Add Bento 1 fresh bentonite consumption

2026-09-06 13:29:06 +02:00
parent ed0e20a04b
commit be8d76507a
11 changed files with 430 additions and 14 deletions
@@ -0,0 +1,60 @@
+# Bento 1 fresh bentonite consumption
+
+`config/bento1-material-consumption.yaml` defines
+`bento1-fresh-bentonite-consumption`. This KPI estimates **fresh bentonite
+consumption** from the two fresh-spreader setpoints in g/m². The third spreader
+uses recycled/recovered bentonite and is deliberately excluded; it is neither
+missing data nor estimated. This is not total bentonite deposited on the product,
+and setpoints are not a measurement of actual mass flow.
+
+The generic `area_application` source sums the configured application signals
+and converts them to kg/h using `sum(g/m²) × nominal_width_m × speed_m/min × 60 / 1000`.
+The existing previous-value integrator then accumulates kilograms only while
+speed is strictly greater than 0.3 m/min. `direct_mass_rate` remains the default
+for existing configurations, including K7. The persistent state schema and
+production-order bootstrap/resume behavior are unchanged; each disjoint run
+starts with a new sample baseline while retaining order totals.
+
+Width comes from the existing ERP workplace status adapter and generic nominal
+width parser. The ERP order must format to the ENLYZE order using
+`Bento 1-{production_order}`. Missing/mismatched ERP context or an unparseable
+width rejects the polling cycle without saving state. ENLYZE product metadata is
+not used for width. The current ERP context supplies width for all runs of that
+same order; this assumes the order's article width stays constant. Historical
+orders no longer present in current ERP workplace status require a separate
+historical context source for offline replay.
+
+Run with:
+
+```sh
+production-analytics run material-poll \
+  --config config/bento1-material-consumption.yaml \
+  --erp-secrets-file secrets/erp.env
+```
+
+`erp_workplace: Bento 1` assumes that exact ERP workplace identifier; verify it
+against the deployment's workplace view. The 20-second maximum sample gap is a
+generic policy shared with K7, intended for a nominal 10-second sampling cadence.
+One missing sample is tolerated: the resulting 20-second interval is integrated
+using the preceding rate and gate. Gaps greater than 20 seconds are not
+integrated, so prolonged or repeated gaps can undercount consumption.
+Missing/nonfinite values from either configured fresh spreader reject the cycle.
+Intervals exceeding the maximum gap and unobserved run tails are not inferred.
+
+Before implementation, the calculation was independently validated manually
+against the same three real ENLYZE signals over the two runs below, yielding
+approximately **147021.9 kg** of fresh bentonite for `Bento 1-12026000814`.
+The post-implementation live replay could not be repeated because ENLYZE
+connectivity was temporarily unavailable (DNS resolution failure), not because
+of a known implementation issue. Exact production-path replay remains a
+follow-up verification rather than a blocker for this milestone.
+
+The regression uses article 180305, description `Bfix NSP 5300,     5,00 x 40 m`
+(width 5.00 m), and the supplied boundaries of runs
+`b675818c-4636-4976-be6c-3e0e24da9e8a` and
+`25d67424-8346-42d4-b844-75b4097382bd` for order `Bento 1-12026000814`.
+It constructs synthetic samples from the supplied aggregates: 1256.0 active
+minutes, 31294.6 m² and 4698.0 g/m², yielding approximately 147021.9 kg.
+It verifies bootstrap across both runs and persisted resume without counting the
+inter-run gap. It is an aggregate plausibility regression, not an independent
+replay of recorded signals; raw reference samples were not supplied.