---
schema_version: "1.0.0"
edition_id: "mp-2026-09-15-morning-0068"
published_at: "2026-09-15T09:00:00.000-04:00"
modified_at: "2026-09-15T09:00:00.000-04:00"
canonical_url: "https://themachinepress.com/edition/2026-09-15"
story_count: 27
lead_story_id: "mp-2026-09-15-001"
---

# The Machine Press — Morning edition

Edition ID: `mp-2026-09-15-morning-0068`  
Published: 2026-09-15T09:00:00.000-04:00  
Canonical edition: https://themachinepress.com/edition/2026-09-15

Continual Search lifted GPT-5.5 root-cause F1 from 0.349 to 0.498 on 50 human-annotated, long-horizon agent failures.

## 1. The Search Kept Going After the First Plausible Failure {#mp-2026-09-15-001}

- Story ID: `mp-2026-09-15-001`
- Type: `lead`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-001/the-search-kept-going-after-the-first-plausible-failure

**Dek:** Continual Search lifted GPT-5.5 root-cause F1 from 0.349 to 0.498 on 50 human-annotated, long-horizon agent failures.

Long agent runs can bury the decisive evidence hundreds of actions away from the visible failure. A one-shot judge may settle early on a diagnosis that sounds plausible while leaving most of the trace unexplored. Continual Search instead asks the judge to keep looking across successive turns for unresolved evidence. Across four existing root-cause benchmarks and the new 50-trial MegaRCA-Mix set, the authors report consistent gains; on MegaRCA-Mix, GPT-5.5 F1 rose more than 40%, from 0.349 to 0.498. The benchmark is research evidence, not proof that automated diagnosis can replace production incident review.

### Why it matters {#why-it-matters-mp-2026-09-15-001}

Continual Search lifted GPT-5.5 root-cause F1 from 0.349 to 0.498 on 50 human-annotated, long-horizon agent failures.

### Limits and context {#limitations-mp-2026-09-15-001}

- The benchmark is research evidence, not proof that automated diagnosis can replace production incident review.

### Claims and sources {#claims-mp-2026-09-15-001}

- Continual Search lifted GPT-5.5 root-cause F1 from 0.349 to 0.498 on 50 human-annotated, long-horizon agent failures. [source-2026-09-15-001] — Qualification: The benchmark is research evidence, not proof that automated diagnosis can replace production incident review.

## 2. The Robot Learned Where Reward Had to Stop {#mp-2026-09-15-002}

- Story ID: `mp-2026-09-15-002`
- Type: `secondary`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-002/the-robot-learned-where-reward-had-to-stop

**Dek:** A reachability-based safety critic cut cumulative safety cost 57% on average while improving task success by 0.13 over SafeVLA.

Safety fine-tuning for vision-language-action models often treats violations as a soft penalty, forcing one objective to trade reward against risk. ShieldVLA instead learns a model-free approximation of a Hamilton–Jacobi reachability value function from visual observations. The critic separates ordinary reward optimization inside the feasible region from recovery behavior near unsafe states; rubric-based vision-language scores provide training targets without dense manual cost labels. Across five navigation and manipulation benchmarks and multiple VLA backbones, the authors report 57% lower cumulative safety cost on average and a 0.13 gain in task success over SafeVLA. Those benchmark results do not constitute a formal guarantee for an untested physical deployment.

### Why it matters {#why-it-matters-mp-2026-09-15-002}

A reachability-based safety critic cut cumulative safety cost 57% on average while improving task success by 0.13 over SafeVLA.

### Limits and context {#limitations-mp-2026-09-15-002}

- Those benchmark results do not constitute a formal guarantee for an untested physical deployment.

### Claims and sources {#claims-mp-2026-09-15-002}

- A reachability-based safety critic cut cumulative safety cost 57% on average while improving task success by 0.13 over SafeVLA. [source-2026-09-15-002] — Qualification: Those benchmark results do not constitute a formal guarantee for an untested physical deployment.

## 3. A Seven-Billion-Parameter Model Opened Its Entire Training Trail {#mp-2026-09-15-003}

- Story ID: `mp-2026-09-15-003`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-003/a-seven-billion-parameter-model-opened-its-entire-training-trail

**Dek:** ZGCM-1 reports a 4.2-fold improvement in 16K pretraining time-to-loss and releases weights, checkpoints, code, data recipes and logs.

ZGCM-1 is a dense seven-billion-parameter foundation model built around the claim that compact models should combine internal reasoning with external tool use instead of trying to memorize the open web. Its recipe mixes sliding-window and full attention, FP8 Muon optimization, progressive context scaling to 256K and interaction traces recast as Markov decision processes. The team reports roughly 4.2× better 16K pretraining time-to-loss and competitive results against larger models on selected math and agentic-search suites. It also releases stage-by-stage weights, checkpoints, code, data recipes and experiment logs; independent replication remains necessary.

### Why it matters {#why-it-matters-mp-2026-09-15-003}

ZGCM-1 reports a 4.2-fold improvement in 16K pretraining time-to-loss and releases weights, checkpoints, code, data recipes and logs.

### Limits and context {#limitations-mp-2026-09-15-003}

- No additional limitation was separately recorded.

### Claims and sources {#claims-mp-2026-09-15-003}

- ZGCM-1 reports a 4.2-fold improvement in 16K pretraining time-to-loss and releases weights, checkpoints, code, data recipes and logs. [source-2026-09-15-003]

## 4. The Optimizer Reached One Good Point Before Spreading Out {#mp-2026-09-15-004}

- Story ID: `mp-2026-09-15-004`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-004/the-optimizer-reached-one-good-point-before-spreading-out

**Dek:** Across 446 pairwise comparisons, a two-stage Bayesian-optimization strategy won 72.9%, tied 21.1% and lost 6.1%.

Multi-objective Bayesian optimization normally tries to approach the Pareto front and preserve diversity at the same time. Under tight evaluation budgets, the authors argue that this split attention can prevent either goal from succeeding. Their converge-then-diversify strategy first drives the search toward a single Pareto point, then spreads solutions across the front. Two implementations based on standard acquisition functions statistically outperformed state-of-the-art methods in 72.9% of 446 pairwise comparisons, tied in 21.1% and lost in 6.1%, with the largest gains in tight-budget and high-dimensional settings.

### Why it matters {#why-it-matters-mp-2026-09-15-004}

Across 446 pairwise comparisons, a two-stage Bayesian-optimization strategy won 72.9%, tied 21.1% and lost 6.1%.

### Limits and context {#limitations-mp-2026-09-15-004}

- No additional limitation was separately recorded.

### Claims and sources {#claims-mp-2026-09-15-004}

- Across 446 pairwise comparisons, a two-stage Bayesian-optimization strategy won 72.9%, tied 21.1% and lost 6.1%. [source-2026-09-15-004]

## 5. The Patent Judge Improved Drafts—and Disagreed by Metric {#mp-2026-09-15-005}

- Story ID: `mp-2026-09-15-005`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-005/the-patent-judge-improved-drafts-and-disagreed-by-metric

**Dek:** Iterative judge feedback helped a low-reasoning drafting agent approach a costlier agent, but attorney agreement varied sharply with the metric.

Vibe Patenting tests an agent that drafts patents and a separately invoked language-model judge that critiques each revision. Judge-guided iteration consistently raised judge-assessed quality, while unguided revision saturated; a lower-reasoning agent approached the score of a more expensive high-reasoning configuration. A professional patent attorney’s review found meaningful agreement with the automated judge, but calibration and agreement changed substantially by metric. The result supports structured critique as an optimization signal while warning that self-scoring is not a substitute for qualified legal judgment.

### Why it matters {#why-it-matters-mp-2026-09-15-005}

Iterative judge feedback helped a low-reasoning drafting agent approach a costlier agent, but attorney agreement varied sharply with the metric.

### Limits and context {#limitations-mp-2026-09-15-005}

- The result supports structured critique as an optimization signal while warning that self-scoring is not a substitute for qualified legal judgment.

### Claims and sources {#claims-mp-2026-09-15-005}

- Iterative judge feedback helped a low-reasoning drafting agent approach a costlier agent, but attorney agreement varied sharply with the metric. [source-2026-09-15-005] — Qualification: The result supports structured critique as an optimization signal while warning that self-scoring is not a substitute for qualified legal judgment.

## 6. The Farm Agent Adapted When the Weather Pattern Moved {#mp-2026-09-15-006}

- Story ID: `mp-2026-09-15-006`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-006/the-farm-agent-adapted-when-the-weather-pattern-moved

**Dek:** Zero-shot language-model agents matched reinforcement learning under familiar weather and adapted better after the environment shifted.

Long-horizon physical management requires an agent to observe, act and recover as conditions change. This study combines planning, tool use, observation and verification in a multi-agent framework and tests it on agricultural management tasks against reinforcement-learning agents. Under the same weather pattern, the zero-shot language-model agents achieved comparable management outcomes; when evaluated after a weather shift, they adapted more effectively than the trained RL policies. The result is a controlled feasibility study, not evidence that general language agents can safely manage farms without domain controls or human oversight.

### Why it matters {#why-it-matters-mp-2026-09-15-006}

Zero-shot language-model agents matched reinforcement learning under familiar weather and adapted better after the environment shifted.

### Limits and context {#limitations-mp-2026-09-15-006}

- The result is a controlled feasibility study, not evidence that general language agents can safely manage farms without domain controls or human oversight.

### Claims and sources {#claims-mp-2026-09-15-006}

- Zero-shot language-model agents matched reinforcement learning under familiar weather and adapted better after the environment shifted. [source-2026-09-15-006] — Qualification: The result is a controlled feasibility study, not evidence that general language agents can safely manage farms without domain controls or human oversight.

## 7. A Lab Agent Kept the Corrections With the Protocol {#mp-2026-09-15-007}

- Story ID: `mp-2026-09-15-007`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-007/a-lab-agent-kept-the-corrections-with-the-protocol

**Dek:** LabAgent recorded executable skills and corrective experience, reproduced a published figure and led tested generalist agents in four life-science domains.

When students leave a lab, tacit fixes and fragile procedures can disappear with them. LabAgent packages methods as executable, verifiable skills and records corrections so later attempts can avoid the same failure. The authors applied it to drug-property prediction, biomedical analysis, protein-variant effects and statistical genetics, reporting first place over tested commercial generalist agents in each domain and successful reproduction of a published figure. The evaluation supports institutional memory as an agent design goal, but broad scientific reliability still depends on transparent protocols and independent reproduction.

### Why it matters {#why-it-matters-mp-2026-09-15-007}

LabAgent recorded executable skills and corrective experience, reproduced a published figure and led tested generalist agents in four life-science domains.

### Limits and context {#limitations-mp-2026-09-15-007}

- No additional limitation was separately recorded.

### Claims and sources {#claims-mp-2026-09-15-007}

- LabAgent recorded executable skills and corrective experience, reproduced a published figure and led tested generalist agents in four life-science domains. [source-2026-09-15-007]

## 8. Synthetic Curves Taught the Model How Trends Combine {#mp-2026-09-15-008}

- Story ID: `mp-2026-09-15-008`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-008/synthetic-curves-taught-the-model-how-trends-combine

**Dek:** TimeThink trained only on generated time-series primitives and outperformed strong baselines on synthetic and real-world compositional questions.

Time-series language models can answer familiar questions while failing when trends, seasonality and other temporal primitives must be composed in a new way. TimeThink generates atomic and composite question-answer pairs with deterministic ground truth, then uses reinforcement learning with verifiable rewards to train explicit reasoning. The model was trained only on synthetic data yet outperformed strong baselines on both synthetic and real-world benchmarks in the authors’ experiments. The finding isolates a useful training mechanism; it does not establish clinical readiness for the high-stakes applications that motivate the work.

### Why it matters {#why-it-matters-mp-2026-09-15-008}

TimeThink trained only on generated time-series primitives and outperformed strong baselines on synthetic and real-world compositional questions.

### Limits and context {#limitations-mp-2026-09-15-008}

- The model was trained only on synthetic data yet outperformed strong baselines on both synthetic and real-world benchmarks in the authors’ experiments.
- The finding isolates a useful training mechanism; it does not establish clinical readiness for the high-stakes applications that motivate the work.

### Claims and sources {#claims-mp-2026-09-15-008}

- TimeThink trained only on generated time-series primitives and outperformed strong baselines on synthetic and real-world compositional questions. [source-2026-09-15-008] — Qualification: The model was trained only on synthetic data yet outperformed strong baselines on both synthetic and real-world benchmarks in the authors’ experiments.

## 9. The Web Agent Learned the Application Before Spending Tokens {#mp-2026-09-15-009}

- Story ID: `mp-2026-09-15-009`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-009/the-web-agent-learned-the-application-before-spending-tokens

**Dek:** On 45 Canvas tasks, OdoBot used 44% and 80% fewer tokens than two agent baselines while beating one on task success.

Conventional web agents repeatedly inspect an interface and reason about low-level interactions. OdoBot instead builds an application behavior model from successful demonstrations, then uses that reusable structure to execute natural-language tasks. On 45 tasks in the Canvas learning-management system, the authors report 44% fewer tokens than Agent-E and 80% fewer than WebVoyager, while also surpassing WebVoyager’s task-success rate. The evaluation is limited to one application and task set, but it suggests that explicit behavior models can replace a costly share of repeated visual reasoning.

### Why it matters {#why-it-matters-mp-2026-09-15-009}

On 45 Canvas tasks, OdoBot used 44% and 80% fewer tokens than two agent baselines while beating one on task success.

### Limits and context {#limitations-mp-2026-09-15-009}

- No additional limitation was separately recorded.

### Claims and sources {#claims-mp-2026-09-15-009}

- On 45 Canvas tasks, OdoBot used 44% and 80% fewer tokens than two agent baselines while beating one on task success. [source-2026-09-15-009]

## 10. The Molecule Builder Added New Fragments Without Retraining {#mp-2026-09-15-010}

- Story ID: `mp-2026-09-15-010`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-010/the-molecule-builder-added-new-fragments-without-retraining

**Dek:** Fraglingo jointly predicts fragment identity and attachment and generalized to inference libraries four times larger than its training vocabulary.

Fragment-based molecular generators often choose a fragment from a fixed vocabulary and predict its attachment separately. Fraglingo represents both decisions in one attachment-aware continuous embedding, retrieves the next fragment by nearest-neighbor search and encodes the growing molecule from its active attachment site. Because inference operates over embeddings rather than fixed identifiers, new fragments can be added without retraining. The authors report stronger joint property control than comparable baselines and generalization to fragment libraries four times larger than the training set. Computational benchmarks do not establish laboratory synthesis or drug efficacy.

### Why it matters {#why-it-matters-mp-2026-09-15-010}

Fraglingo jointly predicts fragment identity and attachment and generalized to inference libraries four times larger than its training vocabulary.

### Limits and context {#limitations-mp-2026-09-15-010}

- Computational benchmarks do not establish laboratory synthesis or drug efficacy.

### Claims and sources {#claims-mp-2026-09-15-010}

- Fraglingo jointly predicts fragment identity and attachment and generalized to inference libraries four times larger than its training vocabulary. [source-2026-09-15-010] — Qualification: Computational benchmarks do not establish laboratory synthesis or drug efficacy.

## 11. The Meta-Agent Cut 1,283 Browser Tools Down to 33 {#mp-2026-09-15-011}

- Story ID: `mp-2026-09-15-011`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-011/the-meta-agent-cut-1-283-browser-tools-down-to-33

**Dek:** AutoTailor’s final tool set reached 90.6% correctness with fallback while cutting request tokens 57.8% and latency 29.4%.

Tool-discovery systems can leave web agents with hundreds of redundant, badly sized APIs. AutoTailor converts successful trajectories into parameterized browser programs, filters them for granularity and likely use, then watches outcomes to add missing capabilities and prune unused ones. On 106 WebArena Postmill tasks, offline filtering reduced 1,283 candidate APIs to 87 and dynamic reselection produced 33. With ReAct fallback, the final set achieved 90.6% correctness versus 87.5% for ReAct alone while reducing request-token cost 57.8% and latency 29.4%.

### Why it matters {#why-it-matters-mp-2026-09-15-011}

AutoTailor’s final tool set reached 90.6% correctness with fallback while cutting request tokens 57.8% and latency 29.4%.

### Limits and context {#limitations-mp-2026-09-15-011}

- No additional limitation was separately recorded.

### Claims and sources {#claims-mp-2026-09-15-011}

- AutoTailor’s final tool set reached 90.6% correctness with fallback while cutting request tokens 57.8% and latency 29.4%. [source-2026-09-15-011]

## 12. The Passenger Saw What the Physical Robot Saw {#mp-2026-09-15-012}

- Story ID: `mp-2026-09-15-012`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-012/the-passenger-saw-what-the-physical-robot-saw

**Dek:** A live robot-to-VR link averaged 29.63-millisecond state updates and 2.28% route-progress error over 20 trials.

Passenger-facing autonomous-vehicle studies often use simulated motion or scripted scenes. This framework links a physical ROS 2 robot vehicle to a Unity application on a VR headset, transmitting state and onboard video separately so the virtual vehicle mirrors real motion and exposes navigation decisions. Across 20 closed-loop trials, the authors report 29.63 ms mean state-update latency, 2.28% mean relative route-progress error, 10.006 video frames per second and 0.25% frame loss, with all monitored decisions reflected correctly. The laboratory-scale setup demonstrates synchronization, not passenger trust or road deployment.

### Why it matters {#why-it-matters-mp-2026-09-15-012}

A live robot-to-VR link averaged 29.63-millisecond state updates and 2.28% route-progress error over 20 trials.

### Limits and context {#limitations-mp-2026-09-15-012}

- The laboratory-scale setup demonstrates synchronization, not passenger trust or road deployment.

### Claims and sources {#claims-mp-2026-09-15-012}

- A live robot-to-VR link averaged 29.63-millisecond state updates and 2.28% route-progress error over 20 trials. [source-2026-09-15-012] — Qualification: The laboratory-scale setup demonstrates synchronization, not passenger trust or road deployment.

## 13. Fifteen Minutes of Installer Help Taught the Robot the Tight Fit {#mp-2026-09-15-013}

- Story ID: `mp-2026-09-15-013`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-013/fifteen-minutes-of-installer-help-taught-the-robot-the-tight-fit

**Dek:** With two millimeters of clearance, the simulated assembly pipeline reached 100% seating after 12–15 minutes of takeover supervision.

Precision construction assembly depends on tacit installer knowledge at contact and failure boundaries. This framework starts with teleoperated demonstrations, then records sparse binary takeovers and acceptance-aligned terminal rewards while a temporally abstract policy learns recovery maneuvers. In stress tests with two millimeters of clearance, randomized pose and friction, the simulated system reached 100% autonomous seating after 12–15 minutes of cumulative installer supervision over three hours of online training; the 95% milestone arrived in about 0.5 and 1.5 hours across two experiments. Physical-site validation remains open.

### Why it matters {#why-it-matters-mp-2026-09-15-013}

With two millimeters of clearance, the simulated assembly pipeline reached 100% seating after 12–15 minutes of takeover supervision.

### Limits and context {#limitations-mp-2026-09-15-013}

- No additional limitation was separately recorded.

### Claims and sources {#claims-mp-2026-09-15-013}

- With two millimeters of clearance, the simulated assembly pipeline reached 100% seating after 12–15 minutes of takeover supervision. [source-2026-09-15-013]

## 14. A 512-Byte Message Beat the Reconstructed Grasp Geometry {#mp-2026-09-15-014}

- Story ID: `mp-2026-09-15-014`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-014/a-512-byte-message-beat-the-reconstructed-grasp-geometry

**Dek:** The outcome-centered representation was 288 times smaller than one RGB-D frame and reached 0.876 AUC on 11,979 simulated grasps.

Networked robot arms often transmit dense geometry even when the action only needs a compact prediction of outcomes. This work learns an action-conditioned stochastic bottleneck that preserves outcome distributions rather than reconstructing the whole scene. Across 11,979 simulated grasps on 13 objects, its representation reached 0.876 AUC for lift success versus 0.542 for a reconstructed-geometry wrench score. The 512-byte interface was 288× smaller than one RGB-D frame and ran in 16 ms per CPU decision. Performance on unseen objects weakened before a feedback update, underscoring the remaining generalization gap.

### Why it matters {#why-it-matters-mp-2026-09-15-014}

The outcome-centered representation was 288 times smaller than one RGB-D frame and reached 0.876 AUC on 11,979 simulated grasps.

### Limits and context {#limitations-mp-2026-09-15-014}

- Networked robot arms often transmit dense geometry even when the action only needs a compact prediction of outcomes.

### Claims and sources {#claims-mp-2026-09-15-014}

- The outcome-centered representation was 288 times smaller than one RGB-D frame and reached 0.876 AUC on 11,979 simulated grasps. [source-2026-09-15-014] — Qualification: Networked robot arms often transmit dense geometry even when the action only needs a compact prediction of outcomes.

## 15. The Simulator Stepped Without Middleware—and Replayed the Same Run {#mp-2026-09-15-026}

- Story ID: `mp-2026-09-15-026`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-026/the-simulator-stepped-without-middleware-and-replayed-the-same-run

**Dek:** GzDRL synchronized actions directly with Gazebo physics, led tested workstation throughput and transferred a policy to a quadrotor without fine-tuning.

Middleware can make reinforcement-learning experiments in Gazebo nondeterministic and difficult to reproduce. GzDRL moves environment stepping into one process, directly synchronizing agent actions with physics updates to support vectorized, repeatable data collection. The authors report the highest workstation throughput among evaluated frameworks, competitive performance with GPU-accelerated simulators on laptop hardware, multi-agent scaling and reproducible runs. A learned policy was also deployed on a physical quadrotor without fine-tuning. The transfer is one validation case, not a general sim-to-real guarantee.

### Why it matters {#why-it-matters-mp-2026-09-15-026}

GzDRL synchronized actions directly with Gazebo physics, led tested workstation throughput and transferred a policy to a quadrotor without fine-tuning.

### Limits and context {#limitations-mp-2026-09-15-026}

- The transfer is one validation case, not a general sim-to-real guarantee.

### Claims and sources {#claims-mp-2026-09-15-026}

- GzDRL synchronized actions directly with Gazebo physics, led tested workstation throughput and transferred a policy to a quadrotor without fine-tuning. [source-2026-09-15-015] — Qualification: The transfer is one validation case, not a general sim-to-real guarantee.

## 16. Two Black-Hole Binaries May Rise Above the Background {#mp-2026-09-15-027}

- Story ID: `mp-2026-09-15-027`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-027/two-black-hole-binaries-may-rise-above-the-background

**Dek:** Population fits identify 3C 66B and Mrk 501 as promising individual gravitational-wave targets and forecast tens of SKAO detections.

A population model fitted to NANOGrav’s 15-year data asks whether electromagnetically identified supermassive-black-hole binaries occupy the rare loud tail of the gravitational-wave distribution. The analysis finds that 3C 66B and Mrk 501 are likely to stand out from the unresolved background, with 3C 66B robustly detectable in the authors’ Square Kilometre Array Observatory forecast. The same models predict tens of individually resolvable binaries and suggest their abundance could distinguish environmental hardening from purely gravitational-wave-driven evolution. These are model-based forecasts, not detections.

### Why it matters {#why-it-matters-mp-2026-09-15-027}

Population fits identify 3C 66B and Mrk 501 as promising individual gravitational-wave targets and forecast tens of SKAO detections.

### Limits and context {#limitations-mp-2026-09-15-027}

- These are model-based forecasts, not detections.

### Claims and sources {#claims-mp-2026-09-15-027}

- Population fits identify 3C 66B and Mrk 501 as promising individual gravitational-wave targets and forecast tens of SKAO detections. [source-2026-09-15-016] — Qualification: These are model-based forecasts, not detections.

## 17. Positive Wigner Functions Fell Below the Vacuum Entropy {#mp-2026-09-15-015}

- Story ID: `mp-2026-09-15-015`
- Type: `ticker`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-015/positive-wigner-functions-fell-below-the-vacuum-entropy

**Dek:** Explicit finite-energy states overturn a conjectured entropy floor while recovering the vacuum limit as energy vanishes.

The paper constructs physical quantum states whose Wigner functions remain nonnegative even though their Shannon entropy falls below the vacuum value. It supplies an explicit finite-energy counterexample and analytic finite-Fock-support families, then derives how the optimal entropy deficit closes as mean photon number approaches zero. The result is a mathematical counterexample and asymptotic analysis, not an experimental preparation report.

### Why it matters {#why-it-matters-mp-2026-09-15-015}

Explicit finite-energy states overturn a conjectured entropy floor while recovering the vacuum limit as energy vanishes.

### Limits and context {#limitations-mp-2026-09-15-015}

- The paper constructs physical quantum states whose Wigner functions remain nonnegative even though their Shannon entropy falls below the vacuum value.
- The result is a mathematical counterexample and asymptotic analysis, not an experimental preparation report.

### Claims and sources {#claims-mp-2026-09-15-015}

- Explicit finite-energy states overturn a conjectured entropy floor while recovering the vacuum limit as energy vanishes. [source-2026-09-15-017] — Qualification: The paper constructs physical quantum states whose Wigner functions remain nonnegative even though their Shannon entropy falls below the vacuum value.

## 18. Four Photons Shared 750 Secret Bits Each Second {#mp-2026-09-15-016}

- Story ID: `mp-2026-09-15-016`
- Type: `ticker`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-016/four-photons-shared-750-secret-bits-each-second

**Dek:** A four-arm, 20-kilometer fiber network supported composable protection against outside and participant attacks.

Two engineered photon-pair sources generated four-qubit GHZ states at more than 5,000 fourfold events per second. Distributed across a four-arm star network totaling 20 km of fiber, the setup reached a maximum asymptotic secret-key rate of 750 ± 10 bits per second. From measured rates and errors, the authors infer an 8.7-megabit finite-key lower bound for an optimized 24-hour run if device statistics remain stationary.

### Why it matters {#why-it-matters-mp-2026-09-15-016}

A four-arm, 20-kilometer fiber network supported composable protection against outside and participant attacks.

### Limits and context {#limitations-mp-2026-09-15-016}

- From measured rates and errors, the authors infer an 8.7-megabit finite-key lower bound for an optimized 24-hour run if device statistics remain stationary.

### Claims and sources {#claims-mp-2026-09-15-016}

- A four-arm, 20-kilometer fiber network supported composable protection against outside and participant attacks. [source-2026-09-15-018] — Qualification: From measured rates and errors, the authors infer an 8.7-megabit finite-key lower bound for an optimized 24-hour run if device statistics remain stationary.

## 19. The Optical Compiler Routed Around Its Own Imperfections {#mp-2026-09-15-017}

- Story ID: `mp-2026-09-15-017`
- Type: `ticker`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-017/the-optical-compiler-routed-around-its-own-imperfections

**Dek:** Hardware-aware subcircuit placement raised coincidence rates by as much as 33.2% across the reported benchmarks.

Linear-optical quantum circuits lose accuracy and photons when phase noise and fabrication imperfections vary across the chip. This compiler incorporates a device’s nonuniform imperfections when mapping subcircuits, selecting placements that reduce their impact rather than assuming uniform hardware. Across the authors’ benchmark suite, the method improved phase-noise robustness and increased coincidence rates by as much as 33.2%. The evaluation demonstrates compilation gains on tested models and devices, not fault-tolerant operation.

### Why it matters {#why-it-matters-mp-2026-09-15-017}

Hardware-aware subcircuit placement raised coincidence rates by as much as 33.2% across the reported benchmarks.

### Limits and context {#limitations-mp-2026-09-15-017}

- The evaluation demonstrates compilation gains on tested models and devices, not fault-tolerant operation.

### Claims and sources {#claims-mp-2026-09-15-017}

- Hardware-aware subcircuit placement raised coincidence rates by as much as 33.2% across the reported benchmarks. [source-2026-09-15-019] — Qualification: The evaluation demonstrates compilation gains on tested models and devices, not fault-tolerant operation.

## 20. Elliptical Galaxies Lost Most of the Gas Below 300,000 Kelvin {#mp-2026-09-15-018}

- Story ID: `mp-2026-09-15-018`
- Type: `ticker`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-018/elliptical-galaxies-lost-most-of-the-gas-below-300-000-kelvin

**Dek:** Reprocessed FUSE spectra point to mass-dependent hot-gas cooling and roughly tenfold suppression before the molecular phase.

A modern reanalysis of archival FUSE spectra measures or limits O VI emission in 29 massive elliptical galaxies and uses it to infer gas cooling through about 300,000 kelvin. The reported comparison suggests cooling from hotter circumgalactic gas is suppressed four to five times more strongly at cluster scale than around groups and ellipticals. Cooling onward to molecular gas appears suppressed by about an order of magnitude across masses, implying additional mixing, conduction or recycling physics.

### Why it matters {#why-it-matters-mp-2026-09-15-018}

Reprocessed FUSE spectra point to mass-dependent hot-gas cooling and roughly tenfold suppression before the molecular phase.

### Limits and context {#limitations-mp-2026-09-15-018}

- No additional limitation was separately recorded.

### Claims and sources {#claims-mp-2026-09-15-018}

- Reprocessed FUSE spectra point to mass-dependent hot-gas cooling and roughly tenfold suppression before the molecular phase. [source-2026-09-15-020]

## 21. Extended Dark Objects Bent the Wave Differently {#mp-2026-09-15-019}

- Story ID: `mp-2026-09-15-019`
- Type: `ticker`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-019/extended-dark-objects-bent-the-wave-differently

**Dek:** Finite lens size changes gravitational-wave diffraction and current data probe model space from 100 to 100,000 solar masses.

Gravitational-wave lensing searches often approximate a compact object as a point. This study calculates the frequency-dependent waveform effects of extended dark lenses such as solitons and dark stars, finding that finite size matters once the lens radius exceeds its Einstein scale. Current gravitational-wave data provide complementary constraints for lens masses from roughly 100 to 100,000 solar masses, while future detectors extend the reach. The work sets model-dependent bounds; it does not report a dark-object detection.

### Why it matters {#why-it-matters-mp-2026-09-15-019}

Finite lens size changes gravitational-wave diffraction and current data probe model space from 100 to 100,000 solar masses.

### Limits and context {#limitations-mp-2026-09-15-019}

- The work sets model-dependent bounds; it does not report a dark-object detection.

### Claims and sources {#claims-mp-2026-09-15-019}

- Finite lens size changes gravitational-wave diffraction and current data probe model space from 100 to 100,000 solar masses. [source-2026-09-15-021] — Qualification: The work sets model-dependent bounds; it does not report a dark-object detection.

## 22. SATURNIX {#mp-2026-09-15-020}

- Story ID: `mp-2026-09-15-020`
- Type: `invention_desk`
- Classification: `editorial`
- Content status: `carried_over`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-020/saturnix

**Dek:** Builds a tactile digital camera around a Raspberry Pi Zero 2 W, an autofocus sensor, a small viewfinder, and mechanical-switch controls while publishing the software and printable hardware files.

Builds a tactile digital camera around a Raspberry Pi Zero 2 W, an autofocus sensor, a small viewfinder, and mechanical-switch controls while publishing the software and printable hardware files.

### Why it matters {#why-it-matters-mp-2026-09-15-020}

An independent builder is turning an improbable idea into a working project.

### Limits and context {#limitations-mp-2026-09-15-020}

- A Desk Pick is an editorial selection, not a product endorsement.

### Claims and sources {#claims-mp-2026-09-15-020}

- This Invention Desk entry makes no independently sourced news claim.

## 23. WURB-2026 {#mp-2026-09-15-021}

- Story ID: `mp-2026-09-15-021`
- Type: `invention_desk`
- Classification: `editorial`
- Content status: `carried_over`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-021/wurb-2026

**Dek:** Combines a Raspberry Pi-class computer, an ultrasonic microphone, local storage, and a web interface into a modular recorder for active or unattended bat monitoring.

Combines a Raspberry Pi-class computer, an ultrasonic microphone, local storage, and a web interface into a modular recorder for active or unattended bat monitoring.

### Why it matters {#why-it-matters-mp-2026-09-15-021}

An independent builder is turning an improbable idea into a working project.

### Limits and context {#limitations-mp-2026-09-15-021}

- A Desk Pick is an editorial selection, not a product endorsement.

### Claims and sources {#claims-mp-2026-09-15-021}

- This Invention Desk entry makes no independently sourced news claim.

## 24. OpenCTD {#mp-2026-09-15-022}

- Story ID: `mp-2026-09-15-022`
- Type: `invention_desk`
- Classification: `editorial`
- Content status: `carried_over`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-022/openctd

**Dek:** Packages conductivity, temperature, and depth sensors with an Arduino-compatible controller, battery, and SD storage inside a user-built housing for nearshore research and education.

Packages conductivity, temperature, and depth sensors with an Arduino-compatible controller, battery, and SD storage inside a user-built housing for nearshore research and education.

### Why it matters {#why-it-matters-mp-2026-09-15-022}

An independent builder is turning an improbable idea into a working project.

### Limits and context {#limitations-mp-2026-09-15-022}

- A Desk Pick is an editorial selection, not a product endorsement.

### Claims and sources {#claims-mp-2026-09-15-022}

- This Invention Desk entry makes no independently sourced news claim.

## 25. Microban {#mp-2026-09-15-023}

- Story ID: `mp-2026-09-15-023`
- Type: `invention_desk`
- Classification: `editorial`
- Content status: `carried_over`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-023/microban

**Dek:** Turns printable body parts, nineteen servomotors, a Raspberry Pi Zero 2 W, batteries, and shared control software into a compact humanoid platform for hands-on robotics work.

Turns printable body parts, nineteen servomotors, a Raspberry Pi Zero 2 W, batteries, and shared control software into a compact humanoid platform for hands-on robotics work.

### Why it matters {#why-it-matters-mp-2026-09-15-023}

An independent builder is turning an improbable idea into a working project.

### Limits and context {#limitations-mp-2026-09-15-023}

- A Desk Pick is an editorial selection, not a product endorsement.

### Claims and sources {#claims-mp-2026-09-15-023}

- This Invention Desk entry makes no independently sourced news claim.

## 26. The First Paid Slot {#mp-2026-09-15-024}

- Story ID: `mp-2026-09-15-024`
- Type: `invention_desk`
- Classification: `house_example`
- Content status: `carried_over`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-024/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-2026-09-15-024}

This placement explains how builders can appear in The Invention Desk without purchasing editorial endorsement.

### Limits and context {#limitations-mp-2026-09-15-024}

- House example - no advertiser paid. Payment will buy placement, never endorsement.

### Claims and sources {#claims-mp-2026-09-15-024}

- This Invention Desk entry makes no independently sourced news claim.

## 27. Put Your Project on the Desk {#mp-2026-09-15-025}

- Story ID: `mp-2026-09-15-025`
- Type: `invention_desk`
- Classification: `house_example`
- Content status: `carried_over`
- Permanent URL: https://themachinepress.com/story/mp-2026-09-15-025/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-2026-09-15-025}

This placement explains how builders can appear in The Invention Desk without purchasing editorial endorsement.

### Limits and context {#limitations-mp-2026-09-15-025}

- Manual intake only. Payment buys placement, never endorsement, and every submission is reviewed.

### Claims and sources {#claims-mp-2026-09-15-025}

- This Invention Desk entry makes no independently sourced news claim.

## Normalized sources

- **source-2026-09-15-001:** [arXiv preprint 2609.13463](https://arxiv.org/abs/2609.13463) — arXiv; primary_research
- **source-2026-09-15-002:** [arXiv preprint 2609.13231](https://arxiv.org/abs/2609.13231) — arXiv; primary_research
- **source-2026-09-15-003:** [arXiv preprint 2609.13356](https://arxiv.org/abs/2609.13356) — arXiv; primary_research
- **source-2026-09-15-004:** [arXiv preprint 2609.13396](https://arxiv.org/abs/2609.13396) — arXiv; primary_research
- **source-2026-09-15-005:** [arXiv preprint 2609.13422](https://arxiv.org/abs/2609.13422) — arXiv; primary_research
- **source-2026-09-15-006:** [arXiv preprint 2609.13436](https://arxiv.org/abs/2609.13436) — arXiv; primary_research
- **source-2026-09-15-007:** [arXiv preprint 2609.13437](https://arxiv.org/abs/2609.13437) — arXiv; primary_research
- **source-2026-09-15-008:** [arXiv preprint 2609.13457](https://arxiv.org/abs/2609.13457) — arXiv; primary_research
- **source-2026-09-15-009:** [arXiv preprint 2609.13491](https://arxiv.org/abs/2609.13491) — arXiv; primary_research
- **source-2026-09-15-010:** [arXiv preprint 2609.13519](https://arxiv.org/abs/2609.13519) — arXiv; primary_research
- **source-2026-09-15-011:** [arXiv preprint 2609.13548](https://arxiv.org/abs/2609.13548) — arXiv; primary_research
- **source-2026-09-15-012:** [arXiv preprint 2609.13224](https://arxiv.org/abs/2609.13224) — arXiv; primary_research
- **source-2026-09-15-013:** [arXiv preprint 2609.13234](https://arxiv.org/abs/2609.13234) — arXiv; primary_research
- **source-2026-09-15-014:** [arXiv preprint 2609.13235](https://arxiv.org/abs/2609.13235) — arXiv; primary_research
- **source-2026-09-15-015:** [arXiv preprint 2609.13243](https://arxiv.org/abs/2609.13243) — arXiv; primary_research
- **source-2026-09-15-016:** [arXiv preprint 2609.13363](https://arxiv.org/abs/2609.13363) — arXiv; primary_research
- **source-2026-09-15-017:** [arXiv preprint 2609.13312](https://arxiv.org/abs/2609.13312) — arXiv; primary_research
- **source-2026-09-15-018:** [arXiv preprint 2609.13367](https://arxiv.org/abs/2609.13367) — arXiv; primary_research
- **source-2026-09-15-019:** [arXiv preprint 2609.13379](https://arxiv.org/abs/2609.13379) — arXiv; primary_research
- **source-2026-09-15-020:** [arXiv preprint 2609.13368](https://arxiv.org/abs/2609.13368) — arXiv; primary_research
- **source-2026-09-15-021:** [arXiv preprint 2609.13369](https://arxiv.org/abs/2609.13369) — arXiv; primary_research