simjecture 0.6.1__tar.gz → 0.6.2__tar.gz

{simjecture-0.6.1 → simjecture-0.6.2}/.gitattributes

@@ -1,3 +1,5 @@
+*.sh text eol=lf
+
 # Preserve autonomous run artifacts byte-for-byte, including model-authored
 # formatting. Generated figures likewise need not satisfy source whitespace
 # checks.

{simjecture-0.6.1 → simjecture-0.6.2}/.github/workflows/ci.yml

@@ -9,6 +9,22 @@ permissions:
   contents: read
 
 jobs:
+  windows-application-boundary:
+    runs-on: windows-latest
+    timeout-minutes: 5
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.12"]
+    steps:
+      - uses: actions/checkout@v7
+      - uses: actions/setup-python@v7
+        with:
+          python-version: ${{ matrix.python-version }}
+      - name: Test shared Windows process ownership without vendor applications
+        run: python scripts/check_cst_windows.py
+      - name: Test legacy CST entrypoint compatibility
+        run: python scripts/check_cst_windows.py --runner src/conjecture_solver/cst_windows.py
+
   documentation:
     runs-on: ubuntu-latest
     timeout-minutes: 10
@@ -28,7 +44,8 @@ jobs:
 
   deterministic-core:
     runs-on: ubuntu-latest
-    timeout-minutes: 15
+    # Pytest takes about 13 minutes; leave time for setup and integration checks.
+    timeout-minutes: 30
     steps:
       - uses: actions/checkout@v7
       - uses: actions/setup-python@v7
@@ -67,7 +84,7 @@ jobs:
       - name: Lint
         run: uv run ruff check .
       - name: Test
-        run: uv run pytest --cov=conjecture_solver --cov-report=term-missing
+        run: uv run pytest --cov=conjecture_solver --cov-report=term-missing --durations=20
       - name: Build source distribution and wheel from it
         run: uv build --out-dir /tmp/simjecture-python-distributions
       - name: Validate packaged workspace install

{simjecture-0.6.1 → simjecture-0.6.2}/CHANGELOG.md

@@ -4,6 +4,110 @@ This project follows semantic versioning. Dates use ISO 8601.
 
 ## Unreleased
 
+## 0.6.2 — 2026-10-16
+
+- Bring the 0.6.2 candidate series into the stable release: project/workspace
+  collections, director-controlled registration recovery and storage grants,
+  explicit experimental solver development, large-build handling and portable
+  LF shell scripts. Preserve existing study identities and scientific review gates.
+
+- Verify fresh-home installations through the downloaded workspace archive and
+  checksum rather than accidentally selecting the source checkout. Update the
+  locked DSH transitive MCP SDK and sharp dependencies to patched versions while
+  retaining the pinned DSH release.
+
+- Retry transient Windows receipt-replacement sharing errors for a bounded time,
+  preserving the previous complete receipt. Cover an actual Windows reader lock
+  and include supervisor/worker logs in failed native integration checks.
+
+- Add an experimental common Windows application interface with discovery,
+  frozen per-job runner bundles, complete input inventories, asynchronous jobs,
+  deadlines, cancellation, process ownership and verified output artifacts.
+  Keep the CST commands on the shared runner and add native CST Python scripts
+  plus an experimental SOLIDWORKS COM part rebuild/export adapter. SOLIDWORKS
+  model commissioning and COMSOL execution remain outside the current validation.
+  Add Linux regression coverage and native Windows process/fixture tests without
+  vendor applications. Jobs use trusted host processes and remain exploratory.
+  Ship a discoverable `cst-geometry` skill covering imported-solid edits, recorded
+  history, full-profile comparisons and assembly checks through authorized native
+  automation; native CST scripts use the generic Windows submission interface.
+  The contributor validated the shared runner on CST 2024.4 with fresh Two Coils
+  solves, native scripting, artifact verification and cancellation while preserving
+  the source model and an unrelated application session.
+
+- Park research workers while awaiting useful events, detect status polling that
+  conceals stalled progress, and make independent report assessment available
+  during research. Use compact, hashed artifact indexes in review packets and
+  expose native-agent accounting coverage. Preserve original scientific receipts
+  and distinguish accepted unresolved reports from accepted scientific claims.
+
+- Preserve autonomous-study tool selections across agent redrafting and restarts.
+  Teach native agents to attach registered tools in study briefs, retain declared
+  instruments, and reject missing or mismatched declared tools before launch.
+  Explicitly selecting ordinary Python still clears a saved tool selection.
+
+- Give new studies all available installed tools by default across chat, browser,
+  CLI and terminal UI. Tool restrictions are an advanced option; solver requirements
+  are separate. Freeze each study's runtime inventory, keep remote tools on their
+  configured workers, and qualify the instruments actually used. Existing studies
+  retain their frozen inventories and methods policies on resume.
+
+- Extend active minimal studies from the workspace, experiment monitor, CLI or local
+  operator API. Deadline changes are live, idempotent and recorded with their
+  original budget; the director replans without restarting numerical jobs. Agents
+  can request time, while grants remain an operator action. Individual experiment
+  limits, report reserves and scientific contracts remain unchanged.
+
+- Accept lists of text for study brief success criteria and constraints, preserving
+  their items as paragraphs and enforcing the existing 20,000-character limit.
+  Clarify native-agent brief instructions and distinguish invalid types from
+  oversized text in validation errors.
+
+## 0.6.2rc2 — 2026-10-14
+
+- Keep token-usage/status inspection working for historical or partial study records
+  without storage allowances; report unknown limits without inventing admission budgets.
+- Let the research director tune remote instrument probe deadlines and retry counts.
+  Preserve the 60-second/one-attempt default, frozen identities and report reserve;
+  record actionable transport failures and show attempts and grants in the run monitor.
+- Keep shell scripts in Unix LF format even with Windows Git checkout settings.
+  Verify that the ITER pack bootstrap launches after a `core.autocrlf=true` checkout.
+- Make experimental solver development a visible, default-on permission for new
+  minimal studies. Preserve older study instructions, inherit explicit choices in
+  continuations, and expose the policy through the GUI, CLI and agent drafting tools.
+- Record optional numerical/physical/diagnostic change provenance against separately
+  registered builds. Show reviewed method scopes in the research monitor and require
+  source-bound production-method approval for newly registered evidence instruments.
+  Preserve restricted hosted-study permissions and reference identities.
+
+- Fix large scientific builds being stopped by a separate 512 MiB file ceiling:
+  use the admitted workspace allowance for individual files and input staging,
+  reject oversized inputs before hashing, and record executor identity and limits.
+- Let operators delegate recorded storage increases to the research director.
+  Grants wake the worker and apply to future admissions; existing reservations,
+  scientific requirements and deadlines remain unchanged. Hosted allowances remain
+  protected by default. Show delegated authority, current allowances and recorded
+  grants or rejections in the research monitor.
+- Fix the generated client's registration argument collision and support newly
+  built instruments on frozen SSH workers. Append machine-scoped identities usable
+  for execution, source inspection and methods review while preserving original
+  instrument and worker identities.
+
+## 0.6.2rc1 — 2026-10-09
+
+- Add optional local workspaces and project collections, with a workspace switcher,
+  shared instructions/files, unchanged conversation paths and frozen context for
+  agent turns and autonomous studies. Expose the same records through JSON CLI
+  commands and scoped HTTP APIs. Existing users open Personal without a setup flow;
+  the hosted trial retains its existing account-scoped interface.
+- Preserve project selection when reloading a new-conversation page and keep
+  study navigation linked to its owning workspace. Collect research tools,
+  machines, benchmarks and connections under Resources; the single-user view
+  keeps these controls expanded by default.
+- Include a project/workspace guide covering CLI operations, context provenance,
+  private data storage and the distinction between local organisation and hosted
+  account isolation. No private device assets are distributed with this candidate.
+
 ## 0.6.1 — 2026-10-08
 
 - Include cylinder-flow validation data in source distributions as well as

{simjecture-0.6.1 → simjecture-0.6.2}/CITATION.cff

@@ -2,8 +2,8 @@ cff-version: 1.2.0
 message: "If you use this software, please cite the exact software version and Git commit."
 title: "Simjecture"
 type: software
-version: 0.6.1
-date-released: 2026-10-08
+version: 0.6.2
+date-released: 2026-10-16
 abstract: >-
   An evidence-governed runtime for autonomous computational experimentation and
   falsification over hypothesis trees.

{simjecture-0.6.1 → simjecture-0.6.2}/CONTRIBUTING.md

@@ -1,5 +1,7 @@
 # Contributing
 
+[中文操作指南：安装、Fork、开发和提交 PR](docs/zh-cn/research-group-guide.md)。
+
 Thanks for helping improve Simjecture. You can contribute without running a full
 research campaign or installing a scientific solver. Useful contributions include:
 

{simjecture-0.6.1 → simjecture-0.6.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: simjecture
-Version: 0.6.1
+Version: 0.6.2
 Summary: Evidence-governed autonomous experimentation and falsification for computational science
 Project-URL: Homepage, https://github.com/tomzhu0225/simjecture
 Project-URL: Documentation, https://drawingsword.com/simjecture/
@@ -57,6 +57,8 @@ Description-Content-Type: text/markdown
 [Research loop](https://drawingsword.com/simjecture/concepts/research-loop.html) · [Recorded studies](https://drawingsword.com/simjecture/demos/index.html) ·
 [Contributing](https://github.com/tomzhu0225/simjecture/blob/main/CONTRIBUTING.md)
 
+[中文安装、使用与开发指南](docs/zh-cn/research-group-guide.md)
+
 **Hypothesize. Simulate. Falsify.**
 
 Simjecture runs autonomous numerical investigations from a human-defined scientific
@@ -83,11 +85,28 @@ claim. New browser studies default to answer; the CLI defaults to repair. Both u
 minimal mode by default: flexible agent strategy, recorded experiments and explicit
 scientific acceptance. A deadline can leave the question unresolved.
 
+New minimal studies allow experimental solver development by default. The agent can
+work in a private source fork while preserving the reference installation. Changed
+builds need independent, source-bound qualification before supporting a claim. Turn
+this off in the study brief or with `--no-allow-solver-development`; see
+[solver development and qualification](https://drawingsword.com/simjecture/how-to/minimal-oversight.html#solver-development-permission).
+
+The optional research director can stop unaffordable experiments, request a replan
+and tune remote instrument probe deadlines and retries. Operators can also delegate
+storage increases for large builds and checkpoints. The monitor shows failures,
+before/after limits and the reasons for changes; scientific acceptance and the
+campaign deadline remain separate. See [director control and recovery](https://drawingsword.com/simjecture/how-to/minimal-oversight.html#instrument-registration-recovery).
+
 Read [how the loop works](https://drawingsword.com/simjecture/concepts/research-loop.html) or
 [how evidence is checked](https://drawingsword.com/simjecture/concepts/evidence-and-claims.html).
 
 ## Follow a real investigation
 
+The [**48 hour FRC campaign on eight GPUs**](docs/research/frc-48-hour-campaign-20261014.md)
+ended with its merger hypothesis unresolved. Its report shows the retained plasma
+fields, numerical controls, model usage, API cost estimates and the controller
+failures exposed by two days of research. Public tables reproduce the accounting.
+
 [**Magnetic-island coalescence with FLASH**](https://drawingsword.com/simjecture/demos/island-coalescence.html)
 shows density, magnetic geometry and a current sheet evolving in a real MHD
 calculation. Follow five resistivity cases, two refinements and their evidence
@@ -160,6 +179,11 @@ The [first-study tutorial](https://drawingsword.com/simjecture/getting-started/f
 counterexample-and-repair workflow. [Headless studies](https://drawingsword.com/simjecture/how-to/headless-studies.html)
 cover terminal launches and automation.
 
+For longer-running work, optional [projects and workspaces](docs/how-to/projects-and-workspaces.md)
+group conversations and shared context, or keep separate local lab environments.
+The default remains a single personal workspace. CLI and HTTP interfaces expose
+the same project operations for external agents and scripts.
+
 ## Bring your instruments and compute
 
 Ordinary Python studies work without an external solver. Optional capabilities
@@ -167,6 +191,13 @@ include WarpX, operator-supplied FLASH, EOS/opacity tools, ITER diagnostics and
 [GPU cylinder flow](https://drawingsword.com/simjecture/how-to/cylinder-flow.html). The cylinder solver and its
 validation records were contributed by [Zifei Meng](https://github.com/ZifeiMengSPH).
 
+An experimental [Windows application interface](docs/how-to/windows-applications.md)
+shares job execution, input snapshots and artifact collection across product adapters.
+It supports [CST magnetostatic jobs](docs/how-to/cst-windows.md), native CST Python
+scripts, and an initial SOLIDWORKS part rebuild/export adapter. These use trusted
+Windows processes and record exploratory jobs; SOLIDWORKS still needs native
+model commissioning.
+
 - [Commission a scientific instrument](https://drawingsword.com/simjecture/how-to/guided-commissioning.html) or
   [add your own capability](https://drawingsword.com/simjecture/how-to/add-a-capability.html).
 - [Run experiments on local or SSH workers](https://drawingsword.com/simjecture/how-to/ssh-workers.html).

{simjecture-0.6.1 → simjecture-0.6.2}/README.md

@@ -8,6 +8,8 @@
 [Research loop](https://drawingsword.com/simjecture/concepts/research-loop.html) · [Recorded studies](https://drawingsword.com/simjecture/demos/index.html) ·
 [Contributing](https://github.com/tomzhu0225/simjecture/blob/main/CONTRIBUTING.md)
 
+[中文安装、使用与开发指南](docs/zh-cn/research-group-guide.md)
+
 **Hypothesize. Simulate. Falsify.**
 
 Simjecture runs autonomous numerical investigations from a human-defined scientific
@@ -34,11 +36,28 @@ claim. New browser studies default to answer; the CLI defaults to repair. Both u
 minimal mode by default: flexible agent strategy, recorded experiments and explicit
 scientific acceptance. A deadline can leave the question unresolved.
 
+New minimal studies allow experimental solver development by default. The agent can
+work in a private source fork while preserving the reference installation. Changed
+builds need independent, source-bound qualification before supporting a claim. Turn
+this off in the study brief or with `--no-allow-solver-development`; see
+[solver development and qualification](https://drawingsword.com/simjecture/how-to/minimal-oversight.html#solver-development-permission).
+
+The optional research director can stop unaffordable experiments, request a replan
+and tune remote instrument probe deadlines and retries. Operators can also delegate
+storage increases for large builds and checkpoints. The monitor shows failures,
+before/after limits and the reasons for changes; scientific acceptance and the
+campaign deadline remain separate. See [director control and recovery](https://drawingsword.com/simjecture/how-to/minimal-oversight.html#instrument-registration-recovery).
+
 Read [how the loop works](https://drawingsword.com/simjecture/concepts/research-loop.html) or
 [how evidence is checked](https://drawingsword.com/simjecture/concepts/evidence-and-claims.html).
 
 ## Follow a real investigation
 
+The [**48 hour FRC campaign on eight GPUs**](docs/research/frc-48-hour-campaign-20261014.md)
+ended with its merger hypothesis unresolved. Its report shows the retained plasma
+fields, numerical controls, model usage, API cost estimates and the controller
+failures exposed by two days of research. Public tables reproduce the accounting.
+
 [**Magnetic-island coalescence with FLASH**](https://drawingsword.com/simjecture/demos/island-coalescence.html)
 shows density, magnetic geometry and a current sheet evolving in a real MHD
 calculation. Follow five resistivity cases, two refinements and their evidence
@@ -111,6 +130,11 @@ The [first-study tutorial](https://drawingsword.com/simjecture/getting-started/f
 counterexample-and-repair workflow. [Headless studies](https://drawingsword.com/simjecture/how-to/headless-studies.html)
 cover terminal launches and automation.
 
+For longer-running work, optional [projects and workspaces](docs/how-to/projects-and-workspaces.md)
+group conversations and shared context, or keep separate local lab environments.
+The default remains a single personal workspace. CLI and HTTP interfaces expose
+the same project operations for external agents and scripts.
+
 ## Bring your instruments and compute
 
 Ordinary Python studies work without an external solver. Optional capabilities
@@ -118,6 +142,13 @@ include WarpX, operator-supplied FLASH, EOS/opacity tools, ITER diagnostics and
 [GPU cylinder flow](https://drawingsword.com/simjecture/how-to/cylinder-flow.html). The cylinder solver and its
 validation records were contributed by [Zifei Meng](https://github.com/ZifeiMengSPH).
 
+An experimental [Windows application interface](docs/how-to/windows-applications.md)
+shares job execution, input snapshots and artifact collection across product adapters.
+It supports [CST magnetostatic jobs](docs/how-to/cst-windows.md), native CST Python
+scripts, and an initial SOLIDWORKS part rebuild/export adapter. These use trusted
+Windows processes and record exploratory jobs; SOLIDWORKS still needs native
+model commissioning.
+
 - [Commission a scientific instrument](https://drawingsword.com/simjecture/how-to/guided-commissioning.html) or
   [add your own capability](https://drawingsword.com/simjecture/how-to/add-a-capability.html).
 - [Run experiments on local or SSH workers](https://drawingsword.com/simjecture/how-to/ssh-workers.html).

{simjecture-0.6.1 → simjecture-0.6.2}/docs/archive/index.md

@@ -61,6 +61,7 @@ classic-evidence-and-claims
 Original scientific records remain intact. Read each report's qualifications and
 corrections together with the claimed result.
 
+- [48 hour FRC campaign on eight GPUs, 2026-10-14](../research/frc-48-hour-campaign-20261014.md)
 - [Run 0004 audit](../research/run-0004.md)
 - [Demo refresh and harness audit, 2026-10-08](../research/demo-audit-20261008.md)
 - [Historical next-step plan](../research/next-steps.md)
@@ -69,6 +70,8 @@ corrections together with the claimed result.
 - [Adaptive aluminium campaign audit, 2026-10-04](../research/adaptive-campaign-audit-20261004.md)
 - [Z-pinch execution and timestep repair qualification, 2026-10-03](../testing/zpinch-repair-qualification-20261003.md)
 - [Research director and adaptive FLASH qualification, 2026-10-03](../testing/research-director-adaptive-20261003.md)
+- [Large-build, remote-registration and director storage repairs, 2026-10-12](../testing/large-build-remote-registration.md)
+- [Research feedback and evidence-packet repair, 2026-10-14](../testing/research-feedback-repair-20261014.md)
 - [Stagnation retrospective, 2026-09-29](../research/stagnation-retrospective-20260929.md)
 - [Stagnation video outline, 2026-09-29](../research/stagnation-video-outline-20260929.md)
 
@@ -83,6 +86,8 @@ corrections together with the claimed result.
 ../research/adaptive-campaign-audit-20261004
 ../testing/zpinch-repair-qualification-20261003
 ../testing/research-director-adaptive-20261003
+../testing/large-build-remote-registration
+../testing/research-feedback-repair-20261014
 ../research/stagnation-retrospective-20260929
 ../research/stagnation-video-outline-20260929
 ```

{simjecture-0.6.1 → simjecture-0.6.2}/docs/getting-started/research-workspace.md

@@ -144,6 +144,11 @@ only when a continuation is submitted.
 
 ## Permanent files, arranged by project
 
+For optional conversation collections and separate local lab environments, see
+[Projects and workspaces](../how-to/projects-and-workspaces.md). You can continue
+using a single conversation without setting either up. The folders below retain
+their historical `projects/` name and each represent one conversation.
+
 Project and study folders use names you can recognize:
 
 ```text

{simjecture-0.6.1 → simjecture-0.6.2}/docs/how-to/continuation-steering.md

@@ -2,6 +2,10 @@
 
 Continuation and steering apply to minimal studies.
 
+For more time on the same active question, use [Extend time](extend-study.md).
+This updates the current study's clock with an operator audit record. A continuation
+is for a finished study, a revised question or a new phase with its own scope.
+
 ## Continue after a deadline or change direction
 
 Open the study in **Autonomous research**, or open its **experiment monitor**.
@@ -79,8 +83,9 @@ simjecture steer --campaign runs/phase-2 \
 
 Guidance is advisory. It does not rewrite commitments, change acceptance rules,
 extend time or interrupt active simulations. Use the existing Pause/Stop controls
-for execution control. For new requirements, a new hypothesis or more time, use
-**Continue investigation**. This keeps the original contracts auditable while
+for execution control. Use **Extend time** for an active study's budget, and
+**Continue investigation** for new requirements, a new hypothesis or an ended study.
+This keeps the original contracts auditable while
 leaving the research agent free to plan its next experiments.
 
 ## Review and status recovery

simjecture-0.6.2/docs/how-to/cst-windows.md

@@ -0,0 +1,189 @@
+# Use an installed Windows CST from WSL
+
+The experimental `simjecture cst` bridge keeps Simjecture on WSL and uses native
+Windows Python to operate an existing CST Studio Suite installation. It can
+inspect saved 0D/1D results and submit self-contained **LF MStatic** projects with
+numeric parameter changes. These commands now use the
+[shared Windows application runner](windows-applications.md), which also offers
+native CST Python scripts. Remote Windows hosts, external input staging and
+direct evidence-stage `lab.run` integration are outside this version.
+
+The operator supplies CST, a compatible Windows Python and the necessary solver
+authorization. Nothing downloads or redistributes CST. CST 2024 includes compatible
+Python libraries and may include an interpreter at `Opera/code/bin/python.exe`.
+Use `--python` if your installation needs a different compatible interpreter.
+
+## Check the interface and inspect an existing result
+
+Run these commands from WSL. A single installation in the standard
+`/mnt/c/Program Files (x86)` directory is detected automatically. Otherwise use
+`--installation` or set `SIMJECTURE_CST_INSTALLATION` to the installed directory.
+Windows absolute paths are also accepted; quote spaces and backslashes.
+
+```bash
+simjecture cst check --installation '/mnt/c/Program Files (x86)/CST Studio Suite 2024'
+
+simjecture cst inspect \
+  --project '/mnt/c/work/examples/Two Coils.cst' \
+  --result '1D Results\Ms Solver\Inductance Matrix\L coil1, coil1'
+```
+
+`check` loads the Windows APIs without launching a frontend or solver. It reports
+`solver_tested=false`; this is not numerical readiness. `inspect` lists available
+0D/1D tree paths and, for requested paths, returns run IDs, parameter combinations,
+axis labels and values. Complex values use `{real, imag}` objects. Missing result
+paths and non-finite values fail explicitly.
+Run ID 0 is CST's current-result view and may duplicate a numbered run; it must
+not be counted as an independent replicate.
+
+Save and close the project before inspection. If it must remain open, use
+`--allow-open` after saving and while no calculation is active. CST's reader
+cannot see unsaved state. It requires unpacked, unprotected projects supported
+by that installed reader version; see the local CST Python documentation.
+
+## Submit and monitor a magnetostatic job
+
+Save and close the source project, and keep its files unchanged during submission.
+Choose a separate job root on a local Windows drive
+mounted in WSL. The operator-side job directory contains the immutable input
+snapshot, writable project copy, request, state, results and Windows logs.
+
+```bash
+simjecture cst submit \
+  --project '/mnt/c/work/examples/Two Coils.cst' \
+  --job-root '/mnt/c/work/cst-jobs' \
+  --parameter r=2 \
+  --result '1D Results\Ms Solver\Inductance Matrix\L coil1, coil1' \
+  --result '1D Results\Ms Solver\Inductance Matrix\L coil1, coil2' \
+  --result '1D Results\Ms Solver\Inductance Matrix\L coil2, coil2' \
+  --timeout 600
+
+# Use the directory returned by submit:
+simjecture cst status '/mnt/c/work/cst-jobs/cst_JOB_ID'
+simjecture cst wait '/mnt/c/work/cst-jobs/cst_JOB_ID'
+simjecture cst cancel '/mnt/c/work/cst-jobs/cst_JOB_ID'
+```
+
+Submission is asynchronous. Jobs serialize per CST installation, including across
+different job roots. Their wall deadline includes queueing. A detached Windows
+supervisor owns a Windows Job Object containing the worker and its descendant
+processes. Timeout, cancellation or supervisor termination closes that job and
+terminates its members. The runner creates its own CST instance; it does not
+connect to the operator's existing frontend or kill every CST process by name.
+The per-installation lock remains held through process cleanup and terminal-state
+recording. The worker disables CST distributed computing after rebuilding and
+immediately before starting the solver. Saved distributed settings are overridden
+in the working copy because DC services and remote solvers are outside the local
+Job Object. Receipts record `solver_execution="local"`.
+
+`cancel` records a request; inspect `status` until a terminal outcome is reported.
+The supervisor runs on Windows and continues enforcing the deadline if the WSL
+caller exits. Windows shutdown and loss of access to the job drive require
+operator reconciliation; a stale `running` receipt must not be reported as a
+confirmed successful execution or cancellation.
+If supervisor creation definitely fails before a process starts, the receipt is
+marked `failed` and the launch marker is removed. Submit again to create a new job;
+terminal jobs are not relaunched. Failures after process creation preserve the
+launch marker and follow the cancellation/reconciliation path.
+
+The source `.cst` and its complete companion `Model` file inventory are hashed
+before and after copying. Both inventories and the copied hashes must agree;
+changed, added or removed files cause submission to fail before a Windows launch.
+The worker also checks the complete queued snapshot and its writable copy before
+opening CST. Each new job freezes the shared runner and adapter source files so
+an update to the client package does not replace a queued job's runtime.
+Existing `Result`, `Temp`, `SP` and mesh-cache directories are not staged. The
+Windows worker deletes results in its writable project copy, sets finite numeric
+parameters, rebuilds and starts the magnetostatic solver. It saves the output and
+extracts the declared result tree paths. The original project is preserved.
+Reference files outside the project are not automatically discovered or staged;
+use a self-contained starting model for this version.
+
+Inspect `worker.log` and `supervisor.log` for failures. A successful receipt binds
+the `results.json` hash; later output modification is detected by `status`.
+The request also records input hashes and the runner and Python executable
+identities. Retain the entire job directory for diagnosis and reproduction.
+
+If startup fails with **WinError 740**, Windows requires elevation for that CST
+executable. This can be caused by the installation's “Run as administrator”
+compatibility setting. Use an operator-authorized launch context appropriate for
+the installation. The bridge does not change Windows compatibility settings,
+request UAC elevation or attach to an existing elevated frontend. Saved-result
+inspection may still work in an ordinary WSL session when frontend startup is
+blocked. Report those checks separately from a successful fresh solve.
+
+## Geometry editing guidance for agents
+
+The discoverable `cst-geometry` skill documents how an authorized host agent can
+inspect and edit geometry through the installed CST Python/VBA interfaces. It
+covers imported solids, recorded modeling history, non-ASCII shape names, full
+section comparisons, smooth joins, and assembly interfaces. Native agents read
+`skills/cst-geometry/SKILL.md`; API agents use `read_skill(name="cst-geometry")`
+and load its linked references as needed. The skill ships with the Python package.
+
+This is reusable workflow guidance, not a geometry-editing CLI command. It does
+not extend `cst submit` beyond LF MStatic. A geometry edit should be saved to a
+separate project and checked by rebuilding, reopening, and reviewing the complete
+profile. An increased tip radius alone can hide new connector faces and sharp
+joins. A successful geometry edit does not establish simulation accuracy.
+
+## Commission the actual model
+
+The official **Two Coils** example is a useful initial instrument: in CST's
+Component Library, find the example and save a local copy. The documented model
+uses metre geometry units, two five-turn current-driven coils and a linear
+relative permeability of 1000. Its existing sweep can contain saved results.
+For bridge acceptance, submit a new job for each parameter combination and
+compare with a manually verified baseline.
+
+Verify the meanings and units of the results, source current, winding orientation,
+magnetic material model, boundaries and convergence controls. For linear magnetic
+models, useful internal checks are matrix reciprocity, positive magnetic energy,
+flux linkage and consistency of field energy with `0.5 Iᵀ L I`. These checks do
+not independently prove numerical accuracy. In particular, a mesh stopping limit
+on energy change does not bound the error in every self- or mutual-inductance entry.
+Rebuilding replays CST model history. Settings applied only through transient VBA
+may be reset; persist intended mesh and solver settings in history and verify the
+effective settings in the working copy and solver logs.
+
+## Regression checks
+
+The ordinary Linux suite (`pytest tests/test_cst.py tests/test_windows_applications.py`)
+exercises snapshot changes, local-solver enforcement, native-script dispatch,
+cleanup ordering and submission failure recovery without CST. Shared native
+Job Object, file-lock and fixture-adapter checks run in CI on Windows Python 3.9
+and 3.12, without vendor applications or extra Python packages:
+
+```powershell
+python scripts/check_cst_windows.py
+```
+
+From WSL, use Windows Python and Windows paths for the script. The optional
+`--runner PATH` selects an extracted wheel's standalone `windows_runtime.py` for
+package validation; its sibling adapters must be present. The original
+`cst_windows.py` entrypoint has separate compatibility checks. These checks complement an actual model
+solve; they do not establish solver availability or numerical accuracy.
+
+## Execution and scientific scope
+
+This is an **operator-side trusted Windows process** integration. The selected
+Linux Bubblewrap/PRoot numerical backend neither executes nor isolates CST.
+Windows account access applies; CPU/RAM reservation and monitoring from Linux
+execution pools do not apply to these jobs. This initial bridge serializes jobs
+and enforces a wall deadline, but does not impose Windows CPU or memory quotas.
+
+Jobs are permanently marked `stage="exploration"` and `evidence_eligible=false`.
+They are not registered `lab.run` capabilities and cannot grant evidence eligibility,
+methods approval or claim support. Native agents with operator-authorized host
+tools may use the bridge for preparation and diagnostic work. Sandboxed API agents
+must not use it to bypass their configured execution boundary.
+
+The `cst` skill is shipped and discoverable even without an installation.
+Building a governed Windows numerical worker requires a separately reviewed
+transport, resource accounting and evidence binding; it is not achieved by pointing
+a Linux capability descriptor at a Windows executable.
+
+Consult the documentation installed with your CST release, including the Python
+`cst.interface` and `cst.results` modules and VBA application commands. Public
+upstream references include the [CST solver overview](https://www.3ds.com/products/simulia/cst-studio-suite/electromagnetic-simulation-solvers)
+and [CST Python automation overview](https://3dswym.3dexperience.3ds.com/post/simulia-community/top-10-q-a-articles-for-january-2024_SfCnArS3TgKOnGOVp4Rl8Q).

simjecture-0.6.2/docs/how-to/extend-study.md

@@ -0,0 +1,73 @@
+# Extend an active study
+
+Use **Extend time** beside the study clock in the workspace or experiment monitor.
+Choose +1, +6 or +12 hours, or enter another amount. Review the current and proposed
+deadline before confirming. The dialog also shows active experiment limits.
+
+An extension adds wall time and can consume more model credit and compute. It
+keeps the investigation, agent context, evidence and original scientific scope.
+The running supervisor observes the new clock without a restart and notifies the
+director to update its plan. The protected report reserve stays the same length.
+If report writing has already started, that bounded pass can finish; research
+resumes under the new clock and the extended study gets a fresh closing report.
+
+## From a terminal or external agent
+
+```bash
+simjecture study budget --campaign ./runs/my-study
+simjecture study extend --campaign ./runs/my-study --hours 12 \
+  --reason "Observe the collision and relaxation window" \
+  --request-key extra-relaxation-time
+```
+
+Both commands return JSON. Repeating an extension with the same request key and
+arguments returns its original receipt without adding time again. Reusing that
+key for different arguments is rejected. A separate intentional extension needs
+a new key. Without `--request-key`, the CLI creates a new key for that invocation.
+
+The local operator HTTP API exposes the same operation:
+
+```text
+POST /api/workspace/extend-study
+X-Simjecture-Token: <token from /api/bootstrap>
+Content-Type: application/json
+
+{"campaign":"<registered campaign ID>","hours":12,
+ "reason":"Observe relaxation","request_key":"extra-relaxation-time"}
+```
+
+Browser requests also send `expected_deadline`, so a concurrent change requires
+reviewing the updated preview. Read-only sessions cannot grant extensions. Hosted
+trial allowances remain controlled by the host; this local operator endpoint does
+not provide a way around public usage limits.
+
+`research.json` is the authoritative campaign clock. Its `budget_extensions` array
+preserves each grant, previous and new deadlines, operator identity, reason and
+retry key. The original launch request remains unchanged. External watchdogs
+should read `simjecture study budget` instead of retaining a launch-time cutoff.
+
+## Requests from the research agent
+
+The worker may call:
+
+```python
+lab.request_time(hours=12, reason="The measured runtime needs another collision window")
+```
+
+The director can return `time_request` with `hours` and a concrete `reason`.
+Neither operation grants time. The request appears beside the study clock for
+the operator to review. Agents must plan against the current budget until an
+extension is granted. Use `--request-id` to grant a request from the terminal.
+
+## What an extension does not change
+
+Individual experiments keep their recorded timeouts and execution deadlines,
+including limits clipped to the earlier campaign cutoff. Extending the study
+does not automatically lengthen those jobs. Inspect their limits in the dialog
+and use retained checkpoints for a new recorded experiment when needed.
+
+Ended, cancelled and expired studies use **Continue investigation** with a new
+budget. Extensions do not revive a terminal study or change an accepted result.
+Existing supervisors from releases without live budget support need one ordinary
+pause/resume using updated code; installing a package does not modify a running
+Python process.