Skip to content

Sweeps and specialist exchange

Use a specialist workflow when the question needs additional cases, external environment evidence or a 42 execution. Start with a working native study; the first Console run covers that baseline.

Question Workflow Before starting
How sensitive is ADCS to the selected inputs? Native ADCS robustness Verified nominal case, physical input bounds and seed
Can I use an externally calculated environment? External environment product A compatible exchange product and its source evidence
How does a captured controller case behave in 42? 42 handoff and local replay A generated handoff; a compatible installation for execution
How do I report costs in another currency? Reporting FX preparation An explicit rate snapshot and freshness policy

Keep the required resources and external tools available before execution. These workflows extend the study's evidence within their documented model limits; a successful exchange does not establish physical validation.

Native ADCS robustness

Native ADCS sizing response now uses the shared Design Studies workflow. First complete a nominal case in a study. From its native evidence, prepare an explicit, grid or sampled ADCS experiment; confirm the resulting plan and budget before execution. Hardware may change between sizing cases. This is neither a fixed-hardware qualification nor a controller-dynamics simulation.

The local CLI accepts an AdcsProposalRequest JSON file through sindri study adcs adcs-request.json. Populate nominal_study_id and nominal_case_id from study status and study cases. For four seeded samples:

{
  "nominal_study_id": "study.replace-with-recorded-id",
  "nominal_case_id": "case.replace-with-recorded-id",
  "strategy": "sampled",
  "seed": 101,
  "prescribed": {"sample_count": 4, "sampling": "latin_hypercube"},
  "preset": "quick"
}

These IDs are placeholders, not an executable example. Inspect the returned populated physical variables, source binding and workload. Confirm with study start using the returned proposal ID/digest/revision and --parent-study set to the returned parent_study_id. There is no hidden nominal solve or second sweep runner. The ordinary sindri run command now performs only the single project run.

Read ADCS evidence limits. Preserve the seed, exact plan, completed-case count and failed/partial cases. The previous embedded sweep aggregate and its pass-fraction thresholds are historical records; new study assessments retain individual requirements and contexts.

External environment product

A specialist-generated environment product is ingested as local project data. Declare it in the study resources, then set the mission's environment.external_environment_resource_id to that resource role. The product must use the supported exchange schema and known evidence source.

Validation checks content digest, mission identity, target orbit, dated coverage when applicable, mode IDs/fractions/attitudes and selected-case applicability. An incompatible product fails resolution; it does not silently fall back to a native calculation. No automatic external-service fetch or generic import command is implied.

Export the current schema suite to inspect specialist contracts:

sindri export-schema --out-dir my-schemas

Coordinate the exact exported exchange schema and resource version with the specialist. A matching envelope is necessary but not sufficient for physical validity; preserve the external method, source and limitations.

42 handoff and local replay

Enable a supported ADCS 42 handoff configuration in the study, execute it, and inspect its indexed handoff/case IDs. Then validate the generated handoff:

sindri adcs-42 validate my-run

Select an independently installed compatible executable once, then run one captured case without repeating native sizing or CAD:

sindri adcs-42 configure --executable /path/to/42
sindri adcs-42 doctor --controller
sindri adcs-42 run my-run --case my_case_id --workspace-root replay-workspaces --timeout 3600

The command creates a canonical child run, separate from the immutable source. Its printed run directory contains indexed execution records, telemetry and reports. Repeating the command creates a fresh attempt. The optional --executable flag must match the configured installation, not override it. sindri resume CHILD_RUN recovers interrupted attempts only when their captured inputs and execution fingerprint still match; terminal attempts require a new run. Preserve the executable revision/digest, execution result, stdout/stderr, parsed output and report references. Validation does not execute 42; fixture output does not prove a real replay. See the 42 boundary.

New Sindri-controller handoffs capture Sindri_Controller.dat alongside the controller contract and fixed bridge source. The runner verifies bridge identity and passes that case's parameters to 42. Changing gains or a mode timeline no longer requires rebuilding the bridge. An older compiled-per-case installation requires an explicit bridge update; incompatible handoffs must be regenerated. Neither validation nor execution downloads or builds 42.

Running 42 in Console

  1. Open Configure → Run, enable 42 verification and select one to eight Cases.
  2. To prepare inputs without executing 42, open Execution policy and select Prepare handoff only (do not execute). Otherwise, the selected cases run after native sizing.
  3. Set the duration, integration step, output interval and per-case timeout for the intended study. Select Save 42 configuration, then validate.
  4. For execution, check Local installation as described below, then use the normal Run action. Follow the native and 42 stages in the run monitor.

Controller and timeline details remain editable in the full YAML editor. Require successful 42 execution is the default execution policy. Turning it off does not relax input/runtime integrity, cancellation or native-run failures.

Set the executable and build receipt under Local installation. These paths are machine settings, not portable study inputs. Check startup probes the installed executable without running a simulation. Runtime identity is checked again at launch and execution, so an earlier readiness check is not a guarantee.

For local use, open Console using the private URL printed by sindri console. Its per-launch credential authorizes edits and execution; a bare URL does not authorize them. Do not share the launcher URL. After restarting Console, open the newly printed URL. The browser removes its credential fragment from the address bar and keeps it only for that browser session, not in project files or run bundles. For a server on another machine, use its configured remote Console connection. The 42 installation belongs to that server, not the computer running the browser.

42 execution currently requires a Linux/WSL installation controlled by the Console process owner or root. Executable, receipt, Model assets, settings and their parent paths must not be writable by other users or groups or use symlinks. Keep the installation in a protected local directory, not a shared writable folder. Build receipts verify identity, not the trustworthiness of arbitrary code.

Sizing and budgets > 42 verification discovers indexed executions automatically. Run captured 42 cases creates a fresh child run from a finalized run's captured hardware, controller and timeline, without repeating sizing or CAD. Use the existing monitor to cancel or inspect its logs. The original run and older attempts remain unchanged; reruns appear alongside the source's recorded executions. Select an execution explicitly when several are available. Link 42 results remains available for compatible older external captures.

Continue with interpreting a 42 execution for telemetry, frames and acceptance state. Use run recovery for an interrupted attempt rather than treating it as a completed result.

Execution success, verified provenance and engineering acceptance are different states. A plot or a successful process does not establish that pointing or actuator requirements pass. Failed captures retain available diagnostics, not a success claim.

Optional reporting FX preparation

This is an explicit network preparation step, separate from offline solving:

sindri cost refresh-fx --output my-project/fx/ecb-latest.yaml --currency GBP

The command includes EUR and canonical USD as well as requested currencies. Existing files require an explicit --force; failed refresh preserves the old file. Declare the snapshot as a project resource, select snapshot mode and its resource role under policies.cost.reporting_fx, and choose a maximum age. Validate before running. Future/too-old snapshots fail rather than falling back to annual averages. Replay retains the accepted captured rates without fetching new ones or reinterpreting freshness.

This changes reporting FX only. It does not provide intraday/procurement prices or improve underlying cost authority; see cost and authority.