Compare designs with a bounded study¶
Use Studies when you want to compare design alternatives or examine how a candidate responds to uncertain inputs. Start from a saved project with its requirements, hardware choices and physical architecture in place. The ordinary Run action still evaluates one project; a study coordinates several cases.
For a new geometry trade, choose Configure → Examples → Design study starter and create an editable copy. It uses a supported two-deck bus with editable width, depth, height and deck spacing. The family stays fixed by that project's geometry binding. Other starters and existing projects keep their saved geometry. The new starter uses its mode schedule for payload time allocation and keeps the original microsatellite mission requirements; changing dimensions or hardware can still produce an infeasible design.
Review a populated plan¶
Open Studies from the Design navigation. Sindri proposes a search strategy, variable bounds, compatible hardware choices, uncertainty assumptions and a budget. The Plan, Design inputs, Uncertainty and Advanced tabs separate setup into focused work areas. The saved project's input catalog supplies the subsystem groups, labels, supported limits and suggested ranges. Geometry, Power, Attitude control, Thermal, Communications, Propulsion and Payload & operations have study adapters. Each input states when it is unavailable and its saved setting remains fixed. Authored requirements and locked choices remain authoritative.
The input summary counts searchable inputs and shows an upper bound on nominal combinations, or Continuous ranges for numerical exploration. It follows your current choices and holds; uncertainty cases are separate. The count does not guarantee that every combination is feasible.
A component selected in the spacecraft is a baseline that a study can compare with compatible alternatives. Fixed in project means an explicit saved constraint applies, including imposed mission EPS hardware IDs. No alternatives means the catalog has no other compatible choice; Unsupported geometry means the current topology has no supported study adapter. Hover or focus a restriction badge for its reason. Hold at the baseline value is a reversible study setting: clear it and review the updated plan to explore that input again. Study holds cannot remove saved project constraints.
A study searches at most 16 inputs at once. Larger projects retain the remaining inputs on baseline holds. Release a hold to include that input and hold another if needed; the catalog remains available across all subsystem tabs.
Project-owned components saved through Configure → Catalog → Customize are available to studies in supported hardware roles. They carry a Project label and retain engineering-estimate evidence; the base component's supplier verification is not inherited. Saving a component does not select it in the configuration or remove mission constraints.
Each study can select up to eight choices per input, including its baseline.
For larger catalogs, Find a model searches all compatible models; deselect
an existing choice before adding another when eight are selected. The suggested
shortlist is a starting point, not the full catalog. Excluded models and their
reasons are captured in the proposal's input catalog. Exports from studies with
project components include their captured definitions and source hashes in
inputs/project-hardware.json.
Thermal compares compatible bus and radiator finishes, including qualified project definitions. Hot and cold design margins can add up to 10 K of reserve above the saved value, within the native 50 K limit. Temperature requirements, environment cases and imposed radiator/heater limits remain fixed. Increased reserve can require more heater power or make a design infeasible; inspect the thermal and power results together. Scenario evaluations retain the selected finishes and installed radiator and heater hardware.
Communications offers compatible radio and antenna models for each RF hardware group. Shared command/telemetry hardware changes together. Add project models in Configure when the packaged catalog has only one compatible model. Transmit power, electrical loads and mass follow the selected model. Contact allocation scale uses 50–100% of the saved contact share while retaining physical contact access, link rates, margins and required data volume. A smaller share can fail data delivery. Optical terminal selections remain fixed in studies.
Propulsion compares compatible thruster assemblies, propellants and tanks, including saved project models. Choose compatible partners together when a model needs a specific propellant or tank. Propellant reserve can add up to ten percentage points above the saved reserve, capped at 50%. Mission delta-v, delivery limits and imposed hardware remain fixed. Undersized tanks remain visible as failed engineering results; scenario evaluations retain the selected hardware and loaded propellant.
Payload & operations offers up to 25% additional payload load allowance. Where the native mode schedule is the time authority, Payload time allocation can add payload time from idle modes while retaining at least half their saved time. Downlink and propulsion time stay fixed. Explicit power timelines, activity windows and scenario duty cycles keep their saved schedule. These controls retain payload quantity, mass, required data volume and performance requirements; they do not estimate improved payload efficiency or science output.
Choose a budget preset, or Custom to enter a case limit (up to 1,024) and compute minutes (up to 240). The compute budget measures elapsed time while the study is active, including execution and comparison work. Pausing stops the clock once active work has settled. Change supported bounds or choices and record a reason for an override, then choose Review updated plan. The Ready to run? panel shows the case and compute limits with the nominal and uncertainty allocation. Changing tabs retains edits; changing settings requires a refreshed plan before confirmation. Source identities and operational limits remain recorded in the plan and evidence export.
For a small first comparison, choose Use a nominal starter at the top. This prepares up to three real designs from the saved project, without uncertainty; it is not sample data and does not start execution. Its settings populate the editor and still require Confirm and start study. A suggested uncertainty interval is an engineering assumption, not measured probability data.
Saving a Configure draft does not change the study's saved source. Save and validate intended project changes before preparing a replacement plan.
| Strategy | Use it for |
|---|---|
| Adaptive | Search within the declared physical choices and bounds |
| Explicit | Evaluate a prescribed list, including repeated observations |
| Grid | Cover every declared combination of axis values |
| Sampled | Cover seeded strata within declared variable domains |
Prescribed experiments must fit their allowance. They cannot report a successful adaptive early stop while leaving declared cases unevaluated. Advanced prescribed JSON is optional; ordinary setup supplies defaults.
Select the Play icon (tooltip: Confirm and start study) once. Confirmation authorizes this plan and allowance. It does not approve a final spacecraft design.
Identify your studies¶
Use the pencil beside a study's title to set a Study name and Tags, for
example “Battery sizing — baseline” with battery and baseline. Names and tags
are saved across sessions and appear in Recorded studies, where you can search
by name, tag or study ID on the current page and filter by status.
The header and list show Started in your browser's local time. Hover over the timestamp for the full date and time zone. This is the original start-confirmation time, including any queue wait; pausing, resuming or extending a study keeps it. Existing studies get their time from the saved event history. Labels are Console annotations: editing them does not change the confirmed plan or captured results.
Follow work and recover¶
The workspace has Results, Execution, and Saved reviews sections. Completed studies open on Results. During execution, meters show case allowance, compute time and active workers. These indicate budget use, not a prediction of percentage complete. Completed work can leave unused allowance.
Open Execution for the case table and links to child runs. Search and status filters in this table apply to the current page; use its pagination for more cases. Run details and activity holds source identities, prior allowances and the event history. Open a child to inspect its normal progress and results, then return to the study. Reloading or returning through saved discovery retains the study identity. All studies → Recorded studies opens a compact list with status and search controls for the current page.
Pause, Play and Stop icons control execution; their tooltips identify the exact action. The green heartbeat beside Running indicates fresh live updates. It stops when updates are stale or recovering and stays still with reduced motion enabled. Pause study stops further admission and settles active work safely. Resume remaining allowance continues the same confirmed allowance; consumed slots and active time are not reset. If a control reports changed state, refresh and inspect the current plan before trying again. A lost response is not proof that another start is needed.
Budget exhaustion, an interrupted case, a blocked capability and an infeasible design have different meanings. Read the stated cause. Continuing after an exhausted allowance requires preparation and explicit confirmation of a new allowance; prior consumption stays recorded. A failed case stays part of the study evidence.
Read the whole population¶
Use the candidate table and two-objective tradeoff plot together. Choose the comparison context before comparing values. Search by candidate or hardware, filter by assessment tier or published front, and sort by score, status or candidate order. These controls apply across the entire selected comparison before pagination. Table options controls visible columns and row density; Rows per page sets the page size. Hide chart gives the table more room. Filtering changes the visible rows and chart marks; it does not redefine the published population or recompute the published front. An attractive point may still have warnings, failed requirements or incomplete evidence. Inspect candidate opens a focused panel with Assessment, Uncertainty, Convergence, and Follow-up studies views. Close it with Close or Escape to return to the same table. Source identities, requirement checks and detailed diagnostics remain available inside expandable sections.
The plot supports keyboard navigation: Tab to a mark, use the arrow keys to move between candidates, and Enter to open evidence. The candidate table provides the same inspection and shortlist actions without requiring a chart interaction.
Keep these evidence meanings separate:
- Nominal evaluates a design at the selected context.
- Fixed-hardware scenarios keep the installed realization fixed while recomputing response across the declared scenarios. All required cases must be accounted for before claiming complete coverage.
- Sizing response may select different hardware. It answers how sizing responds, rather than qualifying one frozen spacecraft.
- Statistical evidence requires explicit distribution assumptions and its declared independent holdout. A bounded corner set or stratified screening sample does not establish a real-world failure probability.
Read coupling contributions and residuals alongside requirement assessments. Missing results or nonconvergence do not become passing scores. Different contexts or evidence tiers do not produce one automatic overall winner.
Save a review and record a choice¶
Select up to four candidates in one comparison context, give the shortlist a name, and choose Save and review shortlist. The saved review retains the full published population as its source; the shortlist is a review view. Reopen it from Saved reviews in the study workspace.
Inspect each candidate's requirements and warnings. Select an eligible candidate, accept its displayed warnings when required, and provide your rationale before Confirm selection. Failed or incomplete candidates remain inspectable but cannot be selected as an eligible design. The recorded choice reopens with its original evidence and warning acceptance. Selection uses the same review workflow as candidate evaluation.
Choose Export complete publication to download the captured results in one action. The ZIP captures the plan, publication and review evidence with a checksum manifest. It can omit large child artifacts while retaining their exact references; read the export inventory before transferring it as a complete engineering record.
Local CLI workflow¶
The CLI uses the same local control store and confirmation rules. Use the same store and trusted project root as the Console when moving between them:
sindri study service status
sindri study propose my-project/study.yaml --nominal
The second command prepares a small nominal comparison without executing it.
For the populated adaptive plan with scenarios, omit --nominal. Use --preset,
--strategy, --setup, --spec and --overrides only when you want to replace
those suggestions. sindri study propose --help describes the bounded inputs.
Review the returned proposal and copy its actual ID, digest and revision into
shell variables PROPOSAL_ID, PROPOSAL_DIGEST and PROPOSAL_REVISION. Then:
sindri study start "$PROPOSAL_ID" --digest "$PROPOSAL_DIGEST" \
--revision "$PROPOSAL_REVISION" --key my-reviewed-study-start
Keep that request key if recovering the same confirmation after a lost response.
The command executes locally unless --queue-only delegates dispatch to an
attached Console. Set STUDY_ID from the returned receipt to inspect it:
sindri study status "$STUDY_ID"
sindri study cases "$STUDY_ID" --limit 20
sindri study events "$STUDY_ID" --limit 100
sindri study export "$STUDY_ID" --output study-review.zip --key my-study-export
sindri study verify-export study-review.zip
Export uses a new destination and never overwrites an existing file. For a
nondefault store, put --store path/to/control.db immediately after study on
every command. Pause, resume and cancellation require the current plan digest
and revision shown by status; inspect them before issuing a control.
Studies are available in new workspaces. Upgrading a workspace preserves any explicit availability choice made by its local owner; paused studies stay paused.
The local owner can use sindri study service disable to stop study admission
safely, and enable to restore availability. The environment setting
SINDRI_STUDIES_ENABLED=false remains a hard disable. Re-enabling does not
silently resume existing work; evidence reads and ordinary runs remain available
where independently healthy.
Older projects and saved evidence¶
Current study.yaml uses contract_version: 2.0.0. Mission and spacecraft
contracts keep their own versions. Embedded sweeps are no longer executable
project fields, even when empty. On a separate copy of an old authored project,
update the study version, remove the embedded declaration, and prepare its
experiment through Studies. Review the resulting variables, sample count, seed,
contexts and assessment meaning; do not assume every old adapter is supported.
Existing supported result bundles, comparison records and historical study snapshots remain read-only. Keep their original bytes and version identities; do not edit them to make them look like current inputs. See bundle compatibility and native ADCS sizing response.