{"resourceId":"pacman-diiid-control-boundaries-preprint-2025","versions":[{"version":"external-80a51f8e92e7b7bc74f6ad75591aa3ae5748553bcca801ea18fbf40b506314b9","resource":{"id":"pacman-diiid-control-boundaries-preprint-2025","title":"PACMAN integrates research control models with explicit timing and failure boundaries","organization":"Princeton University, Princeton Plasma Physics Laboratory and collaborators","sector":"University and national-laboratory scientific instrumentation","geography":"DIII-D, California, United States; collaboration includes Japan","publishedAt":"November 11, 2025, arXiv v1; newly relevant following PPPL's September 2, 2026 report","publicationDate":"2025-11-11","eventDate":null,"sourceName":"arXiv","sourceLabel":"Author preprint; architecture, application tables and limitations inspected","sourceUrl":"https://arxiv.org/html/2511.08818v1","evidenceClass":"academic-research","outcomeClass":"mixed","topics":["developers-agents","infrastructure","data-security","governance-procurement","operating-model"],"finding":"PACMAN demonstrates modular ML control on a research instrument while documenting situations where timing or missing inputs limit applicability.","sledRelevance":"Relevant to university instrument-software teams; transfer concerns architectural evaluation, not a ready-made controller for other laboratories. This is specialized ML, not an LLM copilot.","evidence":"The preprint details five experimental applications with diagnostic, model, controller and actuator stages. Its profile controller uses 4 ms encoding plus up to 10 ms optimization, enabling a 20 ms cycle. These are instrument-specific timings, not a general productivity estimate or controlled comparative trial.","architectureImplications":"Interpretation: keep actuator authorization and safety checks outside learned predictions; validate the complete sensing-to-actuation deadline on target hardware.","governanceImplications":"Interpretation: assign instrument experts authority over model release, operating limits and fallback selection.","securityPrivacyImplications":"Interpretation: isolate instrument control networks, restrict model/configuration changes and authenticate telemetry and deployment artifacts.","caveats":"Inspected 2025 preprint, not the inaccessible journal version. It excludes sub-millisecond vertical-displacement control and warns that inaction can be hazardous. Some detailed application results are in separate papers not inspected. No cross-instrument replication or general failure-rate estimate.","streamIds":["research"],"roles":{"sales":"Interpretation: qualify needs with laboratory directors, instrument engineers, computational scientists and safety owners. Ask which control tasks already have valid sensor feeds, what response deadlines apply and how current experimental software is maintained. The value hypothesis is more maintainable integration of bounded models, subject to local testing. A suitable engagement is an architecture and replay assessment for one existing instrument workflow. The evidence does not justify unattended experiments, guaranteed safety or broadly faster discovery. Explain that conventional control requirements still determine fit and that a successful demonstration elsewhere supplies design questions rather than authorization for deployment.","engineering":"Interpretation: design explicit interfaces for diagnostics, prediction, controller decisions and final actuator requests. Establish local timing budgets and hardware constraints before adopting any model. Replay known traces and inject stale readings, missing diagnostics, invalid predictions and conflicting commands. Evaluate worst-case latency and recovery under the actual controller implementation. A hold, shutdown or alternative action must be justified by instrument physics; do not assume doing nothing is always safe. Keep credentials and configuration authority outside experimental model code. Proposed proof should demonstrate bounded commands, intelligible fault reporting and validated transitions to the approved fallback before supervised physical operation.","delivery":"Interpretation: an instrument operations owner should coordinate staged release with control engineers and disciplinary scientists. Inventory signal sources, supported operating regimes and current manual procedures; preserve approved models and rollback artifacts. Dependencies include a replay environment, safe test windows and personnel qualified to interpret faults. Train operators through accessible state displays and rehearsed recovery procedures. Governance gates should approve simulation, supervised trials and subsequent model changes separately. Proposed acceptance criteria include meeting the agreed worst-case deadline, passing every critical fault-injection case and recording all actuator overrides. Risks include mismatched timescales, silent sensor degradation and inappropriate reuse of another instrument's fallback behavior."},"retrievedAt":"2026-09-08T03:02:22Z","enrichedAt":"2026-09-08T03:06:13Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Interpretation: make fault state and override controls accessible; control and physics expertise remain necessary.","procurementImplications":"Interpretation: require timing evidence, supported interfaces, versioned artifacts and responsibility for maintenance.","operatingModelImplications":"Interpretation: operators own physical release authority; software teams maintain regression and recovery tests.","updateExplanation":"Absent from the full archive. The inspected September 2 PPPL report (https://www.pppl.gov/news/2026/pacman-ai-framework-controlling-fusion-systems-safely-makes-key-decisions-milliseconds) motivated opening the older author manuscript for technical constraints. No claim that the preprint itself changed or that journal and preprint versions are identical.","sourceVerification":{"openedUrl":"https://arxiv.org/html/2511.08818v1","referenceExcerpt":"sometimes inaction can potentially be more dangerous than a bad action.","promptVersion":"sled-research-v3.1","model":null,"basis":"agent-reported inspection"}}}]}