Reports and workflows
Assistant answers do not evaporate when the conversation scrolls. Any turn that ran tools and produced charts or insight cards emits an analysis artifact, and artifacts accumulate in the Reports workspace, where they become structured report documents, a reusable chart library, and exportable deliverables. Workflow templates close the loop in the other direction: they turn recurring operator reports into parameterized prompts that drive the same deterministic tools.
For the console UI around the Reports workspace itself, see Reports in the console.
From answer to artifact
As covered in How it works, each analytical turn packages its outputs: the one-line action summaries, the rendered charts, the insight cards with their full audit trails, and the names of the tools that ran. Plain conversational turns stay chat-only; only turns with real computed content earn an artifact.
The insight card is the unit of evidence. It carries the headline value with units, the scope line (operator, filter state, timeline mode, timestamp), the method, the exact calculation, sample sizes and exclusions with their rules, the table of participating assets, and any low-sample or truncation notes. When an artifact streams into a report, the cards become the report's Findings and Method sections, so every number in the document is traceable to the tool run that produced it.
Report documents
A report document assembles artifacts into an operator-grade deliverable: a posture summary, findings backed by insight cards, charts, and a method section listing the tools, windows, assumptions, and exclusions behind every figure. Because the source engines are deterministic, a report is reproducible: the same questions at the same sim time with the same filters regenerate the same numbers.
The chart library and regeneration
Every chart the assistant renders carries regeneration metadata pointing back at the exact tool call that produced it: the tool name, the arguments, and the scope. Saved charts form a library that can be re-run later against the current scene. Regeneration is honest by construction: it re-executes the producing tool, so a regenerated chart reflects the scene as it is now, and the deterministic engines guarantee that identical inputs still produce identical outputs.
Exports
Reports export as shareable documents, including PDF, for the audiences that never open the console: customers receiving a gateway proposal, regulators receiving coverage evidence, leadership receiving a quarterly summary. The exported document keeps the findings, charts, and the method section, so the audit trail travels with the numbers.
The ten workflow templates
Workflows are reusable report templates. Each one is an operator-grade instruction that names the deterministic tools to run, with a small set of parameters. Running a workflow renders the template into a prompt and feeds it to the chat composer; from there it is a normal assistant turn, with the step trace, insight cards, and artifact capture that implies. Every number in the output comes from a tool run, never from memory.
| Workflow | What it produces | Key parameters |
|---|---|---|
| Daily fleet health | Posture, degraded assets, utilization and latency trends for a shift handover. Runs get_fleet_summary, snapshot statistics, a 24 hour trend, a breakdown by node type, and a busiest-assets chart. | date, fleet |
| Weekly capacity and utilization | Where capacity is consumed, saturated (utilization above 0.8), or stranded across the week, with breakdowns by node and link type and week-over-week deltas when prior data exists. | week start, fleet |
| Coverage comparison | Two constellation designs scored on identical windows via create_orbit_scenario, with reset_scenario between runs, presented side by side with a recommendation and an explicit tiebreak rule. | design A, design B, window, region |
| Gateway proposal | Customer-facing comparison of candidate gateway sites: the full ranked table from compare_candidate_sites, optimizer picks with marginal gains, and a heavy-rain re-run (25 mm/h) showing which sites degrade most. | customer, candidate sites, objective |
| Link budget validation | One link budget walked end to end (EIRP, free-space path loss, receive gain, noise, margin) and reconciled against stored telemetry, flagging model-versus-telemetry disagreement rather than averaging it away. | link, required margin |
| Launch readiness review | Go/no-go checklist: injection orbit coverage, first-contact opportunities, longest gap in the first day, and TT&C station geometry via compare_candidate_sites. | mission, launch date |
| Anomaly investigation | Evidence-driven timeline for one asset: trend onset, before/after scene comparison via seek_time, and a peer breakdown to separate common cause from isolated fault. Ends with root-cause confidence and the telemetry that would confirm it. | asset, window |
| Executive dashboard | Leadership summary: headline metrics, composition donut, trend direction, incidents, and the decisions being asked for, written for a non-operator audience with every figure traceable to a tool result. | period |
| Regulatory coverage | Reproducible service-availability evidence for a country filing: per-station contact scores over identical propagated tracks, gap remediation options, and the exact window, step, and elevation mask. | country, window |
| Orbital trade study | N constellation configurations ranked on one metric with identical scoring, sensitivity notes on the elevation mask and window, and a flag on any advantage within noise of the runner-up. | configurations, primary metric |
Parameter handling
Required parameters that are missing are flagged before the workflow runs. Optional parameters left unset render into the prompt as an explicit instruction: "(not specified; choose a sensible default and state it)". That keeps the contract intact: the agent may choose defaults, but it must disclose every choice as an assumption in the report.
Workflows are honest about data limits
The templates bake the honesty rules in. The weekly capacity workflow tells the agent to report the exact coverage the trend tool returns and say plainly if the stored window is shorter than a week. The comparison workflows demand identical windows, steps, and elevation masks across runs, and require synthetic runs to be labeled synthetic. A workflow cannot make the agent fabricate a number the tools did not produce.
Where to go next
- Response contract: the rules every workflow-driven answer still follows.
- Memory and sessions: where reports, charts, and scenarios persist.
- Examples: a worked mission readiness report and a PDF export walkthrough.