Costs: what a finding costs

A finding's cost is one of four kinds of figure, and the kinds are never added together. A plant manager should be able to check every dollar against their own records, so each figure says what it is and the finding page shows the arithmetic behind it. Give the collection's context model a costs block, and an asset its own rate where it differs:

"costs": { "currency": "USD", "perMinuteDown": 40, "perScrappedPart": 3.5,
           "products": [ { "name": "Cap 38mm", "perScrappedPart": 0.8 } ],
           "runningStates": ["Running"],
           "attribution": [ { "family": "drift", "asset": "ct1", "downtimeReasons": ["cooling water"] } ] },
"assets": [ { "id": "press4", "name": "Press 4", "costPerMinuteDown": 95 } ],
"relations": [ { "from": "ct1", "to": "press4", "type": "feeds" } ]

The editor's collection form holds the costs rates and the asset form holds costPerMinuteDown (section 7.1); attribution is edited in the model's JSON.

The four kinds

1. Attributed: the stops your own records tie to the problem. When costs.attribution has an entry for the finding's kind (its family: drift, band, setpoint, first-out-alarm, an mes- check, ...; optionally only on one asset or one tag), the finding is charged the downtime whose reason is one of downtimeReasons (compared without case, as the state-periods reason column carries it), on its own machine and every machine its feeds relations reach (a cooling tower feeds the presses it cools), during the finding's active span (the start and end its evidence names, not the whole time it stayed open). Each machine's minutes are priced at its own costPerMinuteDown, else perMinuteDown. It reads "USD 719 attributed", the figure "11 stops, 20.6 min". When such a finding has mapped stops, excess over normal is not added to it (excess scrap on those machines is not counted there); when it has none, it is priced as excess instead and says so. Two attributed findings on one source with overlapping spans each price their own span and each names the other as an overlap.
2. Above normal (estimated): for any other process finding on a machine: only what is above that machine's own usual rate.
3. Scrap, per product: (reject rate in the finding's hours − the same product's reject rate in the baseline) × parts made in the finding's hours × the product's price (its products entry, else perScrappedPart; the arithmetic names which). Parts made = good + rejected from the counts shape; a count bucket the window cuts counts in proportion to its overlap. A figure can therefore differ by a part or two from a raw shot count: on Whitfield, A13's window cuts the hourly buckets at 4:24 PM, so it shows 190.6 WD-4510-VB rejected of 1,671.6 made where the shot table counts 191 of 1,670.
4. Downtime: (unplanned minutes per running hour in the finding's hours − the baseline's) × running hours × the per-minute rate. Running hours are the hours less every period that is not running (planned or not); unplanned minutes are periods whose planned is not true and whose state is not in runningStates.
5. The baseline is the 28 days before the finding began.
6. Stops the map ties to a cause are left out of the downtime, in the finding's hours and in the baseline: every stop whose reason a costs.attribution entry names for this machine (an entry with an asset reaches that machine and every machine it feeds; one without reaches all). This is decided by the map alone, and the arithmetic says "N stops with reasons mapped to a cause are left out".
7. Guards: "not priced" beats an invented number. A product is priced only with at least 200 parts made in the finding's hours and 1,000 in the baseline; downtime only with 8 running hours in the finding's hours, 48 in the baseline and 5 unplanned stops there. The difference must be at least 3 standard errors (scrap: a two-proportion test; stop minutes: a compound-Poisson test using the baseline's stop rate and stop lengths). The label names the reason that binds: a product well above normal but made in too few parts is "Not priced: too few parts"; a smaller difference is "Not priced: could be chance"; a rate at or below normal is "No excess over normal"; too few parts or hours otherwise is "Not priced: too little data". Rates in the arithmetic are shown to the cent so every line multiplies out.
8. Each finding is priced on its own: its own window against its own baseline, whatever other findings exist, so its figure is the same on the list, a filtered list, its page, the digest and Ask. When another finding that is open, judged a real issue, or resolved from open or a real issue (its resolve note records the status it had before; one resolved from hidden or suppressed is never named), on the same machine or one related by feeds, has an attributed or above-normal figure over an overlapping window, the arithmetic adds "Overlaps with '