Replaying estimated states
Once the estimated-state lake exists (see 11-estimating-gb-history.md), three surfaces consume it: the desktop app, the web console, and DataExplorer. All of them carry the estimate's fit quality alongside the state - a frame of estimated history is never shown without the quality of the fit that produced it.
The desktop app
Start the bridge server first (13-the-gda-bridge.md),
then launch the app and pick the Live bridge entry in the model picker.
The entry appears whenever a bridge endpoint is configured - the default
(ws://127.0.0.1:8770/ws/bridge) is set in
%LOCALAPPDATA%\GridSim\settings.json (BridgeEndpoint; blank it to hide the
entry).
In bridge mode the transport bar shows three pickers:
- Zone - which case to stream: the six DNO zones,
GB full, orGB spine - GSP-level(the estimation target; pick this for estimated replay). - Start - a historical anchor preset (2 days back ... 30 days back).
Known limitation: the no-forward-inference horizon was widened to
169 hours (7 days + 1 hour), so the shorter presets (2 days ... 1 week) now
fall inside it and the server rejects them (
HORIZON REJECTIONon the status pill). Pick 2 weeks ago or 30 days ago until the presets are reworked. - SRC - the reconstruction the stream replays:
PRESOLVED- the demand-scaled re-solve lake; per-bus state varies each tick, deterministically derived from the model.ESTIMATED- the measurement-fitted WLS lake; every frame was fitted to that period's real telemetry and carries a per-frame quality block.LIVE- a static-network solve with only the system scalars moving.
Switching any picker restarts the stream on the same connection.
The chi-square status pill
On an ESTIMATED stream the bridge status pill shows, per frame:
ESTIMATED * gb-spine * chi-square PASS * J=201.3 n=948 removed=1
- chi-square PASS/FAIL - whether that period's fit was consistent with its declared measurement uncertainties (pill green on PASS, red on FAIL). A FAIL frame still displays, but treat its numbers as uncorroborated until you know why the test failed (see 10-state-estimation.md).
- J - the weighted sum of squared residuals, the number the chi-square gate judges.
- n - how many measurements the period's fit consumed.
- removed - bad-data removals in that period (omitted when zero). Persistent removals at the same location across periods are a measurement or model defect, not noise - the closure report's outlier league tracks exactly this.
Launch arguments
Pre-select bridge mode from the command line - useful for scripted demos:
dotnet run --project src/GridSim.Wpf -- --zone gb-spine --back 14 --bridge-source estimated
--zone matches a case name substring, --back N picks the N-days-back start
preset, --bridge-source presolved|estimated|live pre-selects SRC.
The web console
Configure the bridge endpoint (the GridSim:BridgeEndpoint configuration key
- e.g. via
appsettings.jsonor theGridSim__BridgeEndpointenvironment variable), start the web app, and the home screen shows an Estimated bridge card ("GDA lake * WLS-estimated states (chi-square-gated)") alongside the Live bridge card. It opens:
/dashboard/bridge:gb-spine@estimated
The URL scheme is bridge:<case>[@<source>] - the optional @source suffix
picks the reconstruction: presolved (the default when omitted), estimated,
or live. Any bridge case works, e.g. /dashboard/bridge:gb-full@presolved.
DataExplorer
The lake's quality series are charted in GDA's DataExplorer. From
D:\Work\GDA\v1:
python DataExplorer/server.py
then open http://localhost:8077 (change with --port). The State
estimation group carries two charts, both defaulting to the last 7 days and
scrollable across the whole backfilled range:
- Estimation quality (gb-spine) - per-period objective J against the chi-square gate (left axis), iterations and bad-data removals (right axis). J above the gate = inconsistent fit; watch for J tracking the gate closely (sigma pressure) or removal spikes (a defective source).
- Solve vs estimate divergence (gb-spine) - max deltaVm (pu) and deltaVa (deg) on the left axis, max deltaflow (MW) on the right: where the deterministic solve route and the measurement-fitted route disagree. Persistent divergence localises model or dispatch error.
See also
- 10-state-estimation.md - what J, chi-square and removals mean
- 11-estimating-gb-history.md - building the lake these views read
- 13-the-gda-bridge.md - the server, sources and faults
- 04-the-desktop-app.md - the transport bar and pickers in general