to 57.5 percent. The reported recovery and learning gains are bounded by those evaluations. ### Why it matters {#why-it-matters-mp-2026-10-04-003} MobiAgent separates execution from a learning loop that discovers and refines reusable skills. ### Limits and context {#limitations-mp-2026-10-04-003} - No additional limitation was separately recorded. ### Claims and sources {#claims-mp-2026-10-04-003} - MobiAgent separates execution from a learning loop that discovers and refines reusable skills. [source-2026-10-04-003] ## 4. The Teacher Repaired the Student’s First Wrong Turn {#mp-2026-10-04-004} - Story ID: `mp-2026-10-04-004` - Type: `dispatch` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-004/the-teacher-repaired-the-student-s-first-wrong-turn **Dek:** RC-OPD uses local repairs of a model’s own failed reasoning instead of handing it an unrelated correct solution. Reference-based distillation can teach a correct conclusion without fixing the mistake that produced a student’s failure. RC-OPD locates the earliest substantive error, builds a local correction and tests whether the student can continue from the repaired point. Successful repair chains supervise flawed segments while preserving valid progress. The preprint reports improvements across multiple datasets and model scales, but its abstract does not supply a single aggregate gain; the contribution is the tested diagnosis-and-repair procedure. ### Why it matters {#why-it-matters-mp-2026-10-04-004} RC-OPD uses local repairs of a model’s own failed reasoning instead of handing it an unrelated correct solution. ### Limits and context {#limitations-mp-2026-10-04-004} - The preprint reports improvements across multiple datasets and model scales, but its abstract does not supply a single aggregate gain; the contribution is the tested diagnosis-and-repair procedure. ### Claims and sources {#claims-mp-2026-10-04-004} - RC-OPD uses local repairs of a model’s own failed reasoning instead of handing it an unrelated correct solution. [source-2026-10-04-004] — Qualification: The preprint reports improvements across multiple datasets and model scales, but its abstract does not supply a single aggregate gain; the contribution is the tested diagnosis-and-repair procedure. ## 5. The Skill Package’s Biggest Advantage Was Its Calculator {#mp-2026-10-04-005} - Story ID: `mp-2026-10-04-005` - Type: `dispatch` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-005/the-skill-package-s-biggest-advantage-was-its-calculator **Dek:** FinSkillBench separates procedural documents from executable tools in verifiable financial-agent tasks. FinSkillBench contains 2,603 dated episodes across twelve subtasks with deterministic verifiers. A paired analysis across eight models found curated skill packages raised mean scores by 16.2 points, while skills generated within one episode added only half a point and used more tokens and turns. Testing resources separately, documents added 5.6 points and executable tools 19.5; combining them was subadditive. A second harness reproduced the direction while changing effect sizes. The authors frame the premium as a property of the full model, resources and harness, rather than proof that the underlying model acquired financial expertise. ### Why it matters {#why-it-matters-mp-2026-10-04-005} FinSkillBench separates procedural documents from executable tools in verifiable financial-agent tasks. ### Limits and context {#limitations-mp-2026-10-04-005} - A paired analysis across eight models found curated skill packages raised mean scores by 16.2 points, while skills generated within one episode added only half a point and used more tokens and turns. ### Claims and sources {#claims-mp-2026-10-04-005} - FinSkillBench separates procedural documents from executable tools in verifiable financial-agent tasks. [source-2026-10-04-005] — Qualification: A paired analysis across eight models found curated skill packages raised mean scores by 16.2 points, while skills generated within one episode added only half a point and used more tokens and turns. ## 6. Sixteen-Bit Arithmetic Challenged the Eight-Bit Budget {#mp-2026-10-04-006} - Story ID: `mp-2026-10-04-006` - Type: `dispatch` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-006/sixteen-bit-arithmetic-challenged-the-eight-bit-budget **Dek:** W16A16 reports lower quantization error with comparable or better inference time and energy on Armv7E-M microcontrollers. Lower precision does not automatically deliver the best implementation on every processor. The W16A16 study analyzes Armv7E-M instructions and evaluates sixteen-bit neural-network quantization for regression and classification. It reports roughly ten times lower quantization error than the tested eight-bit schemes while matching or improving measured inference time and energy. The finding is specific to its microcontroller architecture, models and kernels, rather than a universal argument for higher-precision edge inference. ### Why it matters {#why-it-matters-mp-2026-10-04-006} W16A16 reports lower quantization error with comparable or better inference time and energy on Armv7E-M microcontrollers. ### Limits and context {#limitations-mp-2026-10-04-006} - Lower precision does not automatically deliver the best implementation on every processor. ### Claims and sources {#claims-mp-2026-10-04-006} - W16A16 reports lower quantization error with comparable or better inference time and energy on Armv7E-M microcontrollers. [source-2026-10-04-006] — Qualification: Lower precision does not automatically deliver the best implementation on every processor. ## 7. The Next Shot Borrowed a Character and a Different Room {#mp-2026-10-04-007} - Story ID: `mp-2026-10-04-007` - Type: `dispatch` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-007/the-next-shot-borrowed-a-character-and-a-different-room **Dek:** Weave Forcing routes historical video memories separately for subjects and backgrounds without retraining the generator. Interactive video needs to recombine earlier elements, not simply extend the last scene. Weave Forcing divides a prompt into semantic slots, retrieves references for characters and backgrounds separately, and masks compressed historical memories so unrelated content stays hidden. It also adjusts temporal offsets when reference coverage is incomplete. The authors report better cross-shot subject and background consistency with competitive visual quality and text alignment; the abstract gives no universal duration or consistency guarantee. ### Why it matters {#why-it-matters-mp-2026-10-04-007} Weave Forcing routes historical video memories separately for subjects and backgrounds without retraining the generator. ### Limits and context {#limitations-mp-2026-10-04-007} - Interactive video needs to recombine earlier elements, not simply extend the last scene. ### Claims and sources {#claims-mp-2026-10-04-007} - Weave Forcing routes historical video memories separately for subjects and backgrounds without retraining the generator. [source-2026-10-04-007] — Qualification: Interactive video needs to recombine earlier elements, not simply extend the last scene. ## 8. The Visual Tokens Stayed. Half the Work Disappeared. {#mp-2026-10-04-008} - Story ID: `mp-2026-10-04-008` - Type: `dispatch` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-008/the-visual-tokens-stayed-half-the-work-disappeared **Dek:** Patch-to-Prune bypasses selected visual computations while retaining sequence length, token order and pretrained weights. Patch-to-Prune turns activation patching into an inference procedure. Validation-guided sweeps find decoder regions where fixed proxy activations can replace visual-token projection outputs within an accuracy tolerance. The method keeps token positions, attention masks and residual pathways intact. Across four Qwen2.5-VL and LLaVA models and seven multimodal benchmarks, the authors report about 94 percent of dense-model accuracy with 55 percent fewer FLOPs at a three percent tolerance setting. FLOP savings are an evaluated computation measure, not a promised wall-clock speedup on every device. ### Why it matters {#why-it-matters-mp-2026-10-04-008} Patch-to-Prune bypasses selected visual computations while retaining sequence length, token order and pretrained weights. ### Limits and context {#limitations-mp-2026-10-04-008} - FLOP savings are an evaluated computation measure, not a promised wall-clock speedup on every device. ### Claims and sources {#claims-mp-2026-10-04-008} - Patch-to-Prune bypasses selected visual computations while retaining sequence length, token order and pretrained weights. [source-2026-10-04-008] — Qualification: FLOP savings are an evaluated computation measure, not a promised wall-clock speedup on every device. ## 9. The Robot’s Route Had to Read the Work Schedule {#mp-2026-10-04-009} - Story ID: `mp-2026-10-04-009` - Type: `dispatch` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-009/the-robot-s-route-had-to-read-the-work-schedule **Dek:** CORNAV combines architectural drawings, scene graphs and changing exclusion zones for navigation on active construction sites. CORNAV aligns two-dimensional CAD plans with an open-vocabulary scene graph, translates project schedules into time-varying constraints and screens hazardous requests before planning. An A* planner enforces mandatory exclusion zones. In an office and a real construction site, the authors report task success rising from 13.0 to 72.2 percent over semantic retrieval alone, with schedule awareness eliminating hard-zone violations in those tests. The result supports construction-specific context in navigation; it is not general safety certification for changing worksites. ### Why it matters {#why-it-matters-mp-2026-10-04-009} CORNAV combines architectural drawings, scene graphs and changing exclusion zones for navigation on active construction sites. ### Limits and context {#limitations-mp-2026-10-04-009} - The result supports construction-specific context in navigation; it is not general safety certification for changing worksites. ### Claims and sources {#claims-mp-2026-10-04-009} - CORNAV combines architectural drawings, scene graphs and changing exclusion zones for navigation on active construction sites. [source-2026-10-04-009] — Qualification: The result supports construction-specific context in navigation; it is not general safety certification for changing worksites. ## 10. The Model Learned Which Measurement Was Worth Buying {#mp-2026-10-04-010} - Story ID: `mp-2026-10-04-010` - Type: `dispatch` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-010/the-model-learned-which-measurement-was-worth-buying **Dek:** ECHO-k chooses additional modalities from internal representations when the eventual prediction task is unknown. Collecting every available modality can be expensive or slow. ECHO-k uses a pretrained model’s internal representations as proxy targets for cross-modal information, then learns a policy that acquires measurements sequentially. The authors provide guarantees in a stylized linear setting and report improved budgeted downstream performance across several foundation-model backends against task-agnostic, label-free baselines. The theory’s restricted setting and the evaluated budgets matter; the work does not establish that omitted measurements are always dispensable. ### Why it matters {#why-it-matters-mp-2026-10-04-010} ECHO-k chooses additional modalities from internal representations when the eventual prediction task is unknown. ### Limits and context {#limitations-mp-2026-10-04-010} - The theory’s restricted setting and the evaluated budgets matter; the work does not establish that omitted measurements are always dispensable. ### Claims and sources {#claims-mp-2026-10-04-010} - ECHO-k chooses additional modalities from internal representations when the eventual prediction task is unknown. [source-2026-10-04-010] — Qualification: The theory’s restricted setting and the evaluated budgets matter; the work does not establish that omitted measurements are always dispensable. ## 11. One Photograph Became Collision Geometry in Half a Second {#mp-2026-10-04-011} - Story ID: `mp-2026-10-04-011` - Type: `dispatch` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-011/one-photograph-became-collision-geometry-in-half-a-second **Dek:** I2CD predicts convex parts directly, skipping the mesh-repair pipeline before physics simulation and motion planning. I2CD freezes a pretrained image-to-3D backbone and trains a lightweight head whose tokens emit convex polytopes. On 227 held-out objects, the authors report the highest volumetric overlap among eight compared pipelines and six to 37 times faster end-to-end processing. Four physics engines accepted its delivered geometry without post-processing. In a physical xArm7 experiment, it prepared a twenty-object scene in eleven seconds versus 328 for the strongest baseline, with 85 versus 90 successful picks in one hundred trials. Faster geometry preparation did not improve that measured execution success. ### Why it matters {#why-it-matters-mp-2026-10-04-011} I2CD predicts convex parts directly, skipping the mesh-repair pipeline before physics simulation and motion planning. ### Limits and context {#limitations-mp-2026-10-04-011} - Faster geometry preparation did not improve that measured execution success. ### Claims and sources {#claims-mp-2026-10-04-011} - I2CD predicts convex parts directly, skipping the mesh-repair pipeline before physics simulation and motion planning. [source-2026-10-04-011] — Qualification: Faster geometry preparation did not improve that measured execution success. ## 12. The Speech Generator Learned Without a Generative Teacher {#mp-2026-10-04-012} - Story ID: `mp-2026-10-04-012` - Type: `dispatch` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-012/the-speech-generator-learned-without-a-generative-teacher **Dek:** DriftTTS trains a few-step mel-spectrogram generator through distribution matching and its own intermediate rollouts. DriftTTS avoids a pretrained generative teacher, distillation and adversarial discrimination. Its objective operates in a mel-domain feature space that includes a frozen masked-autoencoder trained on the same LJSpeech split. With four function evaluations, the authors report a 3.7 percent word-error rate versus Matcha-TTS’s 3.4 percent. A paired blind listening test gave mean scores of 4.18 and 3.96 respectively, with ground truth at 4.22. Those dataset-specific results support competitive synthesis under this setup, not universal voice quality or robustness. ### Why it matters {#why-it-matters-mp-2026-10-04-012} DriftTTS trains a few-step mel-spectrogram generator through distribution matching and its own intermediate rollouts. ### Limits and context {#limitations-mp-2026-10-04-012} - Those dataset-specific results support competitive synthesis under this setup, not universal voice quality or robustness. ### Claims and sources {#claims-mp-2026-10-04-012} - DriftTTS trains a few-step mel-spectrogram generator through distribution matching and its own intermediate rollouts. [source-2026-10-04-012] — Qualification: Those dataset-specific results support competitive synthesis under this setup, not universal voice quality or robustness. ## 13. The Action Policy Started Learning Before Actions Were Labeled {#mp-2026-10-04-013} - Story ID: `mp-2026-10-04-013` - Type: `dispatch` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-013/the-action-policy-started-learning-before-actions-were-labeled **Dek:** NAVA-WAM sends future-video supervision directly into a robot policy before training it on commanded demonstrations. NAVA-WAM learns action-relevant priors from observation-only videos instead of first learning a separate latent-action model. Future-video supervision passes through joint attention to train the action branch; later, action-labeled robot demonstrations ground that branch for control. Asymmetric attention then enables action-only inference. The preprint reports stronger in-distribution and out-of-distribution results, label efficiency and real-robot generalization across its experiments, without providing one aggregate figure in the abstract. Observation-only pretraining supplements rather than eliminates the need for labeled robot data. ### Why it matters {#why-it-matters-mp-2026-10-04-013} NAVA-WAM sends future-video supervision directly into a robot policy before training it on commanded demonstrations. ### Limits and context {#limitations-mp-2026-10-04-013} - NAVA-WAM learns action-relevant priors from observation-only videos instead of first learning a separate latent-action model. - Asymmetric attention then enables action-only inference. - Observation-only pretraining supplements rather than eliminates the need for labeled robot data. ### Claims and sources {#claims-mp-2026-10-04-013} - NAVA-WAM sends future-video supervision directly into a robot policy before training it on commanded demonstrations. [source-2026-10-04-013] — Qualification: NAVA-WAM learns action-relevant priors from observation-only videos instead of first learning a separate latent-action model. ## 14. The Humanoid Learned the Landing and the Fall {#mp-2026-10-04-014} - Story ID: `mp-2026-10-04-014` - Type: `dispatch` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-014/the-humanoid-learned-the-landing-and-the-fall **Dek:** KungfuAthleteBot repairs video-derived motion and trains disturbance recovery inside the same control policy. Video motion lacks forces and can contain floating, ground penetration and jitter. KungfuAthleteBot corrects reconstructed aerial trajectories and samples physically plausible training starts, allowing a policy to learn feasible actuation instead of rigidly copying an impossible reference. Disturbance rejection and fall recovery are trained with motion tracking in one policy. The authors report about 0.7-second recovery from falls on their humanoid and ablations supporting each component. This is a preprint result for the reported platform; it does not establish unrestricted athletic safety. ### Why it matters {#why-it-matters-mp-2026-10-04-014} KungfuAthleteBot repairs video-derived motion and trains disturbance recovery inside the same control policy. ### Limits and context {#limitations-mp-2026-10-04-014} - This is a preprint result for the reported platform; it does not establish unrestricted athletic safety. ### Claims and sources {#claims-mp-2026-10-04-014} - KungfuAthleteBot repairs video-derived motion and trains disturbance recovery inside the same control policy. [source-2026-10-04-014] — Qualification: This is a preprint result for the reported platform; it does not establish unrestricted athletic safety. ## 15. The Attack Changed the Representation. The Caption Stayed Right. {#mp-2026-10-04-026} - Story ID: `mp-2026-10-04-026` - Type: `dispatch` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-026/the-attack-changed-the-representation-the-caption-stayed-right **Dek:** A controlled vision-language experiment separates corrupted internal signals from the answer a model freely generates. On Qwen2.5-VL-7B-Instruct and two hundred held-out COCO images, a targeted perturbation could lower training loss for a chosen caption while free generation still produced the correct description. The study traces the disagreement through the decoder and reports that susceptible and resistant outcomes diverge there despite similar visual-encoder corruption. It also flags circularity in one probe estimate. The result concerns a particular model and attack setup, and argues for measuring generated behavior alongside internal proxies. ### Why it matters {#why-it-matters-mp-2026-10-04-026} A controlled vision-language experiment separates corrupted internal signals from the answer a model freely generates. ### Limits and context {#limitations-mp-2026-10-04-026} - No additional limitation was separately recorded. ### Claims and sources {#claims-mp-2026-10-04-026} - A controlled vision-language experiment separates corrupted internal signals from the answer a model freely generates. [source-2026-10-04-015] ## 16. A High Score Hid an Impossible Causal Claim {#mp-2026-10-04-027} - Story ID: `mp-2026-10-04-027` - Type: `dispatch` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-027/a-high-score-hid-an-impossible-causal-claim **Dek:** CERTID tests whether a reasoning model knows when observational data cannot identify the effect it is asked to compute. A model can refuse a solvable causal query or confidently supply a formula for an effect that observational data cannot identify. CERTID certifies identifiability from the graph and query, then checks formulas against structural causal models with known interventional distributions. On 1,200 instances spanning four to fifty vertices, three tested frontier models showed a seventeen-fold difference in false claims on non-identifiable questions. Their overall accuracy did not summarize that risk. The preprint evaluates formal graph problems, not unrestricted real-world causal inference, but makes an unsupported answer distinguishable from an ordinary mistake. ### Why it matters {#why-it-matters-mp-2026-10-04-027} CERTID tests whether a reasoning model knows when observational data cannot identify the effect it is asked to compute. ### Limits and context {#limitations-mp-2026-10-04-027} - A model can refuse a solvable causal query or confidently supply a formula for an effect that observational data cannot identify. - Their overall accuracy did not summarize that risk. - The preprint evaluates formal graph problems, not unrestricted real-world causal inference, but makes an unsupported answer distinguishable from an ordinary mistake. ### Claims and sources {#claims-mp-2026-10-04-027} - CERTID tests whether a reasoning model knows when observational data cannot identify the effect it is asked to compute. [source-2026-10-04-016] — Qualification: A model can refuse a solvable causal query or confidently supply a formula for an effect that observational data cannot identify. ## 17. Local Data Did Not Prove Compliant Training {#mp-2026-10-04-015} - Story ID: `mp-2026-10-04-015` - Type: `ticker` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-015/local-data-did-not-prove-compliant-training **Dek:** PoCoFL separates federation topology, admission policy and cryptographic proof. PoCoFL uses commitments and zero-knowledge proofs to check client contributions and whether aggregation follows a selected policy. Four proof-of-concept instantiations cover vanilla, continual, personalized and threshold-encrypted federated learning. The framework addresses compliance beyond keeping raw training data local. ### Why it matters {#why-it-matters-mp-2026-10-04-015} PoCoFL separates federation topology, admission policy and cryptographic proof. ### Limits and context {#limitations-mp-2026-10-04-015} - No additional limitation was separately recorded. ### Claims and sources {#claims-mp-2026-10-04-015} - PoCoFL separates federation topology, admission policy and cryptographic proof. [source-2026-10-04-017] ## 18. A Short-Length Prediction Bought a Place in Line {#mp-2026-10-04-016} - Story ID: `mp-2026-10-04-016` - Type: `ticker` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-016/a-short-length-prediction-bought-a-place-in-line **Dek:** JIL manipulates an output-length estimate used by a size-based serving scheduler. In tests across two datasets and four models, adversarial suffixes cut predicted lengths by up to 83.4 percent and improved attacker-request completion time by up to 1.53 times on average. Coarse prediction intervals reduced the advantage. These are scheduler-specific experiments. ### Why it matters {#why-it-matters-mp-2026-10-04-016} JIL manipulates an output-length estimate used by a size-based serving scheduler. ### Limits and context {#limitations-mp-2026-10-04-016} - No additional limitation was separately recorded. ### Claims and sources {#claims-mp-2026-10-04-016} - JIL manipulates an output-length estimate used by a size-based serving scheduler. [source-2026-10-04-018] ## 19. The Satellite Filtered the Sea Before Sending the Image {#mp-2026-10-04-017} - Story ID: `mp-2026-10-04-017` - Type: `ticker` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-017/the-satellite-filtered-the-sea-before-sending-the-image **Dek:** A marine anomaly pipeline compresses Earth-observation imagery for resource-limited onboard processing. The study combines a self-supervised image encoder with anomaly detection and compares it with traditional detectors. It reports integration work involving Phisat-2 and IMAGIN-e. Prioritizing unusual sea patterns aims to reduce transmission load; the abstract does not establish a universal detection rate or alert latency. ### Why it matters {#why-it-matters-mp-2026-10-04-017} A marine anomaly pipeline compresses Earth-observation imagery for resource-limited onboard processing. ### Limits and context {#limitations-mp-2026-10-04-017} - Prioritizing unusual sea patterns aims to reduce transmission load; the abstract does not establish a universal detection rate or alert latency. ### Claims and sources {#claims-mp-2026-10-04-017} - A marine anomaly pipeline compresses Earth-observation imagery for resource-limited onboard processing. [source-2026-10-04-019] — Qualification: Prioritizing unusual sea patterns aims to reduce transmission load; the abstract does not establish a universal detection rate or alert latency. ## 20. The Right Answer Was Missing From the Menu {#mp-2026-10-04-018} - Story ID: `mp-2026-10-04-018` - Type: `ticker` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-018/the-right-answer-was-missing-from-the-menu **Dek:** A paired benchmark measures rejection separately from classification accuracy. The candidate-coverage study tests Laya and Jev on matched requests with a reference option present or absent. Native rejection and false-rejection rates differed sharply; calibration changed the tradeoff. The evaluation covers reference-label omission and explicitly leaves natural out-of-scope generalization unresolved. ### Why it matters {#why-it-matters-mp-2026-10-04-018} A paired benchmark measures rejection separately from classification accuracy. ### Limits and context {#limitations-mp-2026-10-04-018} - No additional limitation was separately recorded. ### Claims and sources {#claims-mp-2026-10-04-018} - A paired benchmark measures rejection separately from classification accuracy. [source-2026-10-04-020] ## 21. The Attack-Path Draft Went Back to the Planner {#mp-2026-10-04-019} - Story ID: `mp-2026-10-04-019` - Type: `ticker` - Classification: `editorial` - Content status: `updated` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-019/the-attack-path-draft-went-back-to-the-planner **Dek:** CVE2AP turns vulnerability descriptions into formal planning models and uses execution errors for revision. The study reports up to 86.9 percent syntax correctness and 78.6 percent solvability, with planner feedback yielding consistent improvement. Its 93.1 percent semantic-correctness figure uses a model-as-expert judge, so a solvable generated attack path still needs independent semantic review. ### Why it matters {#why-it-matters-mp-2026-10-04-019} CVE2AP turns vulnerability descriptions into formal planning models and uses execution errors for revision. ### Limits and context {#limitations-mp-2026-10-04-019} - No additional limitation was separately recorded. ### Claims and sources {#claims-mp-2026-10-04-019} - CVE2AP turns vulnerability descriptions into formal planning models and uses execution errors for revision. [source-2026-10-04-021] ## 22. Space Drums 2.0 {#mp-2026-10-04-020} - Story ID: `mp-2026-10-04-020` - Type: `invention_desk` - Classification: `editorial` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-020/space-drums-2-0 **Dek:** Moves motion tracking onto two ESP32-S3 drumsticks, adds haptic hits and a two-pedal hub, and sends velocity-sensitive drum events to open PC or Android playback software. Moves motion tracking onto two ESP32-S3 drumsticks, adds haptic hits and a two-pedal hub, and sends velocity-sensitive drum events to open PC or Android playback software. ### Why it matters {#why-it-matters-mp-2026-10-04-020} An independent builder is turning an improbable idea into a working project. ### Limits and context {#limitations-mp-2026-10-04-020} - A Desk Pick is an editorial selection, not a product endorsement. ### Claims and sources {#claims-mp-2026-10-04-020} - This Invention Desk entry makes no independently sourced news claim. ## 23. 3RIC {#mp-2026-10-04-021} - Story ID: `mp-2026-10-04-021` - Type: `invention_desk` - Classification: `editorial` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-021/3ric **Dek:** Pairs a chip-level 65C02 computer with custom hardware, video, sound, floppy, and microSD support with a cycle-honest emulator and browser assembler built around the same machine. Pairs a chip-level 65C02 computer with custom hardware, video, sound, floppy, and microSD support with a cycle-honest emulator and browser assembler built around the same machine. ### Why it matters {#why-it-matters-mp-2026-10-04-021} An independent builder is turning an improbable idea into a working project. ### Limits and context {#limitations-mp-2026-10-04-021} - A Desk Pick is an editorial selection, not a product endorsement. ### Claims and sources {#claims-mp-2026-10-04-021} - This Invention Desk entry makes no independently sourced news claim. ## 24. M3-CRETE {#mp-2026-10-04-022} - Story ID: `mp-2026-10-04-022` - Type: `invention_desk` - Classification: `editorial` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-022/m3-crete **Dek:** Publishes a pallet-scale cartesian motion reference, CAD, BOM, and low-voltage controls for supervised cementitious-extrusion research while explicitly excluding certified structural use. Publishes a pallet-scale cartesian motion reference, CAD, BOM, and low-voltage controls for supervised cementitious-extrusion research while explicitly excluding certified structural use. ### Why it matters {#why-it-matters-mp-2026-10-04-022} An independent builder is turning an improbable idea into a working project. ### Limits and context {#limitations-mp-2026-10-04-022} - A Desk Pick is an editorial selection, not a product endorsement. ### Claims and sources {#claims-mp-2026-10-04-022} - This Invention Desk entry makes no independently sourced news claim. ## 25. DeskNode {#mp-2026-10-04-023} - Story ID: `mp-2026-10-04-023` - Type: `invention_desk` - Classification: `editorial` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-023/desknode **Dek:** Turns a bare ESP32-S3 touchscreen into a USB-connected desk instrument for PC activity and optional room temperature, humidity, and light measurements. Turns a bare ESP32-S3 touchscreen into a USB-connected desk instrument for PC activity and optional room temperature, humidity, and light measurements. ### Why it matters {#why-it-matters-mp-2026-10-04-023} An independent builder is turning an improbable idea into a working project. ### Limits and context {#limitations-mp-2026-10-04-023} - A Desk Pick is an editorial selection, not a product endorsement. ### Claims and sources {#claims-mp-2026-10-04-023} - This Invention Desk entry makes no independently sourced news claim. ## 26. The First Paid Slot {#mp-2026-10-04-024} - Story ID: `mp-2026-10-04-024` - Type: `invention_desk` - Classification: `house_example` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-024/the-first-paid-slot **Dek:** A transparent preview of paid placement with one verified link and no claim of endorsement. A transparent preview of paid placement with one verified link and no claim of endorsement. House example - no advertiser paid. Payment will buy placement, never endorsement. ### Why it matters {#why-it-matters-mp-2026-10-04-024} This placement explains how builders can appear in The Invention Desk without purchasing editorial endorsement. ### Limits and context {#limitations-mp-2026-10-04-024} - House example - no advertiser paid. Payment will buy placement, never endorsement. ### Claims and sources {#claims-mp-2026-10-04-024} - This Invention Desk entry makes no independently sourced news claim. ## 27. Put Your Project on the Desk {#mp-2026-10-04-025} - Story ID: `mp-2026-10-04-025` - Type: `invention_desk` - Classification: `house_example` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-10-04-025/put-your-project-on-the-desk **Dek:** One manually reviewed placement stays active for seven days and remains separate from Desk Picks. One manually reviewed placement stays active for seven days and remains separate from Desk Picks. Manual intake only. Payment buys placement, never endorsement, and every submission is reviewed. ### Why it matters {#why-it-matters-mp-2026-10-04-025} This placement explains how builders can appear in The Invention Desk without purchasing editorial endorsement. ### Limits and context {#limitations-mp-2026-10-04-025} - Manual intake only. Payment buys placement, never endorsement, and every submission is reviewed. ### Claims and sources {#claims-mp-2026-10-04-025} - This Invention Desk entry makes no independently sourced news claim. ## Normalized sources - **source-2026-10-04-001:** [arXiv preprint 2610.03620, original v1](https://arxiv.org/abs/2610.03620v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-002:** [arXiv preprint 2610.03448, original v1](https://arxiv.org/abs/2610.03448v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-003:** [arXiv preprint 2610.03476, original v1](https://arxiv.org/abs/2610.03476v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-004:** [arXiv preprint 2610.03515, original v1](https://arxiv.org/abs/2610.03515v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-005:** [arXiv preprint 2610.03564, original v1](https://arxiv.org/abs/2610.03564v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-006:** [arXiv preprint 2610.03402, original v1](https://arxiv.org/abs/2610.03402v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-007:** [arXiv preprint 2610.03510, original v1](https://arxiv.org/abs/2610.03510v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-008:** [arXiv preprint 2610.03389, original v1](https://arxiv.org/abs/2610.03389v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-009:** [arXiv preprint 2610.03622, original v1](https://arxiv.org/abs/2610.03622v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-010:** [arXiv preprint 2610.03454, original v1](https://arxiv.org/abs/2610.03454v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-011:** [arXiv preprint 2610.03453, original v1](https://arxiv.org/abs/2610.03453v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-012:** [arXiv preprint 2610.03390, original v1](https://arxiv.org/abs/2610.03390v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-013:** [arXiv preprint 2610.03391, original v1](https://arxiv.org/abs/2610.03391v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-014:** [arXiv preprint 2610.03388, original v1](https://arxiv.org/abs/2610.03388v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-015:** [arXiv preprint 2610.03445, original v1](https://arxiv.org/abs/2610.03445v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-016:** [arXiv preprint 2610.03519, original v1](https://arxiv.org/abs/2610.03519v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-017:** [arXiv preprint 2610.03650, original v1](https://arxiv.org/abs/2610.03650v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-018:** [arXiv preprint 2610.03430, original v1](https://arxiv.org/abs/2610.03430v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-019:** [arXiv preprint 2610.03649, original v1](https://arxiv.org/abs/2610.03649v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-020:** [arXiv preprint 2610.03387, original v1](https://arxiv.org/abs/2610.03387v1) — arXiv / submitting research authors; primary_research - **source-2026-10-04-021:** [arXiv preprint 2610.03383, original v1](https://arxiv.org/abs/2610.03383v1) — arXiv / submitting research authors; primary_research ## Corrections - 2026-10-05T08:46:40.372-04:00: Publication note: This October 4 edition was reconstructed on October 5 after the workstation was offline. It uses original October 2 submissions from the standard Sunday-evening arXiv release, rather than claiming those papers were available for the missed 9 AM run. The date and 9 AM publication slot preserve archive chronology; October 5 developments are excluded. The September 27 Invention Desk is carried over because its replacement was generated October 5. (revision 2)