--- schema_version: "1.0.0" edition_id: "mp-2026-08-19-morning-0041" published_at: "2026-08-19T09:00:00.000-04:00" modified_at: "2026-08-19T09:00:00.000-04:00" canonical_url: "https://themachinepress.com/edition/2026-08-19" story_count: 27 lead_story_id: "mp-2026-08-19-001" --- # The Machine Press — Morning edition Edition ID: `mp-2026-08-19-morning-0041` Published: 2026-08-19T09:00:00.000-04:00 Canonical edition: https://themachinepress.com/edition/2026-08-19 Researchers amplified remote timing signals to leak a co-tenant JWT at up to 12 bits per second before Cloudflare deployed stronger isolation. ## 1. The Timer Was Outside the Sandbox {#mp-2026-08-19-001} - Story ID: `mp-2026-08-19-001` - Type: `lead` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-001/the-timer-was-outside-the-sandbox **Dek:** Researchers amplified remote timing signals to leak a co-tenant JWT at up to 12 bits per second before Cloudflare deployed stronger isolation. The attack targeted the production security boundary used by Cloudflare Workers, where language-level isolation and restricted timers are intended to constrain co-located tenants. The researchers combined microarchitectural amplification with remote timing sources, reporting a Spectre-based extraction of a JWT token at up to 12 bits per second with 99.16 percent accuracy—orders of magnitude above the 2-bits-per-minute comparison they cite. The paper also records a coordinated mitigation, not an unpatched disclosure. Cloudflare integrated the V8 Sandbox, improved its Dynamic Process Isolation detection, and deployed hardware-assisted memory-protection keys to separate tenant heaps. The result is evidence about one studied production architecture and attack path; it is not a claim that every edge runtime shares the flaw or that the disclosed route remains open. ### Why it matters {#why-it-matters-mp-2026-08-19-001} Researchers amplified remote timing signals to leak a co-tenant JWT at up to 12 bits per second before Cloudflare deployed stronger isolation. ### Limits and context {#limitations-mp-2026-08-19-001} - The paper also records a coordinated mitigation, not an unpatched disclosure. - The result is evidence about one studied production architecture and attack path; it is not a claim that every edge runtime shares the flaw or that the disclosed route remains open. ### Claims and sources {#claims-mp-2026-08-19-001} - Researchers amplified remote timing signals to leak a co-tenant JWT at up to 12 bits per second before Cloudflare deployed stronger isolation. [source-2026-08-19-001] — Qualification: The paper also records a coordinated mitigation, not an unpatched disclosure. ## 2. The Muscles Learned to Sprint by Holding Still {#mp-2026-08-19-002} - Story ID: `mp-2026-08-19-002` - Type: `secondary` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-002/the-muscles-learned-to-sprint-by-holding-still **Dek:** A physiology-inspired controller held per-muscle threshold lengths across gait phases and trained a simulated skeleton to sprint with a minimal reward. A muscle-actuated skeleton has far more control variables than a conventional rigid-body character, making reinforcement-learning exploration slow and unstable. The proposed lambda-hold controller reduces that burden by setting an equilibrium-point threshold length for each muscle and holding it over part of the gait cycle; a stretch-reflex rule then turns those thresholds into muscle excitation without requiring the policy to issue every low-level command. The authors report that this controller learned human-like sprinting within an hour of training using only a minimal task reward. The work connects an engineering control scheme to the equilibrium-point hypothesis, intermittent control and optimal feedback control, but it remains predictive musculoskeletal simulation. It does not establish a clinical model of human movement or prove that the learned controller matches biological motor control in people. ### Why it matters {#why-it-matters-mp-2026-08-19-002} A physiology-inspired controller held per-muscle threshold lengths across gait phases and trained a simulated skeleton to sprint with a minimal reward. ### Limits and context {#limitations-mp-2026-08-19-002} - The authors report that this controller learned human-like sprinting within an hour of training using only a minimal task reward. - It does not establish a clinical model of human movement or prove that the learned controller matches biological motor control in people. ### Claims and sources {#claims-mp-2026-08-19-002} - A physiology-inspired controller held per-muscle threshold lengths across gait phases and trained a simulated skeleton to sprint with a minimal reward. [source-2026-08-19-002] — Qualification: The authors report that this controller learned human-like sprinting within an hour of training using only a minimal task reward. ## 3. The Humanoid Learned to Walk to the Grasp {#mp-2026-08-19-003} - Story ID: `mp-2026-08-19-003` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-003/the-humanoid-learned-to-walk-to-the-grasp **Dek:** Reinforcement learning pushed a cloned policy past its synthetic-data ceiling and transferred a reach-and-pick skill to a real humanoid. FetchMan begins with synthetic demonstrations across more than 150,000 scenes, then refines the cloned policy with Flow-GRPO and a sparse reward. The authors report 73.3 percent zero-shot success when a Unitree G1 walked toward and grasped a single target in unseen real scenes. The result supports the sim-to-real recipe for one reach-and-pick policy; it does not establish general household manipulation. ### Why it matters {#why-it-matters-mp-2026-08-19-003} Reinforcement learning pushed a cloned policy past its synthetic-data ceiling and transferred a reach-and-pick skill to a real humanoid. ### Limits and context {#limitations-mp-2026-08-19-003} - The result supports the sim-to-real recipe for one reach-and-pick policy; it does not establish general household manipulation. ### Claims and sources {#claims-mp-2026-08-19-003} - Reinforcement learning pushed a cloned policy past its synthetic-data ceiling and transferred a reach-and-pick skill to a real humanoid. [source-2026-08-19-003] — Qualification: The result supports the sim-to-real recipe for one reach-and-pick policy; it does not establish general household manipulation. ## 4. The Gold Standard Had Missed the Hospital’s Own Names {#mp-2026-08-19-004} - Story ID: `mp-2026-08-19-004` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-004/the-gold-standard-had-missed-the-hospital-s-own-names **Dek:** Institution-specific prompting found locally meaningful protected-health identifiers and exposed 227 omissions in the reference annotations. Across 100 pediatric-oncology notes containing 5,322 annotated PHI spans, the best prompted LLM exceeded two purpose-built de-identification systems in the authors’ comparison. Naming institution-specific categories recovered 48 of 61 misses, and re-annotation confirmed 227 additional PHI spans surfaced by model outputs. No tested multi-agent setup beat the calibrated single pass, and the result belongs to one hospital dataset rather than a universal de-identification guarantee. ### Why it matters {#why-it-matters-mp-2026-08-19-004} Institution-specific prompting found locally meaningful protected-health identifiers and exposed 227 omissions in the reference annotations. ### Limits and context {#limitations-mp-2026-08-19-004} - No additional limitation was separately recorded. ### Claims and sources {#claims-mp-2026-08-19-004} - Institution-specific prompting found locally meaningful protected-health identifiers and exposed 227 omissions in the reference annotations. [source-2026-08-19-004] ## 5. The Classifier Gave Up Its Decision Rules {#mp-2026-08-19-005} - Story ID: `mp-2026-08-19-005` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-005/the-classifier-gave-up-its-decision-rules **Dek:** J-Miner turned internal signals into named concepts and compact executable rules that reproduced up to 98.3 percent of model decisions. J-Miner aggregates vocabulary-aligned signals across layers and token positions, then learns explicit rules from the classifier’s own outputs. Across the reported tasks, those rules were 6.0 to 29.5 percentage points more faithful than equally compact word-based rules. Lightweight students with about one twenty-fourth the parameters retained 99.8 percent of mean source accuracy, showing a route to inspectable reuse without proving that every mined concept is causally decisive. ### Why it matters {#why-it-matters-mp-2026-08-19-005} J-Miner turned internal signals into named concepts and compact executable rules that reproduced up to 98.3 percent of model decisions. ### Limits and context {#limitations-mp-2026-08-19-005} - No additional limitation was separately recorded. ### Claims and sources {#claims-mp-2026-08-19-005} - J-Miner turned internal signals into named concepts and compact executable rules that reproduced up to 98.3 percent of model decisions. [source-2026-08-19-005] ## 6. The Auditor Verified the Network Without Seeing It {#mp-2026-08-19-006} - Story ID: `mp-2026-08-19-006` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-006/the-auditor-verified-the-network-without-seeing-it **Dek:** PANDA used zero-knowledge proofs to certify robustness and fairness properties while keeping model parameters private. Built on CROWN’s linear-relaxation bounds, PANDA produces proofs that reveal a claimed property without exposing the model weights. The authors report generating a local-robustness proof for networks above 2.9 million parameters in five minutes and verifying it in ten seconds—four orders of magnitude larger than prior compared systems. These are system benchmarks for supported guarantees, not evidence that a certified model is safe or fair outside the stated property and threat model. ### Why it matters {#why-it-matters-mp-2026-08-19-006} PANDA used zero-knowledge proofs to certify robustness and fairness properties while keeping model parameters private. ### Limits and context {#limitations-mp-2026-08-19-006} - These are system benchmarks for supported guarantees, not evidence that a certified model is safe or fair outside the stated property and threat model. ### Claims and sources {#claims-mp-2026-08-19-006} - PANDA used zero-knowledge proofs to certify robustness and fairness properties while keeping model parameters private. [source-2026-08-19-006] — Qualification: These are system benchmarks for supported guarantees, not evidence that a certified model is safe or fair outside the stated property and threat model. ## 7. The Forecast Had to Cite Evidence for a Future Code {#mp-2026-08-19-007} - Story ID: `mp-2026-08-19-007` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-007/the-forecast-had-to-cite-evidence-for-a-future-code **Dek:** A bounded research workflow combined an EHR prior, medical search and code dictionaries before ranking next-encounter diagnoses. ICD-Deepresearch predicts diagnosis codes that may appear at a future visit, where no source can reveal the answer in advance. It expands candidates from an EHR foundation model, adds a separate direct forecast, validates exact code semantics and writes rationales only after selection. Patient-averaged precision remained about 25 percent, while recall reached 35.09 percent on MIMIC-III and 48.32 percent on MIMIC-IV; physicians nevertheless rated its retrieved documents more useful than the standalone research comparisons. ### Why it matters {#why-it-matters-mp-2026-08-19-007} A bounded research workflow combined an EHR prior, medical search and code dictionaries before ranking next-encounter diagnoses. ### Limits and context {#limitations-mp-2026-08-19-007} - It expands candidates from an EHR foundation model, adds a separate direct forecast, validates exact code semantics and writes rationales only after selection. ### Claims and sources {#claims-mp-2026-08-19-007} - A bounded research workflow combined an EHR prior, medical search and code dictionaries before ranking next-encounter diagnoses. [source-2026-08-19-007] — Qualification: It expands candidates from an EHR foundation model, adds a separate direct forecast, validates exact code semantics and writes rationales only after selection. ## 8. The Wider Interval Was More Honest {#mp-2026-08-19-008} - Story ID: `mp-2026-08-19-008` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-008/the-wider-interval-was-more-honest **Dek:** A regime-aware conformal method traded sharpness for more dependable coverage across 48 inflation, macroeconomic, energy and market series. DRACP combines weighting for shifts and local regimes with an online controller that adjusts the target significance level. It was not the narrowest method: a strongly adaptive comparator produced intervals about 20 percent tighter. DRACP instead came closest to nominal 90 percent coverage at 0.890, never fell below 0.80 on any series and undercovered on 10 of 48 series versus 20 for that tighter comparator. The study frames reliability as an explicit cost in forecast width. ### Why it matters {#why-it-matters-mp-2026-08-19-008} A regime-aware conformal method traded sharpness for more dependable coverage across 48 inflation, macroeconomic, energy and market series. ### Limits and context {#limitations-mp-2026-08-19-008} - It was not the narrowest method: a strongly adaptive comparator produced intervals about 20 percent tighter. ### Claims and sources {#claims-mp-2026-08-19-008} - A regime-aware conformal method traded sharpness for more dependable coverage across 48 inflation, macroeconomic, energy and market series. [source-2026-08-19-008] — Qualification: It was not the narrowest method: a strongly adaptive comparator produced intervals about 20 percent tighter. ## 9. The Sensor Lie Left Footprints on the Bus {#mp-2026-08-19-009} - Story ID: `mp-2026-08-19-009` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-009/the-sensor-lie-left-footprints-on-the-bus **Dek:** SentryBus maps where I2C manipulation may remain observable—and where a host-side monitor loses the ability to distinguish it. The proposed monitor uses transaction timing, read and write sequences, transfer lengths, address behavior and raw-data transitions from the host-facing sensor bus. A dual-sided testbed established timing and capture limits, including a low-cost analyzer that could silently truncate data. Controlled attack trials are still outstanding and the authors claim no detection rate, making the negative boundary result as important as the proposed feature set. ### Why it matters {#why-it-matters-mp-2026-08-19-009} SentryBus maps where I2C manipulation may remain observable—and where a host-side monitor loses the ability to distinguish it. ### Limits and context {#limitations-mp-2026-08-19-009} - No additional limitation was separately recorded. ### Claims and sources {#claims-mp-2026-08-19-009} - SentryBus maps where I2C manipulation may remain observable—and where a host-side monitor loses the ability to distinguish it. [source-2026-08-19-009] ## 10. The Car Compared Two Stories About Where It Was {#mp-2026-08-19-010} - Story ID: `mp-2026-08-19-010` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-010/the-car-compared-two-stories-about-where-it-was **Dek:** Structured narratives turned disagreements between GNSS and independent vehicle sensing into a five-class spoofing detector. The framework converts two independently derived driving states into text-like structured narratives and feeds them to a small language model. Across no-attack, overshoot, stopped, turn-by-turn and wrong-turn cases, the authors report 96.99 percent average accuracy and 97.18 percent F1, with lower latency and memory than their fine-tuned LLM comparisons. Tests on geographically unseen Clemson field data support transfer within the study, not universal resilience to adversarial driving conditions. ### Why it matters {#why-it-matters-mp-2026-08-19-010} Structured narratives turned disagreements between GNSS and independent vehicle sensing into a five-class spoofing detector. ### Limits and context {#limitations-mp-2026-08-19-010} - Tests on geographically unseen Clemson field data support transfer within the study, not universal resilience to adversarial driving conditions. ### Claims and sources {#claims-mp-2026-08-19-010} - Structured narratives turned disagreements between GNSS and independent vehicle sensing into a five-class spoofing detector. [source-2026-08-19-010] — Qualification: Tests on geographically unseen Clemson field data support transfer within the study, not universal resilience to adversarial driving conditions. ## 11. The Driving Model Looked Harder—Then Moved Closer {#mp-2026-08-19-011} - Story ID: `mp-2026-08-19-011` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-011/the-driving-model-looked-harder-then-moved-closer **Dek:** A bounded inference-time attention bias shifted planned trajectories toward a highlighted traffic actor without retraining the model. The experiment added a pre-softmax bias to detector-localized visual tokens in 50 synthetic lane-change scenarios. Trajectory displacement increased with the bias, reaching about 17 centimeters on average and up to roughly 140 centimeters laterally. The reasoning text did not change because the mask never reached that pathway, and the steering tended to move toward the attended actor rather than encode a safe target behavior—an important limit on treating attention control as safety control. ### Why it matters {#why-it-matters-mp-2026-08-19-011} A bounded inference-time attention bias shifted planned trajectories toward a highlighted traffic actor without retraining the model. ### Limits and context {#limitations-mp-2026-08-19-011} - The reasoning text did not change because the mask never reached that pathway, and the steering tended to move toward the attended actor rather than encode a safe target behavior—an important limit on treating attention control as safety control. ### Claims and sources {#claims-mp-2026-08-19-011} - A bounded inference-time attention bias shifted planned trajectories toward a highlighted traffic actor without retraining the model. [source-2026-08-19-011] — Qualification: The reasoning text did not change because the mask never reached that pathway, and the steering tended to move toward the attended actor rather than encode a safe target behavior—an important limit on treating attention control as safety control. ## 12. One Prompt Turned 18.6 Into 5.4 {#mp-2026-08-19-012} - Story ID: `mp-2026-08-19-012` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-012/one-prompt-turned-18-6-into-5-4 **Dek:** OV3D-Bench separated localization from semantics and found that small prompt changes could collapse open-vocabulary 3D detection. Across seven indoor and outdoor datasets, the benchmark removed per-image category oracles and scored localization, semantic robustness and transfer separately. Detectors often placed a box correctly but chose an adjacent label; one model’s AP fell from 18.6 to 5.4 when its prompt expanded from “car” to a detailed photographic phrase. The study also found a target-aware protocol inflated another model’s ScanNet result by 1.9 times, suggesting geometry is ahead of open-vocabulary semantics. ### Why it matters {#why-it-matters-mp-2026-08-19-012} OV3D-Bench separated localization from semantics and found that small prompt changes could collapse open-vocabulary 3D detection. ### Limits and context {#limitations-mp-2026-08-19-012} - No additional limitation was separately recorded. ### Claims and sources {#claims-mp-2026-08-19-012} - OV3D-Bench separated localization from semantics and found that small prompt changes could collapse open-vocabulary 3D detection. [source-2026-08-19-012] ## 13. The Critic Found the Safe Region Before the Equation Tightened It {#mp-2026-08-19-013} - Story ID: `mp-2026-08-19-013` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-013/the-critic-found-the-safe-region-before-the-equation-tightened-it **Dek:** A scheduled method let reinforcement learning shape the value function before progressively enforcing the governing reach-avoid PDE. High-dimensional reach-avoid analysis is difficult for grid solvers, while physics-informed networks may settle in poor residual minima and reinforcement learning may violate the governing equation. The proposed schedule starts with temporal-difference actor-critic learning, then adds PDE and boundary losses gradually. Two case studies reached accuracy comparable to successfully trained PINNs while mitigating their reported failure mode; the paper does not yet establish scaling across broad safety-critical systems. ### Why it matters {#why-it-matters-mp-2026-08-19-013} A scheduled method let reinforcement learning shape the value function before progressively enforcing the governing reach-avoid PDE. ### Limits and context {#limitations-mp-2026-08-19-013} - Two case studies reached accuracy comparable to successfully trained PINNs while mitigating their reported failure mode; the paper does not yet establish scaling across broad safety-critical systems. ### Claims and sources {#claims-mp-2026-08-19-013} - A scheduled method let reinforcement learning shape the value function before progressively enforcing the governing reach-avoid PDE. [source-2026-08-19-013] — Qualification: Two case studies reached accuracy comparable to successfully trained PINNs while mitigating their reported failure mode; the paper does not yet establish scaling across broad safety-critical systems. ## 14. The Hidden State Knew When the Vote Was Wrong {#mp-2026-08-19-014} - Story ID: `mp-2026-08-19-014` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-014/the-hidden-state-knew-when-the-vote-was-wrong **Dek:** A leakage-free decodability score predicted when selecting an answer from internal signals would beat majority voting. CASE trains a linear gate on answer-token hidden states and chooses the highest-ranked sample. Its decodability measure predicted the gain over voting with a reported correlation of 0.75; across general and medical models, selection improved medium-difficulty accuracy by up to 19 points and hard questions by 16.8 points. A conventional probe looked strong only because question identity leaked across evaluation groups, underscoring that the criterion must itself be tested without leakage. ### Why it matters {#why-it-matters-mp-2026-08-19-014} A leakage-free decodability score predicted when selecting an answer from internal signals would beat majority voting. ### Limits and context {#limitations-mp-2026-08-19-014} - A conventional probe looked strong only because question identity leaked across evaluation groups, underscoring that the criterion must itself be tested without leakage. ### Claims and sources {#claims-mp-2026-08-19-014} - A leakage-free decodability score predicted when selecting an answer from internal signals would beat majority voting. [source-2026-08-19-014] — Qualification: A conventional probe looked strong only because question identity leaked across evaluation groups, underscoring that the criterion must itself be tested without leakage. ## 15. The Robot Moved the Clutter Before It Answered {#mp-2026-08-19-026} - Story ID: `mp-2026-08-19-026` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-026/the-robot-moved-the-clutter-before-it-answered **Dek:** PROBE turns visual question answering into an active task where a robot must reveal occluded objects before deciding. PROBE-Bench contains 150 simulated tabletop tasks across six question types, with picking and pushing tools that change the scene between observations. Agentic tool use outperformed perception-only baselines by 8.0 percent on average, and distilled open-weight agents gained another 11.5 percent over their off-the-shelf agent baselines. The authors also report real-tabletop transfer, but the benchmark remains a bounded step toward robots that can answer questions about hidden household state. ### Why it matters {#why-it-matters-mp-2026-08-19-026} PROBE turns visual question answering into an active task where a robot must reveal occluded objects before deciding. ### Limits and context {#limitations-mp-2026-08-19-026} - Agentic tool use outperformed perception-only baselines by 8.0 percent on average, and distilled open-weight agents gained another 11.5 percent over their off-the-shelf agent baselines. ### Claims and sources {#claims-mp-2026-08-19-026} - PROBE turns visual question answering into an active task where a robot must reveal occluded objects before deciding. [source-2026-08-19-015] — Qualification: Agentic tool use outperformed perception-only baselines by 8.0 percent on average, and distilled open-weight agents gained another 11.5 percent over their off-the-shelf agent baselines. ## 16. The Circuit Paid for Every Bit It Erased {#mp-2026-08-19-027} - Story ID: `mp-2026-08-19-027` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-027/the-circuit-paid-for-every-bit-it-erased **Dek:** Renesis converts ordinary netlists into verified energy-recovery circuits and rejects rewrites that worsen either reported energy table. The open-source synthesis tool tracks switching, erasure and observability while mapping a logically reversible result to one of eight adiabatic circuit families. Every transformation is equivalence-checked and must improve at least one cost table without worsening the other. Fifteen of nineteen held-out circuits improved, with a best-arm median energy of 0.91 times the default. The accounting is circuit-level and technology-specific, not a claim that logical reversibility alone removes thermodynamic cost. ### Why it matters {#why-it-matters-mp-2026-08-19-027} Renesis converts ordinary netlists into verified energy-recovery circuits and rejects rewrites that worsen either reported energy table. ### Limits and context {#limitations-mp-2026-08-19-027} - The accounting is circuit-level and technology-specific, not a claim that logical reversibility alone removes thermodynamic cost. ### Claims and sources {#claims-mp-2026-08-19-027} - Renesis converts ordinary netlists into verified energy-recovery circuits and rejects rewrites that worsen either reported energy table. [source-2026-08-19-016] — Qualification: The accounting is circuit-level and technology-specific, not a claim that logical reversibility alone removes thermodynamic cost. ## 17. One Demonstration Wrote the Planner’s World {#mp-2026-08-19-015} - Story ID: `mp-2026-08-19-015` - Type: `ticker` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-015/one-demonstration-wrote-the-planner-s-world **Dek:** PDDL-ART generated symbolic domains and problems from one expert demonstration, then corrected them through syntax, semantics and execution. Across engine-maintenance and household tasks, the VLM-based system reported 93.3 percent average success versus 78.3 percent for its planner baseline. Tool-grounded predicate checks helped with relations not visible in a single frame; performance still depends on the demonstration and available high-level actions. ### Why it matters {#why-it-matters-mp-2026-08-19-015} PDDL-ART generated symbolic domains and problems from one expert demonstration, then corrected them through syntax, semantics and execution. ### Limits and context {#limitations-mp-2026-08-19-015} - Tool-grounded predicate checks helped with relations not visible in a single frame; performance still depends on the demonstration and available high-level actions. ### Claims and sources {#claims-mp-2026-08-19-015} - PDDL-ART generated symbolic domains and problems from one expert demonstration, then corrected them through syntax, semantics and execution. [source-2026-08-19-017] — Qualification: Tool-grounded predicate checks helped with relations not visible in a single frame; performance still depends on the demonstration and available high-level actions. ## 18. Tissue Context Corrected the Cell Label {#mp-2026-08-19-016} - Story ID: `mp-2026-08-19-016` - Type: `ticker` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-016/tissue-context-corrected-the-cell-label **Dek:** Loki-OT used region-level tissue estimates as soft guidance when visually similar cells confused a cell classifier. On an independent breast-cancer cohort, the method reduced patient-level error and improved F1 in epithelium-rich mimicry regions using 278 weak region estimates. The work is a computational pathology benchmark, not a clinical diagnostic validation. ### Why it matters {#why-it-matters-mp-2026-08-19-016} Loki-OT used region-level tissue estimates as soft guidance when visually similar cells confused a cell classifier. ### Limits and context {#limitations-mp-2026-08-19-016} - The work is a computational pathology benchmark, not a clinical diagnostic validation. ### Claims and sources {#claims-mp-2026-08-19-016} - Loki-OT used region-level tissue estimates as soft guidance when visually similar cells confused a cell classifier. [source-2026-08-19-018] — Qualification: The work is a computational pathology benchmark, not a clinical diagnostic validation. ## 19. The World Model Searched Without Training on Imagined Data {#mp-2026-08-19-017} - Story ID: `mp-2026-08-19-017` - Type: `ticker` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-017/the-world-model-searched-without-training-on-imagined-data **Dek:** QWM used predicted trajectories to select actions while keeping policy and value learning grounded in real transitions. Test-time search over imagined rollouts improved sample efficiency and performance on Robomimic and LIBERO in the authors’ comparison. Avoiding direct training on imagined transitions reduces one source of compounding bias but does not eliminate world-model error during action selection. ### Why it matters {#why-it-matters-mp-2026-08-19-017} QWM used predicted trajectories to select actions while keeping policy and value learning grounded in real transitions. ### Limits and context {#limitations-mp-2026-08-19-017} - Avoiding direct training on imagined transitions reduces one source of compounding bias but does not eliminate world-model error during action selection. ### Claims and sources {#claims-mp-2026-08-19-017} - QWM used predicted trajectories to select actions while keeping policy and value learning grounded in real transitions. [source-2026-08-19-019] — Qualification: Avoiding direct training on imagined transitions reduces one source of compounding bias but does not eliminate world-model error during action selection. ## 20. One Aerial Pass Estimated the Debris {#mp-2026-08-19-018} - Story ID: `mp-2026-08-19-018` - Type: `ticker` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-018/one-aerial-pass-estimated-the-debris **Dek:** DebrisHeightNet converted post-hurricane RGB imagery into spatial volume estimates without LiDAR at deployment time. Across ten regions from five hurricanes, the uncalibrated model correlated at 0.87 with an independent UAV survey and came within 30 percent of reported hauled debris where parametric forecasts overshot by 2.7 to 4.8 times. Its training target was fused supervision, not ground truth, and the authors explicitly quantify that limitation. ### Why it matters {#why-it-matters-mp-2026-08-19-018} DebrisHeightNet converted post-hurricane RGB imagery into spatial volume estimates without LiDAR at deployment time. ### Limits and context {#limitations-mp-2026-08-19-018} - Its training target was fused supervision, not ground truth, and the authors explicitly quantify that limitation. ### Claims and sources {#claims-mp-2026-08-19-018} - DebrisHeightNet converted post-hurricane RGB imagery into spatial volume estimates without LiDAR at deployment time. [source-2026-08-19-020] — Qualification: Its training target was fused supervision, not ground truth, and the authors explicitly quantify that limitation. ## 21. The Tests Got a Contract Before They Got Code {#mp-2026-08-19-019} - Story ID: `mp-2026-08-19-019` - Type: `ticker` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-019/the-tests-got-a-contract-before-they-got-code **Dek:** An agent wrote preconditions, postconditions and undefined behavior before generating a test suite. On production bugs from Google, spec-driven generation improved bug detection by 9.8 percentage points and branch coverage by 2.5 points over the direct agent baseline. An LLM judge also preferred many generated suites, a result that complements but does not replace executable bug and coverage measurements. ### Why it matters {#why-it-matters-mp-2026-08-19-019} An agent wrote preconditions, postconditions and undefined behavior before generating a test suite. ### Limits and context {#limitations-mp-2026-08-19-019} - An LLM judge also preferred many generated suites, a result that complements but does not replace executable bug and coverage measurements. ### Claims and sources {#claims-mp-2026-08-19-019} - An agent wrote preconditions, postconditions and undefined behavior before generating a test suite. [source-2026-08-19-021] — Qualification: An LLM judge also preferred many generated suites, a result that complements but does not replace executable bug and coverage measurements. ## 22. openDogV3 {#mp-2026-08-19-020} - Story ID: `mp-2026-08-19-020` - Type: `invention_desk` - Classification: `editorial` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-020/opendogv3 **Dek:** Supplies CAD, code, and a bill of materials for a PLA-printed quadruped with motor-driven joints, closed-loop controls, and an inverse-kinematics walking mode. Supplies CAD, code, and a bill of materials for a PLA-printed quadruped with motor-driven joints, closed-loop controls, and an inverse-kinematics walking mode. ### Why it matters {#why-it-matters-mp-2026-08-19-020} An independent builder is turning an improbable idea into a working project. ### Limits and context {#limitations-mp-2026-08-19-020} - A Desk Pick is an editorial selection, not a product endorsement. ### Claims and sources {#claims-mp-2026-08-19-020} - This Invention Desk entry makes no independently sourced news claim. ## 23. OpenKnit {#mp-2026-08-19-021} - Story ID: `mp-2026-08-19-021` - Type: `invention_desk` - Classification: `editorial` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-021/openknit **Dek:** Aims to turn digital garment files into knitted pieces on an open-source machine; its smaller Wally120 design is easier to assemble, but the project remains early beta hardware. Aims to turn digital garment files into knitted pieces on an open-source machine; its smaller Wally120 design is easier to assemble, but the project remains early beta hardware. ### Why it matters {#why-it-matters-mp-2026-08-19-021} An independent builder is turning an improbable idea into a working project. ### Limits and context {#limitations-mp-2026-08-19-021} - A Desk Pick is an editorial selection, not a product endorsement. ### Claims and sources {#claims-mp-2026-08-19-021} - This Invention Desk entry makes no independently sourced news claim. ## 24. PicoGUS {#mp-2026-08-19-022} - Story ID: `mp-2026-08-19-022` - Type: `invention_desk` - Classification: `editorial` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-022/picogus **Dek:** Uses an RP2040 microcontroller to emulate several ISA sound cards and a period CD-ROM interface for retro PCs, with open hardware files and assembled cards available. Uses an RP2040 microcontroller to emulate several ISA sound cards and a period CD-ROM interface for retro PCs, with open hardware files and assembled cards available. ### Why it matters {#why-it-matters-mp-2026-08-19-022} An independent builder is turning an improbable idea into a working project. ### Limits and context {#limitations-mp-2026-08-19-022} - A Desk Pick is an editorial selection, not a product endorsement. ### Claims and sources {#claims-mp-2026-08-19-022} - This Invention Desk entry makes no independently sourced news claim. ## 25. Phoniebox {#mp-2026-08-19-023} - Story ID: `mp-2026-08-19-023` - Type: `invention_desk` - Classification: `editorial` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-023/phoniebox **Dek:** Turns RFID cards into selectors for local audio, playlists, podcasts, and web streams on a Raspberry Pi, with USB-reader setups and optional physical controls. Turns RFID cards into selectors for local audio, playlists, podcasts, and web streams on a Raspberry Pi, with USB-reader setups and optional physical controls. ### Why it matters {#why-it-matters-mp-2026-08-19-023} An independent builder is turning an improbable idea into a working project. ### Limits and context {#limitations-mp-2026-08-19-023} - A Desk Pick is an editorial selection, not a product endorsement. ### Claims and sources {#claims-mp-2026-08-19-023} - This Invention Desk entry makes no independently sourced news claim. ## 26. The First Paid Slot {#mp-2026-08-19-024} - Story ID: `mp-2026-08-19-024` - Type: `invention_desk` - Classification: `house_example` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-024/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-2026-08-19-024} This placement explains how builders can appear in The Invention Desk without purchasing editorial endorsement. ### Limits and context {#limitations-mp-2026-08-19-024} - House example - no advertiser paid. Payment will buy placement, never endorsement. ### Claims and sources {#claims-mp-2026-08-19-024} - This Invention Desk entry makes no independently sourced news claim. ## 27. Put Your Project on the Desk {#mp-2026-08-19-025} - Story ID: `mp-2026-08-19-025` - Type: `invention_desk` - Classification: `house_example` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-08-19-025/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-2026-08-19-025} This placement explains how builders can appear in The Invention Desk without purchasing editorial endorsement. ### Limits and context {#limitations-mp-2026-08-19-025} - Manual intake only. Payment buys placement, never endorsement, and every submission is reviewed. ### Claims and sources {#claims-mp-2026-08-19-025} - This Invention Desk entry makes no independently sourced news claim. ## Normalized sources - **source-2026-08-19-001:** [arXiv preprint 2608.17043](https://arxiv.org/abs/2608.17043) — arXiv; primary_research - **source-2026-08-19-002:** [arXiv preprint 2608.17030](https://arxiv.org/abs/2608.17030) — arXiv; primary_research - **source-2026-08-19-003:** [arXiv preprint 2608.17027](https://arxiv.org/abs/2608.17027) — arXiv; primary_research - **source-2026-08-19-004:** [arXiv preprint 2608.17051](https://arxiv.org/abs/2608.17051) — arXiv; primary_research - **source-2026-08-19-005:** [arXiv preprint 2608.17063](https://arxiv.org/abs/2608.17063) — arXiv; primary_research - **source-2026-08-19-006:** [arXiv preprint 2608.17070](https://arxiv.org/abs/2608.17070) — arXiv; primary_research - **source-2026-08-19-007:** [arXiv preprint 2608.17075](https://arxiv.org/abs/2608.17075) — arXiv; primary_research - **source-2026-08-19-008:** [arXiv preprint 2608.17079](https://arxiv.org/abs/2608.17079) — arXiv; primary_research - **source-2026-08-19-009:** [arXiv preprint 2608.17082](https://arxiv.org/abs/2608.17082) — arXiv; primary_research - **source-2026-08-19-010:** [arXiv preprint 2608.17092](https://arxiv.org/abs/2608.17092) — arXiv; primary_research - **source-2026-08-19-011:** [arXiv preprint 2608.17095](https://arxiv.org/abs/2608.17095) — arXiv; primary_research - **source-2026-08-19-012:** [arXiv preprint 2608.17110](https://arxiv.org/abs/2608.17110) — arXiv; primary_research - **source-2026-08-19-013:** [arXiv preprint 2608.17117](https://arxiv.org/abs/2608.17117) — arXiv; primary_research - **source-2026-08-19-014:** [arXiv preprint 2608.17124](https://arxiv.org/abs/2608.17124) — arXiv; primary_research - **source-2026-08-19-015:** [arXiv preprint 2608.17129](https://arxiv.org/abs/2608.17129) — arXiv; primary_research - **source-2026-08-19-016:** [arXiv preprint 2608.17139](https://arxiv.org/abs/2608.17139) — arXiv; primary_research - **source-2026-08-19-017:** [arXiv preprint 2608.17146](https://arxiv.org/abs/2608.17146) — arXiv; primary_research - **source-2026-08-19-018:** [arXiv preprint 2608.17151](https://arxiv.org/abs/2608.17151) — arXiv; primary_research - **source-2026-08-19-019:** [arXiv preprint 2608.17163](https://arxiv.org/abs/2608.17163) — arXiv; primary_research - **source-2026-08-19-020:** [arXiv preprint 2608.17165](https://arxiv.org/abs/2608.17165) — arXiv; primary_research - **source-2026-08-19-021:** [arXiv preprint 2608.17177](https://arxiv.org/abs/2608.17177) — arXiv; primary_research