Glossary¶

This repository draws on four fields that rarely share a reader: discrete-event simulation, industrial control, Lean Six Sigma, and applied machine learning. A sentence in ARCHITECTURE.md can carry a term from each. This page is the single place where each one is defined.

Two kinds of entry appear below, and the difference matters.

An entry marked local is defined by this repository. Where a term has a looser meaning in the wider industry, the meaning here is the narrow one, and docs/style/ste-terms.yml holds the machine-checked form of that rule. The prose gate rejects a declared synonym in a task or reference topic.

An entry marked external names a term owned by a standard or by an established body of practice. This page states the sense in which the term is used here. It does not quote the standard, and it is not a substitute for reading it. Section 11 names the sources this repository validates its statistics against. It is not a source list for every external term on this page.

1. Simulation and determinism¶

Term	Kind	Meaning here
twin	local	The synchronized model of the operation. Three couplings make it a twin: it recalibrates from telemetry, its divergence from reality is itself a finding, and accepted changes flow back as config.
digital shadow	external	A model fed by telemetry that sends nothing back. This repository builds a twin, not a shadow, and the README states the difference.
discrete-event simulation	external	A model that advances time by jumping to the next scheduled event, rather than by fixed increments. SimPy is the kernel, per decision D1.
sim clock	local	The virtual clock service every timestamp derives from. Code outside the kernel package reads an injected `Clock`, never a wall clock.
paced clock	local	A mode that emits events in wall-clock time at a chosen rate, for watching a run. Pacing changes when an event appears, never which event or in what order.
wall clock	local	Real time. Doctrine D-02 permits reading it in exactly four places, and in none of them does the value reach an event payload, the hashed tape, or control flow.
run seed	local	The single integer governing every stochastic stream in a run.
child seed	local	A per-subsystem stream derived from the run seed by a content-addressed derivation, so a stream's identity comes from its name and not from a draw order.
tape	local	The ordered event log a run produces. Two runs that agree byte for byte have the same tape.
event log	local	The `events.ndjson` file a run writes. One event per line, in the canonical total order.
run manifest	local	The record of what produced a run. Doctrine D-01 splits it into a hashed core and a provenance sidecar.
hashed core	local	The manifest fields that enter the log hash: seed, config hash, schema snapshot hash, scenario id, mode, tick rate, horizon, warmup, and fault schedule hash.
provenance sidecar	local	The manifest fields deliberately kept out of the hash, written to `manifest.json`. It holds wall-clock start and finish, git provenance, platform, and package versions.
what-if	local	A twin experiment that applies a config change and measures the delta against a baseline run.
counterfactual	local	A replay of a recorded run through a changed config. Distinct from a what-if, which starts fresh.
warmup	local	The leading span of a run excluded from statistics, so measurements are not taken while the model is still filling.
horizon	local	The simulated duration of a run.
scenario	local	A named, versioned run definition. `SCN-F1` is the determinism scenario the two-run hash match is asserted on.
fault injection	local	Introducing a fault from the labeled fault catalog into a run. The catalog supplies the ground-truth labels the detection comparison in D7 is scored against.
special cause	local	Variation from an assignable source, injected from the fault catalog.
common cause	local	The inherent variation of a stable process. Telling a rare common-cause tail from a genuine special cause is the judgment the statistical engine exists to make.

Determinism here is scoped by doctrine D-05, which names two tiers. Tier one is byte-identical logs on one platform at one pinned dependency set. Tier two is identical business events across platforms, with continuous fields agreeing inside a measured tolerance. Neither tier claims the operation is predictable. Reproducing a run and forecasting the next hour are different properties.

2. Industrial control and the plant floor¶

Term	Kind	Meaning here
ISA-95	external	The standard for integrating enterprise and control systems. Its level model, from L0 sensors to L4 business systems, gives the topic hierarchy every device publishes into.
Purdue Model	external	The reference network segmentation for industrial control, layering L0 to L5. Section 7 of ARCHITECTURE.md puts it in the compose topology, not only in a diagram.
ISA-88	external	The batch control standard. It supplies the recipe model used by the upstream factory at phase P3i.
IEC 62443	external	The industrial automation and control systems security series. It supplies the zone and conduit model at phase 6a15.
OT	external	Operational technology. The plant-floor segment: devices, gateways, and the broker they publish to.
IT	external	Information technology. The enterprise segment: the historian, the statistical engine, the agent, and the dashboard.
bridge	local	The single crossing point between the OT and IT segments. Milestone RA-b lands the test asserting no device container is reachable from the IT segment.
edge gateway	local	The device that collects local telemetry, buffers it across a link outage, and publishes north to the site broker.
PLC	external	Programmable logic controller. The default path models one rather than talking to one, and the README says so.
OPC UA	external	The OPC Unified Architecture protocol for industrial data exchange. A bridge is a roadmap milestone at P5, not a current capability.

3. Telemetry and messaging¶

Term	Kind	Meaning here
UNS	local	The unified namespace: the ISA-95 topic hierarchy every device publishes into, and the single place the current state of the operation is readable.
MQTT	external	The publish and subscribe messaging protocol the fleet runs over. Three brokers fill three roles, per decisions D3 to D5.
Sparkplug B	external	The Eclipse specification, at version 3.0.0, that fixes the payload format and the session lifecycle over MQTT. Every normative statement in it carries a `tck-id`, so compliance is testable.
birth certificate	local	The Sparkplug NBIRTH or DBIRTH message declaring a node or device online, and declaring its full metric model.
death certificate	local	The Sparkplug NDEATH or DDEATH message. NDEATH arrives through the MQTT Last Will, which marks a node stale rather than leaving it silently frozen.
report by exception	external	Publishing a metric only when it changes, rather than on every cycle. NDATA and DDATA carry these updates after the birth certificate has declared the model.
metric aliasing	external	Replacing a metric name with a short integer after the birth certificate has bound the two, which cuts payload size on every later message.
store and forward	local	Buffering telemetry during an outage and replaying it on reconnect.
historian	local	The L2 time-series system of record. It is append-only and event-sourced, so a recorded run can be replayed rather than only summarized.
device twin	local	The desired and reported state pair held for each device.
EPCIS	external	The GS1 standard for supply chain event data. Its 2.0 event vocabulary lands at phase P3, and it answers what, when, where, and why for a traceable object.

4. Lean Six Sigma and statistics¶

Term	Kind	Meaning here
takt	local	Available time divided by customer demand. The rate the operation must run at to meet demand.
cycle time	external	The time one unit takes to pass through a step.
WIP	external	Work in process. The count of units inside the system and not yet finished.
OEE	external	Overall equipment effectiveness, the product of availability, performance, and quality.
bottleneck	external	The resource whose capacity sets the throughput of the whole line.
SPC	external	Statistical process control. Judging a process by whether its variation is stable, rather than by whether a single reading is inside a specification.
control chart	external	A time-ordered plot with control limits derived from the process itself, not from the specification.
Western Electric rules	external	A published set of control chart patterns that identify an out-of-control signal. The engine emits the rule number, not a generic alarm.
Nelson rules	external	A second published pattern set covering the same job, with different sensitivities.
violation	local	A control chart rule breach, identified by rule number.
finding	local	A typed judgment the statistical engine raises, carrying a severity and its evidence window. Never called an alert, an insight, or an observation.
process capability	external	How well a stable process fits inside its specification limits. Cp and Cpk describe short-term capability, Pp and Ppk the long-term form.
Gage R and R	external	Gage repeatability and reproducibility. It separates variation coming from the measurement system from variation coming from the process.
MSA	external	Measurement system analysis, the family Gage R and R belongs to. It asks whether the measurement can be trusted before the process is judged.
repeatability	external	Variation seen when one operator measures the same part twice.
reproducibility	external	Variation seen between operators measuring the same part.
ANOVA	external	Analysis of variance. It splits total variation into named sources, and it is the method behind the Gage R and R error terms.
assumption checker	local	The component that tests the preconditions of a hypothesis test before the test runs, so a p-value is not reported for a test whose assumptions failed.
value stream map	external	The Lean drawing of material and information flow across a process, with the waiting time between steps made visible.
SIPOC	external	Suppliers, inputs, process, outputs, customers. The one-page scoping table for a process. Here the process column is mined and the rest is declared.
swimlane map	external	A process drawing with one lane per resource pool, so every handoff between pools is visible.
Pugh matrix	external	A ranking of candidate changes against a datum, scoring each criterion better, same, or worse. Here the datum is the current state.
stoplight analysis	external	A green, yellow, or red class per candidate, from its impact and its cost. Here the matrix is declared in config and gated.
run chart	external	A time-ordered plot judged by run patterns about the median, used before a series is long enough for control limits.
box plot	external	The five-number summary drawn as a box and whiskers, with outliers past the Tukey fences listed.
concentration diagram	external	Defect counts drawn on the physical layout, so location patterns become visible.
MANOVA	external	Multivariate analysis of variance. Several continuous Ys tested against discrete Xs at once.
X-Y map	external	The Measure-phase table linking each input X to the critical Ys it is hypothesized to move.
HODDOW	external	Hour-of-day, day-of-week aggregation. WOMSOY is its week-of-month, season-of-year twin.
Gantt chart	external	A timeline with one lane per resource and one bar per activity occurrence.
operator-machine chart	external	The Gantt chart filtered to one operator and one machine, showing their interaction.
CPM	external	Critical path method. The forward and backward pass giving the critical path and per-task slack.
PERT	external	Three-point schedule estimation: mean (a + 4m + b) / 6 with a per-task variance.
fishbone diagram	external	The 7-M cause skeleton behind an effect. Also called Ishikawa.
five whys	external	An ordered why-chain from effect to root cause, each step carrying its evidence.
fault tree	external	AND and OR gates over basic events, with minimal cut sets computed.
SMED	external	Single-minute exchange of die. The method for cutting changeover time.
COPQ	external	Cost of poor quality. The money a defect costs once scrap, rework, and warranty are counted.
capability report	local	The generated HTML report for a time window, naming the Gage R and R convention it used.

The Gage R and R F-test needs a stated convention, because the published ones disagree about the error term for the operator effect. The README records the split as the CRAN documentation for the R SixSigma package states it. This repository builds both error terms, tests each against its own published example, and names the convention in every generated report.

5. Warehouse and distribution operations¶

Term	Kind	Meaning here
receiving	local	Taking material off an inbound vehicle and recording it. The first station modeled, at phase P1.
putaway	local	Moving received material to its storage location. One word, no hyphen.
slotting	local	Assigning a SKU to a storage location.
cross-dock	local	Moving inbound material to an outbound door without storing it.
AMR	local	Autonomous mobile robot.
ASRS	external	Automated storage and retrieval system.
sortation	external	The automated diverting of units to lanes by destination.
palletizer	external	The cell that builds a pallet from incoming cases.
cartonization	external	Choosing which carton sizes an order packs into.
perfect order	local	An order complete, on time, damage free, and correctly invoiced.
OTIF	external	On time in full. The share of orders delivered complete and on schedule.
fill rate	external	The share of demand met from stock on hand.
RFID portal	local	A fixed reader at a doorway that reads tags as material passes. RF physics and read-zone modeling land at phase P3.
read zone	local	The physical volume in which a portal can read a tag. Modeling it is what makes a missed read a modeled event rather than a random one.

6. Planning, inventory, and supply¶

Term	Kind	Meaning here
S&OP	external	Sales and operations planning. The monthly cycle that reconciles demand, supply, and finance into one plan.
SIOP	external	Sales, inventory, and operations planning. The same cycle, named to keep inventory explicit.
S&OE	external	Sales and operations execution. The weekly tick that handles what the monthly plan did not foresee.
ATP	external	Available to promise. Whether stock exists to commit to an order.
CTP	external	Capable to promise. Whether capacity exists to make and deliver it.
ABC and XYZ segmentation	external	Two rankings of the same catalog. ABC ranks items by value contribution, so control effort follows the money. XYZ ranks them by demand variability, which is a different question.
MEIO	external	Multi-echelon inventory optimization. Deciding where in a network safety stock sits, rather than setting each site alone.
newsvendor	external	The single-period stocking model that balances the cost of too much against the cost of too little.
guaranteed service time	external	The service-time framing of multi-echelon stock placement.
risk pooling	external	Holding stock centrally so variability across locations partly cancels.
conformal prediction	external	A method that turns a point forecast into an interval with a stated coverage guarantee. It is what makes a forecast consumable by an optimizer.
forecasting arena	local	The harness that runs every forecasting model over the same backtest windows on this project's own series, and publishes the table. It imports no winner from a benchmark paper.

7. Process mining¶

twinflow-procmine is written in this repository under Apache-2.0, because PM4Py is AGPL-3.0 and section 13 of that license reaches a network-served work. Doctrine ruling D-14 settles it, and decision D8 records the reasoning.

Term	Kind	Meaning here
process discovery	external	Deriving a process model from an event log.
directly-follows graph	external	The graph of which activity follows which, and how often. The simplest discovered model, and the input to richer ones.
inductive miner	external	A discovery algorithm that splits the log recursively and returns a model with sound structure by construction.
conformance checking	external	Measuring how far a real log departs from a reference model.
token-based replay	external	Conformance measured by pushing the log through a Petri net and counting the tokens missing or left behind.
alignment	external	Conformance measured by finding the cheapest correspondence between a trace and a model run. Computed here as A star search over the synchronous product net.
variant analysis	external	Grouping traces by their activity sequence, so the common paths and the rare ones separate.
rework loop	local	A repeated activity in a trace that indicates work redone rather than work advanced.
ground-truth process model	local	The twin's own designed process. Because this repository owns the reference model, discovery can be scored on how well it recovers a known answer. No external library closes that loop.

8. The agent and its accuracy stack¶

The target is that the agent never states a number it did not get from an execution. Seven layers enforce it, and the README lists all seven.

Term	Kind	Meaning here
execution-grounded answer	local	A quantitative answer produced by running generated SQL or Python against the historian in a sandbox, and reporting the execution result.
governed semantic metric	local	A metric defined once in YAML with exact SQL, so the model picks a metric rather than writing its own aggregation.
execution-based verification	local	Running every generated query before the answer ships, and feeding errors, empty results, and out-of-range magnitudes back for bounded retries.
constrained decoding	external	Restricting token generation so the output must match a schema. It gives the local model path the same structural guarantee a hosted structured-output API gives.
self-consistency	local	Sampling several query programs for a hard question and executing all of them. The modal result answers, and no majority marks the question.
grounding checker	local	The component that refuses to ship a sentence carrying a number matching no logged query result id.
calibrated abstention	local	Saying the twin lacks the data to answer reliably, below a threshold set by calibration rather than by taste.
eval suite	local	The versioned set of operational questions with known answers that the agent is scored against.
MCP	external	The Model Context Protocol. The server, its threat model, and its red-team suite land at phase P3.

Layers three to seven ship as twinflow-accuracy, which installs with no LLM SDK. A team that already has an agent can take the grounding checker alone.

9. Repository craft¶

Term	Kind	Meaning here
brick	local	A package that installs alone and drags in nothing else. `test_brick_isolated.py` in each package asserts it.
milestone	local	One idea, with an id that never changes and is never removed. Reordering is allowed and recorded. Deletion is not.
work package	local	The unit of delivery inside a phase, tracked in `roadmap.yaml` with a status.
phase	local	A group of work packages ending at one tagged release. One minor version per phase.
VAL-GATE	local	A validation gate. A named assertion, its falsifier, and the command that runs it.
standing gate	local	A gate that re-runs at every later phase exit, not only at the phase that introduced it.
gate status	local	One of `declared`, `specified`, or `implemented`. The required field set widens with each, so a subsystem must declare its gates one phase before building them.
doctrine ruling	local	A binding cross-cutting decision in DOCTRINE.md, cited as `D-01` and up, that every design section obeys. Where a section disagrees, the ruling wins.
technology decision	local	A recorded choice of tool or approach in section 1 of ARCHITECTURE.md, cited as `D1` and up. Distinct from a doctrine ruling despite the similar id.
decision register	local	One of the three places a decision lives. adr/index.md is the index of record and holds the id range each register uses.
metric marker	local	An HTML comment pair holding a quantitative result. `TBD` means unmeasured, and a release cannot be tagged while it owes a marker.
known-answer test	local	A test that checks generated values against a frozen corpus, so a change in the generator is caught rather than absorbed.
property-based test	local	A test that asserts an invariant over generated inputs, rather than over one example. Hypothesis is the tool.
phase-exit runner	local	The command that runs every gate in force at a phase. A gate in that set which is not `implemented` fails the run rather than being skipped.

10. Declared verbs¶

docs/style/ste-terms.yml declares six verbs and the nouns above. One verb carries one meaning, and the prose gate rejects the listed synonym for it in a task or reference topic.

Verb	Means here	Not
publish	Send a message to a UNS topic	emit, push, send out, broadcast
replay	Re-execute a recorded event log	rerun, playback, re-simulate
recalibrate	Adjust twin parameters from observed telemetry	retune, adjust, sync up
inject	Introduce a fault from the fault catalog into a run	trigger, simulate a failure, cause
raise	Create a finding	fire, throw, generate an alert, surface
seed	Initialize a stochastic stream from a seed value	initialize randomness, set the random state

11. External sources¶

Each row names a source this repository validates a statistic against, and what that source covers. Nothing on this page quotes those documents, and several of them are sold rather than published.

Source	Covers
ISA-95, ISA-88, and IEC 62443, from the International Society of Automation and the IEC	The level model, the batch recipe model, and the industrial security zone model
The Purdue Enterprise Reference Architecture	The network layering L0 to L5
Eclipse Sparkplug B, version 3.0.0	The payload format, the session lifecycle, and the `tck-id` compliance statements
GS1 EPCIS 2.0	The supply chain event vocabulary
NIST Statistical Reference Datasets	Certified values for univariate statistics, ANOVA, and regression
NIST/SEMATECH e-Handbook of Statistical Methods, chapter 6	Control charts, process capability, and acceptance sampling
The CRAN documentation for the R SixSigma package	The two published Gage R and R F-test error terms
The AIAG Measurement Systems Analysis manual	The measurement analysis conventions. Sold rather than published, and reported here only as the SixSigma documentation states it

For how this repository decides what a source can be cited for, read doctrine ruling D-11 on validation evidence.