Dashboard Guide¶
Prerequisites¶
Requires Streamlit + Plotly.
Product posture¶
Dashboard = inspect, MCP = author
The dashboard is optimized for inspection, comparison, and visualization. Use MCP-managed workspaces and agent prompts for scenario authoring, then use the dashboard to explore and compare results.
- Use the dashboard to browse runs, compare variants, inspect parity, and review visualizations.
New Runkeeps advanced manual authoring available, but hides it behind an explicit toggle by default.
Launch¶
Fast smoke flow (fp run --backend both)¶
fp run examples/baseline_smoke.yaml --backend both --with-drift --output-dir artifacts/dashboard_smoke
fp dashboard --artifacts-dir artifacts/dashboard_smoke --port 8501
fp triage fppy-report artifacts/dashboard_smoke/<scenario>_<timestamp>
fp triage parity-hardfails artifacts/dashboard_smoke/<scenario>_<timestamp>
fp run --backend both creates a top-level run directory with convenience copies plus a nested parity run that contains canonical per-engine artifacts:
artifacts/dashboard_smoke/<scenario>_<timestamp>/
parity_report.json
PABEV.TXT
parity/<scenario>_<timestamp>/work_fpexe/PABEV.TXT
parity/<scenario>_<timestamp>/work_fppy/PABEV.TXT
Pages¶
Parity¶
- Input artifacts:
- required:
parity_report.json,work_fpexe/PABEV.TXT,work_fppy/PABEV.TXT - optional:
triage_hardfails.csv(orwork_fppy/triage_hardfails.csv),triage_hardfails_summary.json,work_fppy/support_gap_map.csv,work_fppy/support_gap_top.md,work_fppy/fppy_report.json,work_fppy/triage_summary.json,work_fppy/triage_issues.csv,parity_regression.json - Shows top-level parity status (
status,exit_code,hard_fail_cells, max/median/p90 abs diff). - Reads report metadata fields
schema_versionandproducer_versionwhen present inparity_report.json. - Backward compatibility: legacy reports missing these fields are still loaded; dashboard fallbacks remain
unknown/n/awhere metadata is unavailable. - Renders variable x period abs-diff heatmap with start-period control.
- Surfaces hard-fail samples with filter + jump-to-series workflow.
- Compares fp.exe vs fppy series for selected variable with abs-diff overlay.
- Displays unsupported-command impact and example statements from
fppy_report.jsonwhen present. - Highlights solve controls and mode signals (
eq_flags_preset, source,eq_use_setupsolve, iteration stats). - Detects zero-filled forecast-window variables and flags likely 1-iteration trap conditions.
- Provides direct downloads for triage/support-gap artifacts and regression report when present.
- Includes an agent handoff panel for parity follow-up.
Run Panels¶
- Select multiple completed runs for side-by-side small-multiple charts
- Filter runs by backend, bundle, or scenario name
- Choose variables and transform modes (level, % change, level change, % of denominator)
- Compare runs via diff or % diff overlays against a reference run
- Save and load named chart presets for quick recall
- Export multi-panel PNG snapshots
- Reads LOADFORMAT output and derived series when available
Home¶
- Run listing and artifact discovery
- Quick counts for total runs and runs with outputs
- Entry point for navigation to other pages
- Good landing surface after agent-created runs complete
Compare Runs¶
- Select two completed runs
- Compute run comparison metrics and top movers
- View delta bar chart and side-by-side forecast comparison
- Download comparison CSV
- Includes an agent handoff panel for explanation or visualization follow-up
New Run¶
- Defaults to an agent-first handoff flow
- Keeps advanced manual authoring behind an explicit toggle
- Managed workspaces remain available when direct editing is needed
Data Update¶
- Build a new runnable
FM/bundle underartifacts/by runningfp data update-fred - Shows
data_update_report.jsonfor quick verification
Equation Graph¶
- Upload or point to
fminput.txt - Build dependency graph summaries
- Inspect upstream/downstream nodes and export adjacency view
Equations¶
- Browse all behavioral equations, identities, and generated variables
- Filter by section (equations, identities, generated vars) or search by variable name
- View equation specifications in readable format with expandable details
Tweak Scenario¶
- Pick a completed run as a starting point
- Adjust variable overrides (add, remove, or change values)
- See a diff summary of what changed before running
- Re-run the tweaked scenario and immediately compare against the baseline
- Save tweaked config as a new YAML file
Historical Fit¶
- Select a completed run with estimated equations
- View summary metrics: equation count, average R², average Durbin-Watson
- Identify weak equations (low R²) and DW-flagged equations
- Browse a styled table of all equation fit metrics with color-coded R² and DW cells
- R² bar chart visualization (requires matplotlib)
Dictionary¶
- Search variables and equations by name, code, or description
- Inspect variable definitions, categories, and source-map links
- Browse equation specifications with cross-linked variable details
- Manage dictionary overlays: view, edit, and save per-scenario overrides
- Export equation detail tables as CSV
- Auto-discovers overlay paths from selected runs and shared extensions
Sensitivity Analysis¶
- Pick a completed run as the base case
- Sweep one override variable across a min/max range with configurable step count
- Set sweep method (
CHGSAMEPCT,SAMEVALUE,CHGSAMEABS) - Run sensitivity and see fan chart plus response table for tracked variables