---
schema_version: "1.0.0"
edition_id: "mp-2026-10-02-morning-0085"
published_at: "2026-10-02T09:00:00.000-04:00"
modified_at: "2026-10-02T09:00:00.000-04:00"
canonical_url: "https://themachinepress.com/edition/2026-10-02"
story_count: 27
lead_story_id: "mp-2026-10-02-001"
---

# The Machine Press — Morning edition

Edition ID: `mp-2026-10-02-morning-0085`  
Published: 2026-10-02T09:00:00.000-04:00  
Canonical edition: https://themachinepress.com/edition/2026-10-02

Causal Memory Policy intervened on retrieval itself after store-level tests left memory utility unidentified.

## 1. Half the Needed Memories Were Never Given a Chance {#mp-2026-10-02-001}

- Story ID: `mp-2026-10-02-001`
- Type: `lead`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-001/half-the-needed-memories-were-never-given-a-chance

**Dek:** Causal Memory Policy intervened on retrieval itself after store-level tests left memory utility unidentified.

Memory systems often estimate whether a stored item matters only after the retriever surfaces it. Causal Memory Policy shows that this creates a positivity failure: store-level interventions cannot identify the value of a memory that is never retrieved. In the authors' tests, utility was unidentified for 54 percent of required LongMemEval memories and 67 percent on LoCoMo, including in a deployed system. Reserving context slots for memories sampled with known propensities raised required-versus-nonrequired discrimination from 0.54 to 0.66 AUC. Even then, utility estimated for one query did not predict value on unseen queries, limiting its use for irreversible retention decisions.

### Why it matters {#why-it-matters-mp-2026-10-02-001}

Causal Memory Policy intervened on retrieval itself after store-level tests left memory utility unidentified.

### Limits and context {#limitations-mp-2026-10-02-001}

- Memory systems often estimate whether a stored item matters only after the retriever surfaces it.
- Causal Memory Policy shows that this creates a positivity failure: store-level interventions cannot identify the value of a memory that is never retrieved.
- Even then, utility estimated for one query did not predict value on unseen queries, limiting its use for irreversible retention decisions.

### Claims and sources {#claims-mp-2026-10-02-001}

- Causal Memory Policy intervened on retrieval itself after store-level tests left memory utility unidentified. [source-2026-10-02-001] — Qualification: Memory systems often estimate whether a stored item matters only after the retriever surfaces it.

## 2. A Smaller Model Saw the Hallucination First {#mp-2026-10-02-002}

- Story ID: `mp-2026-10-02-002`
- Type: `secondary`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-002/a-smaller-model-saw-the-hallucination-first

**Dek:** Cross-model hidden-state probes matched or beat generators at locating the onset of semantic drift.

Rather than label each output token independently, the study probes internal activations to locate the onset and continuation span of a hallucination. The authors report substantial precision-recall AUC improvements over random baselines despite extreme class imbalance. Their cross-model setup lets one model inspect hidden representations elicited by another model's generation; an external observer matched or exceeded the generator's own onset detection, even when the observer was smaller. The result is a preprint evaluation of the tested model pairs and probes, not evidence that smaller models generally know truth better or that access to internal states is available across ordinary hosted systems.

### Why it matters {#why-it-matters-mp-2026-10-02-002}

Cross-model hidden-state probes matched or beat generators at locating the onset of semantic drift.

### Limits and context {#limitations-mp-2026-10-02-002}

- The result is a preprint evaluation of the tested model pairs and probes, not evidence that smaller models generally know truth better or that access to internal states is available across ordinary hosted systems.

### Claims and sources {#claims-mp-2026-10-02-002}

- Cross-model hidden-state probes matched or beat generators at locating the onset of semantic drift. [source-2026-10-02-002] — Qualification: The result is a preprint evaluation of the tested model pairs and probes, not evidence that smaller models generally know truth better or that access to internal states is available across ordinary hosted systems.

## 3. Fifteen Practice Rounds Took Robot Success to 95 Percent {#mp-2026-10-02-003}

- Story ID: `mp-2026-10-02-003`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-003/fifteen-practice-rounds-took-robot-success-to-95-percent

**Dek:** RPG revised symbolic skills and prompts from simulated execution feedback without updating model weights.

Reconstruct, Practice, Go Real identifies capabilities in offline robot data, builds related simulation tasks and uses execution feedback, privileged simulator state and videos to diagnose failures. Candidate skill and prompt revisions are tested across tasks before retention. On 22 manipulation tasks, reported success rose from 28.6 percent after the first practice round to 95 percent after 15, above the tested baselines; after calibration and hardware adaptation, the frozen system completed all 30 physical trials across three tasks. These are source-reported results on a limited task and hardware set.

### Why it matters {#why-it-matters-mp-2026-10-02-003}

RPG revised symbolic skills and prompts from simulated execution feedback without updating model weights.

### Limits and context {#limitations-mp-2026-10-02-003}

- No additional limitation was separately recorded.

### Claims and sources {#claims-mp-2026-10-02-003}

- RPG revised symbolic skills and prompts from simulated execution feedback without updating model weights. [source-2026-10-02-003]

## 4. Agentic Search Missed the Papers Scientists Said Inspired Them {#mp-2026-10-02-004}

- Story ID: `mp-2026-10-02-004`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-004/agentic-search-missed-the-papers-scientists-said-inspired-them

**Dek:** Embedding retrieval reached 0.48 Recall@20; an agent calling that retriever scored 0.42.

ScholarCatalyst asked 184 lead authors of 207 computer-science papers which earlier work did or could have advanced their projects, with rationales and a time-bounded literature pool. Agentic search reached 0.42 Recall@20, below embedding retrieval at 0.48, while the strongest tested agent reached 0.51 despite possible training exposure to completed papers. The benchmark turns scientific inspiration into an author-judged retrieval task; it does not establish that the labeled papers were the only valid catalysts.

### Why it matters {#why-it-matters-mp-2026-10-02-004}

Embedding retrieval reached 0.48 Recall@20; an agent calling that retriever scored 0.42.

### Limits and context {#limitations-mp-2026-10-02-004}

- The benchmark turns scientific inspiration into an author-judged retrieval task; it does not establish that the labeled papers were the only valid catalysts.

### Claims and sources {#claims-mp-2026-10-02-004}

- Embedding retrieval reached 0.48 Recall@20; an agent calling that retriever scored 0.42. [source-2026-10-02-004] — Qualification: The benchmark turns scientific inspiration into an author-judged retrieval task; it does not establish that the labeled papers were the only valid catalysts.

## 5. No Open Model Cleared 42 Percent on Unhinted Security Commands {#mp-2026-10-02-005}

- Story ID: `mp-2026-10-02-005`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-005/no-open-model-cleared-42-percent-on-unhinted-security-commands

**Dek:** KaliBench tests 8,504 intent-to-command pairs across 1,642 Kali Linux tools.

KaliBench measures whether models can translate analyst intent into executable command lines, with deterministic canonicalization, alias-aware scoring, sandbox execution and human refinement. Across 24 configurations, no open-weight model exceeded 42 percent exact-command accuracy without tool hints. Fine-tuning and verifiable-reward training substantially improved an eight-billion-parameter model to performance comparable with a tested 685-billion-parameter mixture-of-experts model. Exact-command accuracy is a benchmark metric, not authorization to run generated security commands unsupervised.

### Why it matters {#why-it-matters-mp-2026-10-02-005}

KaliBench tests 8,504 intent-to-command pairs across 1,642 Kali Linux tools.

### Limits and context {#limitations-mp-2026-10-02-005}

- Exact-command accuracy is a benchmark metric, not authorization to run generated security commands unsupervised.

### Claims and sources {#claims-mp-2026-10-02-005}

- KaliBench tests 8,504 intent-to-command pairs across 1,642 Kali Linux tools. [source-2026-10-02-005] — Qualification: Exact-command accuracy is a benchmark metric, not authorization to run generated security commands unsupervised.

## 6. A Visual Harness Reached Perfect Action Efficiency {#mp-2026-10-02-006}

- Story ID: `mp-2026-10-02-006`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-006/a-visual-harness-reached-perfect-action-efficiency

**Dek:** Lossless observation memory let one multimodal model finish all 25 public ARC-AGI-3 games.

VISTA preserves past visual observations in original form and lets a multimodal model retrieve and reorganize them during long-horizon reasoning. On ARC-AGI-3, the authors report that it raised Claude Opus 5.0's Relative Human Action Efficiency from 40.68 to 100, completing all 25 public games with 57.4 percent fewer actions than first-time human participants. It also improved results on three other visual-game and puzzle benchmarks. The score depends on the reported harness, model and public game set and is not a general measure of visual intelligence.

### Why it matters {#why-it-matters-mp-2026-10-02-006}

Lossless observation memory let one multimodal model finish all 25 public ARC-AGI-3 games.

### Limits and context {#limitations-mp-2026-10-02-006}

- The score depends on the reported harness, model and public game set and is not a general measure of visual intelligence.

### Claims and sources {#claims-mp-2026-10-02-006}

- Lossless observation memory let one multimodal model finish all 25 public ARC-AGI-3 games. [source-2026-10-02-006] — Qualification: The score depends on the reported harness, model and public game set and is not a general measure of visual intelligence.

## 7. Avatar Animation Lost Three Orders of CPU Cost {#mp-2026-10-02-007}

- Story ID: `mp-2026-10-02-007`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-007/avatar-animation-lost-three-orders-of-cpu-cost

**Dek:** A shared linear blendshape basis replaced heavy per-frame neural decoding across three avatar systems.

GALA distills pretrained Gaussian-avatar animation into a shallow coefficient predictor and a linear blend of identity-independent basis shapes. Block-local PCA under a rendering-aware metric controls fidelity and memory. Across three facial and full-body avatar systems, the authors report up to three orders of magnitude lower CPU animation cost while preserving most rendering quality and reaching as much as 60 frames per second on mobile devices. The claim applies to the evaluated models and held-out identities, not every avatar architecture.

### Why it matters {#why-it-matters-mp-2026-10-02-007}

A shared linear blendshape basis replaced heavy per-frame neural decoding across three avatar systems.

### Limits and context {#limitations-mp-2026-10-02-007}

- The claim applies to the evaluated models and held-out identities, not every avatar architecture.

### Claims and sources {#claims-mp-2026-10-02-007}

- A shared linear blendshape basis replaced heavy per-frame neural decoding across three avatar systems. [source-2026-10-02-007] — Qualification: The claim applies to the evaluated models and held-out identities, not every avatar architecture.

## 8. Three-Dimensional Generation Used 70 Percent Fewer Tokens {#mp-2026-10-02-008}

- Story ID: `mp-2026-10-02-008`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-008/three-dimensional-generation-used-70-percent-fewer-tokens

**Dek:** Sliding-window slice latents preserved cross-sections and long-range topology in one stage.

SILSA represents shapes with overlapping slice latents along three axes rather than fragmented voxel tokens. A sparse decoder and shared volumetric anchor lattice coordinate the slice streams, while persistence-diagram supervision tracks topology between neighbors. Against the strongest reported baseline, SILSA improved PSNR by 8.7 percent, coverage by 5.96 points and Betti error by 9.2 percent while using 70 percent fewer tokens than the next-most compact baseline. These are benchmarked generation results, not guarantees for arbitrary geometry.

### Why it matters {#why-it-matters-mp-2026-10-02-008}

Sliding-window slice latents preserved cross-sections and long-range topology in one stage.

### Limits and context {#limitations-mp-2026-10-02-008}

- These are benchmarked generation results, not guarantees for arbitrary geometry.

### Claims and sources {#claims-mp-2026-10-02-008}

- Sliding-window slice latents preserved cross-sections and long-range topology in one stage. [source-2026-10-02-008] — Qualification: These are benchmarked generation results, not guarantees for arbitrary geometry.

## 9. The Best Data Agent Excelled on Only a Third of Tasks {#mp-2026-10-02-009}

- Story ID: `mp-2026-10-02-009`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-009/the-best-data-agent-excelled-on-only-a-third-of-tasks

**Dek:** Argo-Bench hides ground truth behind 235 tables and grades decisions by simulated consequences.

Argo-Bench simulates a 2024 New York food-delivery platform at scale, exports it as a 235-table, 7.5-billion-row ERP warehouse and asks agents to reconstruct facts before actions such as fraud bans, incentives or back pay. Each of its 210 tasks has an executable reference solution. The strongest of 14 tested frontier and open-weight models scored at least 95 on only 34.8 percent of tasks and averaged 59.5 points. The warehouse is synthetic but designed from public data, literature and filings.

### Why it matters {#why-it-matters-mp-2026-10-02-009}

Argo-Bench hides ground truth behind 235 tables and grades decisions by simulated consequences.

### Limits and context {#limitations-mp-2026-10-02-009}

- The strongest of 14 tested frontier and open-weight models scored at least 95 on only 34.8 percent of tasks and averaged 59.5 points.

### Claims and sources {#claims-mp-2026-10-02-009}

- Argo-Bench hides ground truth behind 235 tables and grades decisions by simulated consequences. [source-2026-10-02-009] — Qualification: The strongest of 14 tested frontier and open-weight models scored at least 95 on only 34.8 percent of tasks and averaged 59.5 points.

## 10. Synthetic Spatial Hints Transferred to Real Images {#mp-2026-10-02-010}

- Story ID: `mp-2026-10-02-010`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-010/synthetic-spatial-hints-transferred-to-real-images

**Dek:** Privileged coordinates used only in training raised six real-world vision benchmarks by 3.23 points on average.

Where-OPD gives a teacher textual spatial guidance from procedurally generated scenes while the student sees only the image and question. The teacher locates and combines evidence across relevant regions; the student learns that behavior without human grounding labels or an external teacher model. Training only on synthetic scenes improved counting, document and chart tasks and produced a 3.23-point average gain across six real-world perception benchmarks. The transfer is measured on the reported models and tests.

### Why it matters {#why-it-matters-mp-2026-10-02-010}

Privileged coordinates used only in training raised six real-world vision benchmarks by 3.23 points on average.

### Limits and context {#limitations-mp-2026-10-02-010}

- Where-OPD gives a teacher textual spatial guidance from procedurally generated scenes while the student sees only the image and question.
- Training only on synthetic scenes improved counting, document and chart tasks and produced a 3.23-point average gain across six real-world perception benchmarks.

### Claims and sources {#claims-mp-2026-10-02-010}

- Privileged coordinates used only in training raised six real-world vision benchmarks by 3.23 points on average. [source-2026-10-02-010] — Qualification: Where-OPD gives a teacher textual spatial guidance from procedurally generated scenes while the student sees only the image and question.

## 11. Molecules Became Valid by Grammar Construction {#mp-2026-10-02-011}

- Story ID: `mp-2026-10-02-011`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-011/molecules-became-valid-by-grammar-construction

**Dek:** Higher-order production rules captured rings while remaining compatible with ordinary sequence models.

Higher-order Grammar Representation lifts molecules into combinatorial complexes and serializes them as context-free production rules. On the new ring-enriched RingDiv benchmark and five generation tests, the authors report 100 percent validity by construction and leading distributional alignment; its foundation model also led mean AUC across seven MoleculeNet benchmarks under probing and full fine-tuning. The representation's advantage is benchmarked against the selected baselines and does not itself establish chemical usefulness or synthesizability.

### Why it matters {#why-it-matters-mp-2026-10-02-011}

Higher-order production rules captured rings while remaining compatible with ordinary sequence models.

### Limits and context {#limitations-mp-2026-10-02-011}

- The representation's advantage is benchmarked against the selected baselines and does not itself establish chemical usefulness or synthesizability.

### Claims and sources {#claims-mp-2026-10-02-011}

- Higher-order production rules captured rings while remaining compatible with ordinary sequence models. [source-2026-10-02-011] — Qualification: The representation's advantage is benchmarked against the selected baselines and does not itself establish chemical usefulness or synthesizability.

## 12. Video Motion Controls Moved Into Three Dimensions {#mp-2026-10-02-012}

- Story ID: `mp-2026-10-02-012`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-012/video-motion-controls-moved-into-three-dimensions

**Dek:** Artists compose camera paths and local object handles in a shared scene scaffold.

Generative Cinematographer lifts one image into an editable 3D scaffold, where camera paths and local handles describe foreground motion relative to the scene. Guidance maps project those controls into a pretrained video model while retaining consistent handle identities and world coordinates as the camera moves. The authors train lightweight guidance and LoRA components and report more consistent camera-relative motion and geometry across diverse scenes. The system approximates non-rigid motion without a physics simulator and is an authoring tool, not a physical-motion model.

### Why it matters {#why-it-matters-mp-2026-10-02-012}

Artists compose camera paths and local object handles in a shared scene scaffold.

### Limits and context {#limitations-mp-2026-10-02-012}

- The system approximates non-rigid motion without a physics simulator and is an authoring tool, not a physical-motion model.

### Claims and sources {#claims-mp-2026-10-02-012}

- Artists compose camera paths and local object handles in a shared scene scaffold. [source-2026-10-02-012] — Qualification: The system approximates non-rigid motion without a physics simulator and is an authoring tool, not a physical-motion model.

## 13. Choosing the Right Tool Did Not Mean Finishing the Task {#mp-2026-10-02-013}

- Story ID: `mp-2026-10-02-013`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-013/choosing-the-right-tool-did-not-mean-finishing-the-task

**Dek:** HumanoidToolBench spans 18 tasks and exposes a gap from selection to mobile execution.

HumanoidToolBench combines tool choice, manipulation and locomotion across 18 simulated tasks, with ToolBook providing 3,100 demonstrations from simulation and a real Unitree G1. Seven simulated policies and three real-robot policies showed a substantial gap between selecting a suitable tool and completing the task. Focused probes also found lower selection accuracy on unseen tools and continued execution under unrelated instructions. The benchmark is a controlled evaluation, not a safety certification for humanoid deployment.

### Why it matters {#why-it-matters-mp-2026-10-02-013}

HumanoidToolBench spans 18 tasks and exposes a gap from selection to mobile execution.

### Limits and context {#limitations-mp-2026-10-02-013}

- The benchmark is a controlled evaluation, not a safety certification for humanoid deployment.

### Claims and sources {#claims-mp-2026-10-02-013}

- HumanoidToolBench spans 18 tasks and exposes a gap from selection to mobile execution. [source-2026-10-02-013] — Qualification: The benchmark is a controlled evaluation, not a safety certification for humanoid deployment.

## 14. One-Second Screening Cut a Third of Water Alerts {#mp-2026-10-02-014}

- Story ID: `mp-2026-10-02-014`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-014/one-second-screening-cut-a-third-of-water-alerts

**Dek:** A training-free first tier preserved reviewer accuracy while reducing LLM review load by 35 to 38 percent.

HydroJEV tested a fast probabilistic screen for distinguishing cyberattacks, physical faults, transients and faulty sensors in simulated water networks. Across four sealed rounds, the training-free model matched a rule tree in distribution and beat a supervised classifier on unseen event subtypes. Passing only benign verdicts confirmed by the rule tree spared an LLM reviewer 35 to 38 percent of windows without macro-F1 loss, and transfer tests stayed within a declared non-inferiority margin. Results come from benchmark networks, not live utility operations.

### Why it matters {#why-it-matters-mp-2026-10-02-014}

A training-free first tier preserved reviewer accuracy while reducing LLM review load by 35 to 38 percent.

### Limits and context {#limitations-mp-2026-10-02-014}

- Passing only benign verdicts confirmed by the rule tree spared an LLM reviewer 35 to 38 percent of windows without macro-F1 loss, and transfer tests stayed within a declared non-inferiority margin.
- Results come from benchmark networks, not live utility operations.

### Claims and sources {#claims-mp-2026-10-02-014}

- A training-free first tier preserved reviewer accuracy while reducing LLM review load by 35 to 38 percent. [source-2026-10-02-014] — Qualification: Passing only benign verdicts confirmed by the rule tree spared an LLM reviewer 35 to 38 percent of windows without macro-F1 loss, and transfer tests stayed within a declared non-inferiority margin.

## 15. Pottery Documentation Felt Forty Times Faster {#mp-2026-10-02-026}

- Story ID: `mp-2026-10-02-026`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-026/pottery-documentation-felt-forty-times-faster

**Dek:** Four open-source modules scan, ink, vectorize and lay out archaeological drawings.

PyPottery combines handwriting extraction, automated inking, semantic vectorization and layout generation for ceramic documentation. On 50 hand-drawn sheets containing 240 drawings from the Terramara di Montale, usability participants reported a median perceived speedup of 40 times, with a 17.5-to-120 range. The figure is user-perceived workflow acceleration, not a stopwatch guarantee, and the paper frames the suite as redistributing expert cognitive work rather than replacing archaeological judgment.

### Why it matters {#why-it-matters-mp-2026-10-02-026}

Four open-source modules scan, ink, vectorize and lay out archaeological drawings.

### Limits and context {#limitations-mp-2026-10-02-026}

- The figure is user-perceived workflow acceleration, not a stopwatch guarantee, and the paper frames the suite as redistributing expert cognitive work rather than replacing archaeological judgment.

### Claims and sources {#claims-mp-2026-10-02-026}

- Four open-source modules scan, ink, vectorize and lay out archaeological drawings. [source-2026-10-02-015] — Qualification: The figure is user-perceived workflow acceleration, not a stopwatch guarantee, and the paper frames the suite as redistributing expert cognitive work rather than replacing archaeological judgment.

## 16. Encrypted Reinforcement Learning Stopped Drifting {#mp-2026-10-02-027}

- Story ID: `mp-2026-10-02-027`
- Type: `dispatch`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-027/encrypted-reinforcement-learning-stopped-drifting

**Dek:** A zero-mean projection kept polynomial approximations inside their safe range without ciphertext bootstrapping.

Fully homomorphic reinforcement learning replaces nonlinear operations with polynomials, which can recursively amplify Bellman error. The Homomorphic Advantage Operator centers temporal-difference targets to remove the uniform value baseline that drives this drift while preserving action rankings. Across tabular, encrypted CartPole and logistics tests, reported agents had zero boundary breaches in all seeds, versus breaches in three of five regularized seeds and 83.8 percent of unstabilized episodes. These controlled results do not establish production security or broad RL performance.

### Why it matters {#why-it-matters-mp-2026-10-02-027}

A zero-mean projection kept polynomial approximations inside their safe range without ciphertext bootstrapping.

### Limits and context {#limitations-mp-2026-10-02-027}

- These controlled results do not establish production security or broad RL performance.

### Claims and sources {#claims-mp-2026-10-02-027}

- A zero-mean projection kept polynomial approximations inside their safe range without ciphertext bootstrapping. [source-2026-10-02-016] — Qualification: These controlled results do not establish production security or broad RL performance.

## 17. Forward KL Kept More High-Value Actions in Play {#mp-2026-10-02-015}

- Story ID: `mp-2026-10-02-015`
- Type: `ticker`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-015/forward-kl-kept-more-high-value-actions-in-play

**Dek:** FERPO improved policies without differentiating the critic with respect to actions.

FERPO derives an entropy- and KL-regularized target action distribution from critic values, then fits the actor with forward KL and self-normalized importance sampling. MuJoCo Playground and ManiSkill tests showed competitive performance, sample-efficiency gains and faster actor updates than REPPO.

### Why it matters {#why-it-matters-mp-2026-10-02-015}

FERPO improved policies without differentiating the critic with respect to actions.

### Limits and context {#limitations-mp-2026-10-02-015}

- No additional limitation was separately recorded.

### Claims and sources {#claims-mp-2026-10-02-015}

- FERPO improved policies without differentiating the critic with respect to actions. [source-2026-10-02-017]

## 18. A Continuous Latent Kept Parallel Tokens Connected {#mp-2026-10-02-016}

- Story ID: `mp-2026-10-02-016`
- Type: `ticker`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-016/a-continuous-latent-kept-parallel-tokens-connected

**Dek:** HC-DLM improved structured reasoning and perplexity at matched model size.

Hierarchical Continuous Diffusion Language Models use one persistent latent trajectory while decoded tokens feed back as a scaffold. On Sudoku, Countdown and LM1B, the reported model beat discrete and continuous diffusion baselines in puzzle accuracy and generative perplexity.

### Why it matters {#why-it-matters-mp-2026-10-02-016}

HC-DLM improved structured reasoning and perplexity at matched model size.

### Limits and context {#limitations-mp-2026-10-02-016}

- No additional limitation was separately recorded.

### Claims and sources {#claims-mp-2026-10-02-016}

- HC-DLM improved structured reasoning and perplexity at matched model size. [source-2026-10-02-018]

## 19. A Discriminator Replaced the Student's Extra Diffusion Model {#mp-2026-10-02-017}

- Story ID: `mp-2026-10-02-017`
- Type: `ticker`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-017/a-discriminator-replaced-the-student-s-extra-diffusion-model

**Dek:** DMAD matched distributions through log-density ratios and reached one-step ImageNet FID 1.04.

DMAD learns the density-ratio signal for few-step visual distillation with two discriminator heads instead of maintaining an auxiliary student-score model. The authors report leading compared image, video and audio-video results, including one-step ImageNet-64 FID 1.04.

### Why it matters {#why-it-matters-mp-2026-10-02-017}

DMAD matched distributions through log-density ratios and reached one-step ImageNet FID 1.04.

### Limits and context {#limitations-mp-2026-10-02-017}

- No additional limitation was separately recorded.

### Claims and sources {#claims-mp-2026-10-02-017}

- DMAD matched distributions through log-density ratios and reached one-step ImageNet FID 1.04. [source-2026-10-02-019]

## 20. Quasi-Newton Training Scaled to 136 Million Parameters {#mp-2026-10-02-018}

- Story ID: `mp-2026-10-02-018`
- Type: `ticker`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-018/quasi-newton-training-scaled-to-136-million-parameters

**Dek:** Positive-definite curvature estimates avoided line searches and GPU-unfriendly decompositions.

SoftServe builds diagonal and Kronecker-factored curvature estimates that remain positive definite under nonconvexity, using coupled Newton-Schulz matrix operations. It often reached lower losses than Adam, Muon and SOAP on ill-conditioned tasks including a 136-million-parameter physics-informed diffusion model.

### Why it matters {#why-it-matters-mp-2026-10-02-018}

Positive-definite curvature estimates avoided line searches and GPU-unfriendly decompositions.

### Limits and context {#limitations-mp-2026-10-02-018}

- SoftServe builds diagonal and Kronecker-factored curvature estimates that remain positive definite under nonconvexity, using coupled Newton-Schulz matrix operations.

### Claims and sources {#claims-mp-2026-10-02-018}

- Positive-definite curvature estimates avoided line searches and GPU-unfriendly decompositions. [source-2026-10-02-020] — Qualification: SoftServe builds diagonal and Kronecker-factored curvature estimates that remain positive definite under nonconvexity, using coupled Newton-Schulz matrix operations.

## 21. One Transport Controller Planned for a Shifted Start {#mp-2026-10-02-019}

- Story ID: `mp-2026-10-02-019`
- Type: `ticker`
- Classification: `editorial`
- Content status: `new`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-019/one-transport-controller-planned-for-a-shifted-start

**Dek:** Distributionally robust bridges traded some nominal performance for better perturbed-input recovery.

Distributionally Robust Schrödinger Bridge optimizes one stochastic controller against ambiguity in the initial distribution. Wasserstein and Sinkhorn variants improved robustness on two-dimensional transport and image translation, with an explicit nominal-performance tradeoff.

### Why it matters {#why-it-matters-mp-2026-10-02-019}

Distributionally robust bridges traded some nominal performance for better perturbed-input recovery.

### Limits and context {#limitations-mp-2026-10-02-019}

- No additional limitation was separately recorded.

### Claims and sources {#claims-mp-2026-10-02-019}

- Distributionally robust bridges traded some nominal performance for better perturbed-input recovery. [source-2026-10-02-021]

## 22. Space Drums 2.0 {#mp-2026-10-02-020}

- Story ID: `mp-2026-10-02-020`
- Type: `invention_desk`
- Classification: `editorial`
- Content status: `carried_over`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-020/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-2026-10-02-020}

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

### Limits and context {#limitations-mp-2026-10-02-020}

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

### Claims and sources {#claims-mp-2026-10-02-020}

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

## 23. 3RIC {#mp-2026-10-02-021}

- Story ID: `mp-2026-10-02-021`
- Type: `invention_desk`
- Classification: `editorial`
- Content status: `carried_over`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-021/3ric

**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-2026-10-02-021}

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

### Limits and context {#limitations-mp-2026-10-02-021}

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

### Claims and sources {#claims-mp-2026-10-02-021}

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

## 24. M3-CRETE {#mp-2026-10-02-022}

- Story ID: `mp-2026-10-02-022`
- Type: `invention_desk`
- Classification: `editorial`
- Content status: `carried_over`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-022/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-2026-10-02-022}

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

### Limits and context {#limitations-mp-2026-10-02-022}

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

### Claims and sources {#claims-mp-2026-10-02-022}

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

## 25. DeskNode {#mp-2026-10-02-023}

- Story ID: `mp-2026-10-02-023`
- Type: `invention_desk`
- Classification: `editorial`
- Content status: `carried_over`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-023/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-2026-10-02-023}

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

### Limits and context {#limitations-mp-2026-10-02-023}

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

### Claims and sources {#claims-mp-2026-10-02-023}

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

## 26. The First Paid Slot {#mp-2026-10-02-024}

- Story ID: `mp-2026-10-02-024`
- Type: `invention_desk`
- Classification: `house_example`
- Content status: `carried_over`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-024/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-2026-10-02-024}

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

### Limits and context {#limitations-mp-2026-10-02-024}

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

### Claims and sources {#claims-mp-2026-10-02-024}

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

## 27. Put Your Project on the Desk {#mp-2026-10-02-025}

- Story ID: `mp-2026-10-02-025`
- Type: `invention_desk`
- Classification: `house_example`
- Content status: `carried_over`
- Permanent URL: https://themachinepress.com/story/mp-2026-10-02-025/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-2026-10-02-025}

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

### Limits and context {#limitations-mp-2026-10-02-025}

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

### Claims and sources {#claims-mp-2026-10-02-025}

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

## Normalized sources

- **source-2026-10-02-001:** [arXiv preprint 2610.02070](https://arxiv.org/abs/2610.02070) — arXiv; primary_research
- **source-2026-10-02-002:** [arXiv preprint 2610.02066](https://arxiv.org/abs/2610.02066) — arXiv; primary_research
- **source-2026-10-02-003:** [arXiv preprint 2610.02204](https://arxiv.org/abs/2610.02204) — arXiv; primary_research
- **source-2026-10-02-004:** [arXiv preprint 2610.02202](https://arxiv.org/abs/2610.02202) — arXiv; primary_research
- **source-2026-10-02-005:** [arXiv preprint 2610.02206](https://arxiv.org/abs/2610.02206) — arXiv; primary_research
- **source-2026-10-02-006:** [arXiv preprint 2610.02200](https://arxiv.org/abs/2610.02200) — arXiv; primary_research
- **source-2026-10-02-007:** [arXiv preprint 2610.02207](https://arxiv.org/abs/2610.02207) — arXiv; primary_research
- **source-2026-10-02-008:** [arXiv preprint 2610.02201](https://arxiv.org/abs/2610.02201) — arXiv; primary_research
- **source-2026-10-02-009:** [arXiv preprint 2610.02122](https://arxiv.org/abs/2610.02122) — arXiv; primary_research
- **source-2026-10-02-010:** [arXiv preprint 2610.02117](https://arxiv.org/abs/2610.02117) — arXiv; primary_research
- **source-2026-10-02-011:** [arXiv preprint 2610.02186](https://arxiv.org/abs/2610.02186) — arXiv; primary_research
- **source-2026-10-02-012:** [arXiv preprint 2610.02180](https://arxiv.org/abs/2610.02180) — arXiv; primary_research
- **source-2026-10-02-013:** [arXiv preprint 2610.02089](https://arxiv.org/abs/2610.02089) — arXiv; primary_research
- **source-2026-10-02-014:** [arXiv preprint 2610.02048](https://arxiv.org/abs/2610.02048) — arXiv; primary_research
- **source-2026-10-02-015:** [arXiv preprint 2610.02072](https://arxiv.org/abs/2610.02072) — arXiv; primary_research
- **source-2026-10-02-016:** [arXiv preprint 2610.02074](https://arxiv.org/abs/2610.02074) — arXiv; primary_research
- **source-2026-10-02-017:** [arXiv preprint 2610.02198](https://arxiv.org/abs/2610.02198) — arXiv; primary_research
- **source-2026-10-02-018:** [arXiv preprint 2610.02193](https://arxiv.org/abs/2610.02193) — arXiv; primary_research
- **source-2026-10-02-019:** [arXiv preprint 2610.02188](https://arxiv.org/abs/2610.02188) — arXiv; primary_research
- **source-2026-10-02-020:** [arXiv preprint 2610.02182](https://arxiv.org/abs/2610.02182) — arXiv; primary_research
- **source-2026-10-02-021:** [arXiv preprint 2610.02043](https://arxiv.org/abs/2610.02043) — arXiv; primary_research