Both VirtualTagEngine and ScriptedAlarmEngine share a pattern: the
BuildReadCache helper iterates the script's declared input set, reading
from _valueCache with a fallback to _upstream.ReadTag. When an upstream
tag hasn't yet delivered its first subscription push, ReadTag returns a
DataValueSnapshot with a null Value and BadNotConnected quality. User
scripts then cast `(double)ctx.GetTag(path).Value` unconditionally and
throw NullReferenceException — once per evaluation tick until the cache
fills, spamming the log with identical stack traces. The existing catch
block recovered (kept the prior state) but didn't silence the churn.
Add AreInputsReady(cache) to both engines: return true only when every
entry has a non-null Value and a non-Bad StatusCode (Good + Uncertain
are both considered ready). Skip script evaluation when the check
returns false — the engine holds the prior state (alarm) or the prior
snapshot (virtual tag) until upstream delivers. Eliminates the cold-
start NRE spam at root without changing the script-engine contract.
Also: fix $changeLines.Count in test-galaxy.ps1 — PowerShell's
Set-StrictMode -Version 3.0 errors on .Count when Where-Object returns
0 or 1 items. Wrap in `@(...)` to force an array; same pattern the
sibling _common.ps1 already uses in Write-Summary for the same reason.
Task #112 — the Galaxy live E2E now passes 3/7 stages (probe + source
read + reverse-bridge-ACL). The remaining 4 stages (virtual-tag,
subscribe-sees-change, alarm-fires, history-read) are deployment-
specific: MoveInBatchID is idle in this Galaxy + its AccessLevel blocks
writes + it's not historized. Cold-start behaviour is now correct, so
once the seed points at a live attribute those stages should light up.
Tests: 36/36 VirtualTags.Tests + 47/47 ScriptedAlarms.Tests green.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
69e1d320ac