This plant's own quality numbers¶

Thirteen numbers in this MES used to be literals in the source: where a process stops being capable, how many readings a control limit needs, how fine a gauge has to be, how many serials a certificate prints, how deep the packaging goes. Each one was a judgment somebody made once, and most of them had a comment beside them arguing for the number rather than stating it — which is what a judgment call sounds like before anybody calls it configuration.

They are this plant's now. They are seeded from [quality] in the plant pack, owned by the plant's database from then on, and edited on Quality › Configuration by somebody holding quality.define — with the value the plant is actually running on beside each one. They follow one rule above all the others:

The literal that was in the source is the shipped default, unchanged. A plant that writes none of these keys behaves exactly as it did.

That is deliberate and it is the whole rollout plan. The setting lands, no plant behaves differently, and the plants that want something else start asking. Making a number configurable and moving it would be two changes, and the second one needs its own argument.

The keys¶

Key	Ships	What it decides
`hold_rules`	`[1, 2, 3, 4]`	Which SPC rules raise a quality hold, from 1 to 5. Every rule is drawn on the chart whatever this says — decision 0036. Rule 5, a moving range beyond its upper limit, is the one the default leaves out: it is drawn and recorded on every plant and holds only where a plant adds `5`
`major_rules`	`[1]`	Which of those open a major non-conformance rather than a minor one. The words come from this plant's severity list
`cpk_capable`	`1.33`	The Cpk at or above which this plant says capable
`cpk_marginal`	`1.0`	The Cpk at or above which it says marginal rather than not capable. Must be below `cpk_capable`
`spc_min_points`	`12`	The fewest points control limits are drawn from — readings on a chart of individuals, samples on an X-bar and R chart (the sampled case). The pallet certificate prints whatever this says
`spc_history`	`200`	How far back a chart and the rules look
`gauge_ratio_adequate`	`10`	How many times finer than the tolerance a gauge must resolve to be called adequate
`gauge_ratio_floor`	`4`	Below this a gauge is too coarse to judge the tolerance at all. Must not be above `gauge_ratio_adequate`
`gauge_default_interval_days`	`365`	The calibration interval a new gauge gets when nobody says otherwise
`coa_serials_listed`	`200`	How many serials a pallet certificate prints before it says how many more there are
`serial_digits`	`6`	How many digits a generated serial carries after its prefix
`nc_code_prefix`	`"NC"`	What a non-conformance is called: `NC-00017`, or `NCR-00017`
`containment_max_depth`	`6`	How deep this plant's packaging goes

[quality]
# A plant that inspects hourly, grades a four-of-five trend as major, and
# follows ANSI Z540 rather than AIAG.
spc_history = 60
major_rules = [1, 3]
gauge_ratio_adequate = 4
gauge_ratio_floor = 2

fsmes pack check reads every one of them offline, and the Configuration page refuses the same values in the same sentences — one wording for one rule, so a screen cannot accept what a pack file cannot hold. It refuses a number that would leave the thing it decides unable to decide anything — control limits from one reading, a serial with no digits, a Cpk bar that makes marginal unreachable — and it refuses nothing else. A plant that wants twenty-five readings behind its limits is answering its own question and is not being second-guessed.

What stays the product's, and why¶

Not everything near these numbers is a plant's to answer. The test is decision 0035's: ask what breaks if two plants answer differently. If nothing outside the plant breaks, it is the plant's. If a number, a topic or an API field would mean something different at the two plants, it is the product's.

The five SPC rules, their numbering and their windows — the four Western Electric rules on the individuals chart and rule 5 on the moving range. A plant that renumbered them would publish SpcSignal.rule = 3 while meaning something nobody else means by rule 3.
The range constants, d2 and D4 at a subgroup of two. 1.128 and 3.267 are what the method is; a plant that changed them would be drawing a different chart under the same name.
The arithmetic behind Cp, Cpk and Pp. Only the English word beside the figure moves. A Cpk of 1.21 is a Cpk of 1.21 everywhere; whether this plant calls that marginal is its own business.
The separator in a serial number. The scan that recovers a counter from serials a plant has already issued reads PREFIX-digits; a plant that changed the hyphen would start numbering again at one over labels already on pallets. The width is the plant's; the hyphen is not.
The width of the number in a non-conformance code. The prefix is the plant's word. Five digits is the product's shape.
A hard ceiling of twelve above containment_max_depth. That number is also the guard that stops a containment walk running away over a cycle in the data. A plant chooses how deep its packaging goes; it does not get to switch off the guard.

Two things that turned out not to be the plant's either¶

Some judgments are not one answer per plant. They are one answer per thing, and a plant that has to give one answer for all of them is being asked the wrong question. Two of these were:

How much warning a gauge wants — gauges.warn_days, thirty by default. A quarterly calibration wants a fortnight's warning and an annual one wants two months, and one plant owns both. It is a column on the gauge, set when the gauge is registered and left blank for the default. Before this column existed, the gauges screen decided due soon in JavaScript at three separate places, so the shop floor's definition of the phrase lived in the browser and the server did not know it.

Whether a material is counted in pieces on a certificate — materials.counted_in_pieces, false by default. This one was a bug as much as a gap: services/coa.py computed pallet capability only for materials whose code started UT-, which is one plant's numbering convention living in product code. Any plant not numbering its pieces that way got an empty capability block on every pallet certificate and no error anywhere. The migration sets the flag true for exactly the materials that prefix chose, so nothing about an existing plant's certificates changes — it is the same answer, given honestly by a flag instead of guessed from a name.

Editing a plant's own quality numbers¶

Quality › Configuration (/dashboard/config/quality) lists every section with its keys, the value this plant is running on, and whether that value is the product's default or one the plant set. Eleven of its twelve sections are these numbers, and each one is a box you type in.

Open Quality › Configuration. Each row's Set to column holds one box per key, with the key's name beside it.
Type the new value and press Save. There is one Save per row, because the Cpk bars and the gauge ratios are each one judgment written as two numbers — a screen that saved half of one would make marginal unreachable until you had typed the other half.
It is in force at once. No approval step, no restart, and no pack to re-apply: the next chart drawn, the next serial issued and the next certificate printed read the new number.

You need the quality.define capability, which the built-in Administrator and Agent roles hold. Without it the page shows you every value and no box: a number nobody can read is a number nobody can argue with, and the reading is open to anybody who may see the plant.

Every change is written to the audit trail — who, when, what it was and what it became. Undo is typing the old number back. There is deliberately no revision history here, because nothing in this MES records the Cpk bar that was in force when I was judged: a non-conformance stores its severity, and a chart is drawn fresh every time. That is the difference between these numbers and the severity vocabulary, which does have a draft → approve lifecycle, because the words it holds are written onto records that outlive it.

What the pack still does, and what it no longer does¶

fsmes pack apply seeds each key the pack carries, once. After that the database owns it, and a later apply leaves it exactly as it is — the same rule every other kind a pack seeds already keeps, and for the same reason: a pack that reached back into a number somebody deliberately changed on a running plant would be the pack overruling the plant.

So editing plant.toml and re-applying does not move a number this plant has already taken ownership of. If that is what you want, change it on the screen; the pack file is how a new plant starts, not how a running one is steered. fsmes pack status reports the difference.

Where the value lives, and what the screen can honestly say¶

The screen has two answers about any key — the product's default, unchanged or this plant set it — and it says only those two. A row written by fsmes pack apply is the second of them, because the plant did set it, in its pack; naming which pack would be a guess, since by the time a plant is serving the file it was built from is not recorded anywhere the running process can see.

Underneath, a number is read in three layers, in this order:

the row in this plant's plant_settings table — what its administrator or its pack wrote;
the setting the pack compiled into the environment (MES_QUALITY_*);
the literal this version of the product ships.

Which is why a plant that has never touched the page and never applied a pack behaves exactly as it did before any of this existed — and why upgrading to this version moves no data: a plant already running on spc_min_points = 25 from its pack keeps drawing limits from twenty-five readings through layer two, with an empty table.

A characteristic inspected several pieces at a time¶

A specification can say how many pieces are measured at a time — sample_size in the pack's quality_specs.json, on POST /quality/specs, and one field on the Specifications tab. Left out, or one, it means what every specification written before this existed means: one piece at a time, charted exactly as it always was. Above one the readings are a subgroup and the chart is X-bar and R, which is decision 0040.

Nothing in the table above changes its name, but three of these keys answer about a point rather than a reading, and on a sampled characteristic a point is a sample:

spc_min_points counts samples. Twelve means twelve samples — sixty bottles if they are taken five at a time — not twelve bottles. A plant that samples five at a time every fifteen minutes waits three hours for its first limits, and that is the right wait: twelve points is twelve looks at the process however many pieces each look holds.
spc_history counts samples the same way, so a chart's window is the same number of points whichever kind it is.
hold_rules and major_rules are unchanged, and the rules run on the sample means. Four readings in a row above the centre line is not a signal; four samples in a row is. One sample is one evaluate, so a sample whose five readings are all interesting raises at most one hold, and the hold's evidence names the sample, its five readings and the five rows they are stored in.

A sampled chart has one more signal, and it is the same rule 5 an individuals chart has: a range beyond its upper limit — the spread inside one sample here, the gap between two readings there. It is drawn and recorded on every plant and raises a hold where hold_rules includes 5, like any other rule. It is worth including: a sample whose range is beyond D4·R̄ means the five pieces disagree, so the mean they average to is not describing anything and the X-bar limits beside it were widened by that sample itself. The bottling lab pack names 5 on both its lists for exactly that reason.

Capability is worked out from the within-process sigma, R̄/d2, and not from the spread of the means, which is smaller by root n. A Cp computed the other way would read about √5 too high on a sample of five, and the SPC screen's Sigma (within) figure is the one capability uses with the mean's own spread in its tooltip.