@dreamforgestudiollc/emet 0.2.4 → 0.2.6

package/gems/lean-six-sigma/guides/ASSUMPTIONS.md

@@ -0,0 +1,45 @@
+## Assumption flags
+
+Name the assumptions a result needs before the result is used. If an assumption was not checked, say it was not checked. Do not treat an unchecked assumption as if it held. Do not invent a test result to paper over a gap.
+
+These are the flags this package uses. Attach the ones that apply. Leave the others off the result.
+
+### Normal model
+
+Cp, Cpk, Pp, Ppk, and the z interval use a normal model when you want the handbook meaning of the index. Say whether you checked normality. If you did not, write "normality not checked". NIST/SEMATECH e-Handbook section 6.1.6 states that those capability indexes assume a normal distribution, and that the sample should be large. This package does not run a normality test for you.
+
+### Stable process
+
+Capability and control limits describe a process whose mean and spread are steady over the data you used. If the chart is not stable, or you did not look, write "stability not checked". Do not quote a Cpk from a data set you already called unstable.
+
+### Sigma kind
+
+Within-subgroup sigma (from a rational subgroup, often R-bar / d2) and overall sigma (the standard deviation of all individuals) are different inputs. Cp and Cpk take the within sigma when that is what you claim. Pp and Ppk take the overall sigma. If the label and the formula disagree, write "sigma label mismatch" and do not relabel the input.
+
+### Independence
+
+t, chi-square, ANOVA, and regression treat observations as independent unless you say otherwise. Paired data are not two independent samples. Write "pairing" when the row is a pair, and use a paired method. This version names the paired t and does not work that statistic. See the hypothesis selection guide.
+
+### Equal variance
+
+Two-sample t and one-way ANOVA are sensitive to unequal spread. Write "equal variance not checked" unless you checked it. This version works the one-way ANOVA sums of squares and does not test the variances for you.
+
+### Binomial or Poisson
+
+A p chart and an np chart use a binomial model: a unit conforms or it does not, trials are independent, and the subgroup size is the n you used. A c chart and a u chart use a Poisson model: counts, and a constant area of opportunity for a c chart. NIST/SEMATECH sections 6.3.3.1 and 6.3.3.2 are the references. If the opportunity changes and you used a c chart, say so and consider a u chart.
+
+### Sample size and the margin of error
+
+The sample-size formulas in this package are normal approximations for a margin of error. They are not a power calculation for a hypothesis test. Write "margin of error, not power" on the plan.
+
+### Small n and zero spread
+
+Regression with fewer than three points still returns a fit when the x values are not all equal, and you flag small n. If every x is equal, there is no slope. Do not report one.
+
+### Critical values and p-values
+
+This version returns a t statistic, a chi-square statistic, ANOVA sums of squares and F, and a regression fit. It does not return a p-value. A critical value is used only when you supply it. The package does not look up a t table for you.
+
+### Illustrations
+
+Worked examples and filled template rows are illustrations. They are not your process. Copy the arithmetic shape. Supply your own inputs.

package/gems/lean-six-sigma/guides/BELTS.md

@@ -0,0 +1,52 @@
+# Belt coaching
+
+These notes coach the work. They do not certify a Yellow Belt, a Green Belt, or a Black Belt. Ask which belt the user is working as on this project. Offer the matching notes. Store the answer on the project document only when the user asks to save it.
+
+## Yellow Belt
+
+This note does not certify a belt. A Yellow Belt on this project can draft and fill:
+
+- Project charter
+- SIPOC
+- Voice of the customer and the CTQ tree
+- Data collection plan
+- A 5S sheet for one area
+- A waste walk
+
+Coach question to offer: "Do you want a Green Belt or a Black Belt with you before a hypothesis test, a sample-size claim, or a financial result?"
+
+Stay with the user while they collect. Do not fill a blank they have not answered. Do not choose their hypothesis test.
+
+## Green Belt
+
+This note does not certify a belt. A Green Belt on this project can run Define through Control with the templates, the Lean tools, and the hand formulas in `guides/FORMULAS.md`.
+
+That includes the tollgates, the measurement plan, the fishbone and Five Whys, the tests the formula guide works in full, the pilot, the control plan, and the response plan.
+
+Coach question to offer: "Do you want a Black Belt with you for a designed experiment, a Gage R&R decision, or a savings number finance has not signed?"
+
+A Gage R&R result is recorded only when the user already has one. This version does not calculate percent of tolerance. A designed experiment stays a plan until the user supplies the design and the run order.
+
+## Black Belt
+
+This note does not certify a belt. A Black Belt on this project can use the full set: the DMAIC path, the DMADV note, DOE planning, the assumption flags in `guides/ASSUMPTIONS.md`, and the coaching questions above with a Yellow Belt or a Green Belt.
+
+Coaching means asking for the data and the assumption, and leaving the other person's blank fields blank. It does not mean completing their charter, picking their cause, or signing their savings.
+
+Before a number is used, ask which formula, which inputs, and which assumption was not checked. If they want a p-value, say this version does not return one. They may compare a statistic to a critical value they supply, using the formula guide.
+
+## Coaching
+
+This note does not certify a belt. Use these questions in order. Stop when the user stops.
+
+1. Is this an existing process (DMAIC) or a new design (DMADV)?
+2. Which phase are you in? If you do not know, we start at Define.
+3. Which belt are you working as on this project: Yellow, Green, or Black?
+4. What is the CTQ, and what is the unit?
+5. Which fields are still blank? They stay blank.
+6. Which formula are you using, and which input is missing?
+7. Which assumption did you check, and which did you not?
+8. Who in finance has signed a hard-savings number? If nobody has, the cell stays blank.
+9. Do you want this filed as a sealed record? File it only if you say yes. A calculator does not write the record.
+
+Do not grade the user. Do not say they are ready to sit for a certification.

package/gems/lean-six-sigma/guides/DMADV.md

@@ -0,0 +1,56 @@
+## DMADV
+
+DMADV is the path for a new design (Design for Six Sigma): Define, Measure, Analyze, Design, Verify. DMAIC is the path for an existing process. If the work is a new design, keep the DMADV label. Do not copy DMADV results into a DMAIC phase.
+
+The same rules apply. Ask for data. Leave a blank blank. Name assumptions you did not check. Illustrations are not your data. Use `guides/FORMULAS.md` when a formula applies. The five client calculators return numbers and do not write a record. Say when a predicted capability is only as good as the inputs you supplied.
+
+A Black Belt coaching a design team asks who the customer is, which CTQ is measurable, and which validation row came from a trial. This package does not certify a belt.
+
+## Define
+
+Define the new design in customer and business terms.
+
+- Customer: who will use the design, in the words the sponsor uses.
+- Business case: why a new design is required, and what the current process cannot do. Leave cost figures blank until finance or the user supplies them.
+- Charter: problem in design language (the need, not a chosen concept), goal, scope in, scope out, team, champion, start, and target end.
+- The champion's acceptance is a name the user supplies. This package does not sign.
+
+Open `guides/TEMPLATES.md#Project charter` and mark the sheet "DMADV" so it is not filed as a DMAIC charter.
+
+## Measure
+
+Measure sets the CTQs and the targets before a concept is chosen.
+
+- Voice statements in the customer's words.
+- Need, driver, and a CTQ with target, lower spec, upper spec, unit, and measurement method.
+- A Kano label only when the user supplies one.
+- Targets the user sets. A target copied from a similar product is still the user's number, and the source is written on the sheet.
+- If no voice has been collected, write the dated plan and leave the CTQ specs blank.
+
+## Analyze
+
+Analyze compares design concepts the user lists. It does not invent a concept.
+
+- List the concepts the team is willing to consider. Leave the list empty until they name them.
+- For each concept, record the risks the user names: safety, compliance, capability, cost, and time. An empty risk cell stays empty.
+- Score only ratings the user enters. See the FMEA template if they want RPN.
+- Write the concept they select, and the reason in their words. If they have not selected one, the gate stays open.
+
+## Design
+
+Design records the selected design at a level a pilot can build.
+
+- The selected concept and the features that meet each CTQ.
+- Predicted capability only from inputs the user supplies (mean, sigma, specs), computed with `guides/FORMULAS.md#Cp` and the Cpk section. Name whether sigma is within-subgroup or overall. If normality was not checked, say so.
+- A pilot plan: units, site, measures, and the decision rule the user states.
+- A DOE plan if they will experiment, or the words "no designed experiment on this project". This version does not build a design matrix.
+
+## Verify
+
+Verify uses validation data the user collected, then leaves a control plan for the new process.
+
+- Validation rows: what was run, how many, and the results they recorded. Do not fill results from the prediction.
+- Compare results to the CTQ specs. A miss stays a miss.
+- Control plan for the new process: step, CTQ, spec, method, sample, owner. Limits from stable validation data, or "limits not set".
+- Response plan and process owner.
+- Financial worksheet from measured results. Soft savings stay labeled soft.

package/gems/lean-six-sigma/guides/DMAIC.md

@@ -0,0 +1,127 @@
+## DMAIC
+
+DMAIC is the path for an existing process: Define, Measure, Analyze, Improve, Control. A new design uses the DMADV note in `guides/DMADV.md` and keeps that label.
+
+Work one phase at a time. If you do not know the phase, start at Define and say that is where you started. Copy blanks from `guides/TEMPLATES.md`. Compute from `guides/FORMULAS.md` with your own data. Five client calculators return numbers for capability, DPMO and yield, control limits, sample size, and basic statistics. They do not write a record. File a sealed record only when you ask.
+
+An illustration in a template is not your data. Leave a field blank until you supply the fact. A tollgate item with no answer stays open. You may defer an item only by writing the deferral and the date.
+
+Project figures stay in this install. Do not send them to another host, an inbox, or a service.
+
+## Define
+
+Define names the problem, the goal, and the boundaries. It does not name a cause and it does not name a solution.
+
+1. Write the problem statement: what is wrong, where, when, and how large. Include a baseline, or write "baseline not measured yet". Keep the cause and the remedy out of this sentence. Those belong in Analyze and Improve, after the data.
+2. Write the goal as baseline, target, and date. Leave the baseline blank until Measure records it.
+3. Write scope in and scope out. Name the process steps that are inside the project and the sites, products, or shifts that are outside it.
+4. Name the team, the roles, and the champion. The champion accepts the charter. This package does not sign for them.
+5. Write the business case in the customer's terms and in the company's terms. Money waits for the financial worksheet and for a person in finance.
+6. Build a SIPOC: suppliers, inputs, about five to seven process steps, outputs, customers.
+7. Record the voice of the customer in their words, then a CTQ that can be measured: target, lower spec, upper spec, unit, and method. If you do not have a voice yet, write a dated plan to collect it.
+8. Open `guides/TEMPLATES.md#Project charter` and fill only the lines you know.
+
+Coach note. A Yellow Belt can draft the charter, the SIPOC, and the CTQ tree. Ask whether a Green Belt or a Black Belt should read the problem statement before the champion accepts it.
+
+## Define tollgate
+
+Read each line. Leave it open until the user supplies it, or until the user writes a deferral and a date.
+
+- Problem statement with what, where, when, and magnitude. No cause. No solution.
+- Baseline number, or the words "baseline not measured yet".
+- Goal with target and date.
+- Scope in and scope out.
+- Team, roles, and champion.
+- Business case.
+- SIPOC.
+- At least one CTQ, or a dated plan to collect the voice of the customer.
+- Charter status: the name of the person the user says has accepted it.
+
+Do not close a line because a similar project usually has that answer.
+
+## Measure
+
+Measure turns the CTQ into a number you can trust, then records the baseline from data you collected.
+
+1. Write an operational definition for each measure: what is counted, the unit, the inclusion rule, and the exclusion rule. Two people using the definition should reach the same count on the same item.
+2. Fill the data collection plan: continuous or discrete, source, method, who collects, when, and the stratification factors (shift, line, product, or others you name). Leave sample size blank until you calculate it from `guides/FORMULAS.md#Sample size for a proportion` or `#Sample size for a mean`, or until you supply a size from a standard your sponsor already uses.
+3. Plan the measurement system before you trust the baseline. Use `guides/TEMPLATES.md#MSA and Gage R&R plan`. This version records the plan and any result you already have. It does not calculate Gage R&R.
+4. Map the process or the value stream with times the user measured. Lead time and process time stay blank until those times exist. See `guides/LEAN.md#Value stream map`.
+5. Compute the baseline only from the rows you collected. DPMO, yield, and capability come from `guides/FORMULAS.md` with those inputs. Name the formula. If an input is missing, stop and ask.
+6. State the assumptions beside any index: stable process, and, for Cp and Cpk, a normal model unless you say you did not check normality.
+
+Coach note. Do not quote a sigma level from a training table. The count path is DPMO. Inverse normal is not computed in this version. See `guides/FORMULAS.md#Sigma level`.
+
+## Measure tollgate
+
+- Operational definition for each measure.
+- Data collection plan, including who collects and when.
+- MSA plan. A result only if you already have one.
+- Process map or value stream map built from observed steps.
+- Baseline computed from your data, with the formula named.
+- Sample size either calculated from the formula guide or supplied by you, with the source written down.
+- Assumptions named, including any you did not check.
+
+## Analyze
+
+Analyze keeps a cause only when the evidence you bring supports it. The package does not pick the cause.
+
+1. Build a cause list with the fishbone. Categories may be people, method, machine, material, measurement, and environment. Drop or add a category. Every cause is in your words.
+2. Run Five Whys on the causes you choose to push on. Stop at the why you stop at. Leave the countermeasure blank until you write it.
+3. Choose a test from `guides/TEMPLATES.md#Hypothesis test selection`. You pick. The guide names assumptions. For a test this package works by hand, follow the matching section in `guides/FORMULAS.md`. For a test marked named only, write the name and the assumptions, and compute it with your own textbook. This version does not return a p-value.
+4. Write the cause you accept for Improve. If you do not accept one yet, the item stays open.
+
+Coach note. A Green Belt can run the basic tests in the formula guide. Ask whether a Black Belt should review a designed experiment, a Gage decision, or a savings number finance has not signed.
+
+## Analyze tollgate
+
+- Fishbone or an equivalent cause list from you.
+- Five Whys for the causes you selected.
+- The analysis you chose, with assumptions written beside the result.
+- The cause you accept for Improve, in your words.
+- Any test you did not run left as "not run", not as a result.
+
+## Improve
+
+Improve changes the process against the accepted cause, on a pilot first.
+
+1. Write countermeasures that name the accepted cause they address. A countermeasure with no cause stays a draft.
+2. Write the pilot plan: where, how many units or how many days, what you will measure, and what "better" means using the goal from Define.
+3. If you will run a designed experiment, fill `guides/TEMPLATES.md#DOE planning`. If you will not, write "no designed experiment on this project". This version does not build a design matrix. You supply the run order.
+4. Score an FMEA for the change with severity, occurrence, and detection you enter from 1 to 10. RPN is the product only when all three numbers are present. See `guides/FORMULAS.md#RPN`.
+5. Record pilot results from the pilot data only. Update the financial worksheet from those results. Hard savings and soft savings stay in their own rows. Do not annualize unless you supply the number of periods and ask for the product.
+6. Use Lean tools where they match the cause: 5S, setup reduction, pull, mistake-proofing, or a kaizen cycle. See `guides/LEAN.md`.
+
+## Improve tollgate
+
+- Countermeasures tied to accepted causes.
+- Pilot plan.
+- DOE plan, or the words "no designed experiment on this project".
+- FMEA scored only with your ratings.
+- Pilot results from your data.
+- Financial worksheet updated from those results, with soft savings still labeled soft.
+- The name of the person in finance who validated a hard-savings number, or a blank if nobody has.
+
+## Control
+
+Control hands a stable method to a process owner, with limits from your data and a response when a point crosses a limit.
+
+1. Write the control plan: step, CTQ, spec, measurement method, sample size and frequency, chart type, and owner.
+2. Write the response plan: which limit, the immediate action, containment, escalation, and the condition for resuming. A point is out of limits only when you supply the point and the limits, or when you compare a supplied point to limits you just computed from supplied data.
+3. Set chart limits from a stable set of your data, using `guides/FORMULAS.md`, or write "limits not set".
+4. Name the process owner and the handoff date.
+5. Record lessons you actually learned. Do not fill lessons from a template example.
+6. Leave the benefits in the state the worksheet records.
+7. Write how often the owner will look at the chart.
+
+This version calculates limits. It does not apply Western Electric or Nelson run rules. Add those only when you choose a published rule and apply it to points you supply.
+
+## Control tollgate
+
+- Control plan.
+- Response plan.
+- Control limits from your stable data, or the words "limits not set".
+- Process owner and handoff.
+- Lessons you recorded.
+- Benefits in the worksheet's own words.
+- Monitoring frequency.

package/gems/lean-six-sigma/guides/FORMULAS.md

@@ -0,0 +1,330 @@
+# Formulas
+
+Compute these from inputs you supply. A missing input stops the calculation. Five client calculators return numbers for capability, DPMO and yield, control limits, sample size, and the basic statistics in this guide. They do not write a record. Takt time, OEE, Little's Law, rolled throughput yield, first pass yield, and a sigma level stay hand calculations. Illustrations use the test vectors below. They are not your process data.
+
+Control-chart factors for subgroup sizes 2 through 10 are the table in NIST/SEMATECH e-Handbook section 6.3.2.1, checked against that page on 2026-10-11. For n = 5 the page prints A2 = 0.577, D3 = 0, D4 = 2.115. Some pocket cards print D4 for n = 5 as 2.114. This package uses 2.115.
+
+| n | A2 | D3 | D4 |
+| --- | --- | --- | --- |
+| 2 | 1.880 | 0 | 3.267 |
+| 3 | 1.023 | 0 | 2.575 |
+| 4 | 0.729 | 0 | 2.282 |
+| 5 | 0.577 | 0 | 2.115 |
+| 6 | 0.483 | 0 | 2.004 |
+| 7 | 0.419 | 0.076 | 1.924 |
+| 8 | 0.373 | 0.136 | 1.864 |
+| 9 | 0.337 | 0.184 | 1.816 |
+| 10 | 0.308 | 0.223 | 1.777 |
+
+## DPMO
+
+Definition. DPO = defects / (units * opportunities per unit). DPMO = DPO * 1,000,000.
+
+Assumptions. You counted defects, units, and opportunities the way the operational definition says. This is not first pass yield of units.
+
+Worked example. Test vector: 10 defects, 100 units, 5 opportunities per unit. DPO = 10 / (100 * 5) = 10 / 500 = 0.02. DPMO = 0.02 * 1,000,000 = 20000.
+
+Reference. The count definition is the one used with the ASQ Certified Six Sigma Black Belt body of knowledge (DMAIC Measure). Page pin is part of the 2027-10-11 source review. https://www.asq.org/cert/six-sigma-black-belt
+
+## Opportunity yield
+
+Definition. Opportunity yield = 1 - DPO. It is the share of opportunities with no defect.
+
+Assumptions. Same counts as DPMO. Do not call this first pass yield.
+
+Worked example. Test vector: DPO = 0.02. Opportunity yield = 1 - 0.02 = 0.98.
+
+Reference. Same count path as DPMO. https://www.asq.org/cert/six-sigma-black-belt
+
+## First pass yield
+
+Definition. First pass yield = (units in - units that failed) / units in.
+
+Assumptions. You supply units in and units that failed. A unit can fail once in this ratio. It is not DPO.
+
+Worked example. Test vector for the definition check: units in = 100, units that failed = 2. First pass yield = (100 - 2) / 100 = 0.98. If you do not have both counts, leave the yield blank. The 0.98 here is from unit counts, not from the DPMO vector.
+
+Reference. Standard throughput yield of units, kept separate from opportunity yield so the two 0.98 results are not treated as the same measurement. https://www.asq.org/cert/six-sigma-black-belt
+
+## Rolled throughput yield
+
+Definition. RTY is the product of the step yields you list.
+
+Assumptions. Each step yield is a yield you already computed or measured. Do not invent a step yield.
+
+Worked example. Test vector: 0.9, 0.8, and 0.95. RTY = 0.9 * 0.8 * 0.95 = 0.684.
+
+Reference. Product of step yields, as used in the Black Belt process-yield material. https://www.asq.org/cert/six-sigma-black-belt
+
+## Sigma level
+
+Definition. Long-term z is the standard-normal quantile whose upper tail equals the defect rate, when you ask for that model. A shifted sigma equals that z plus a shift you choose. This package does not insert 1.5 when you omit the shift.
+
+This version does not compute the inverse normal. Leave the sigma fields blank. Report DPMO and opportunity yield from the counts. The phrase "about 3.4 DPMO at six sigma" is the common Motorola table with a 1.5 shift, rounded. It is a label for that convention. It is not a count from your project, and this package does not calculate it.
+
+Assumptions. A sigma level needs a named model. If you do not name one, do not state a sigma level.
+
+Worked example. Test vector is the hand DPMO path: 10 defects, 100 units, 5 opportunities. DPO = 0.02. DPMO = 20000. Opportunity yield = 0.98. Sigma level: not computed.
+
+Reference. Count path as in the DPMO section. Convention label is not a NIST computed vector. https://www.asq.org/cert/six-sigma-black-belt
+
+## Cp
+
+Definition. Cp = (USL - LSL) / (6 * sigma).
+
+Assumptions. Normal model if you want the handbook meaning. Stable process. Sigma is the within-subgroup sigma when you call the index Cp. NIST/SEMATECH 6.1.6. If normality was not checked, say so. https://www.itl.nist.gov/div898/handbook/pmc/section1/pmc16.htm
+
+Worked example. Test vector: USL = 10, LSL = 0, sigma = 1. Cp = (10 - 0) / (6 * 1) = 10/6.
+
+## Cpu
+
+Definition. Cpu = (USL - mean) / (3 * sigma).
+
+Assumptions. Same normal model and stability notes as Cp. Needs a mean and an upper spec.
+
+Worked example. Test vector: USL = 10, mean = 7, sigma = 1. Cpu = (10 - 7) / 3 = 1.
+
+Reference. NIST/SEMATECH 6.1.6. https://www.itl.nist.gov/div898/handbook/pmc/section1/pmc16.htm
+
+## Cpl
+
+Definition. Cpl = (mean - LSL) / (3 * sigma).
+
+Assumptions. Same notes as Cpu, with a lower spec.
+
+Worked example. Test vector: mean = 7, LSL = 0, sigma = 1. Cpl = (7 - 0) / 3 = 7/3.
+
+Reference. NIST/SEMATECH 6.1.6. https://www.itl.nist.gov/div898/handbook/pmc/section1/pmc16.htm
+
+## Cpk
+
+Definition. Cpk = the smaller of Cpu and Cpl.
+
+Assumptions. Same normal model and stability notes. Cpk is not larger than Cp when both specs exist and the mean is inside them.
+
+Worked example. Test vector: Cpu = 1, Cpl = 7/3. Cpk = 1.
+
+Reference. NIST/SEMATECH 6.1.6. https://www.itl.nist.gov/div898/handbook/pmc/section1/pmc16.htm
+
+## Pp
+
+Definition. Pp uses the Cp algebra with the overall sigma of the individuals: Pp = (USL - LSL) / (6 * sigma overall).
+
+Assumptions. Do not relabel a within sigma as overall. If the label disagrees with the formula, write "sigma label mismatch" and keep the number you actually computed under the name of the sigma you passed.
+
+Worked example. Test vector: USL = 10, LSL = 0, overall sigma = 1. Pp = 10/6.
+
+Reference. Same algebra as NIST/SEMATECH 6.1.6, with the overall sigma named as such. https://www.itl.nist.gov/div898/handbook/pmc/section1/pmc16.htm
+
+## Ppk
+
+Definition. Ppk is the smaller of the one-sided indexes computed with the overall sigma.
+
+Assumptions. Same label rule as Pp.
+
+Worked example. Test vector: mean = 7, USL = 10, LSL = 0, overall sigma = 1. Upper index = 1. Lower index = 7/3. Ppk = 1.
+
+Reference. Same algebra as Cpk in NIST/SEMATECH 6.1.6, using overall sigma. https://www.itl.nist.gov/div898/handbook/pmc/section1/pmc16.htm
+
+## Sigma from range
+
+Definition. sigma hat = R-bar / d2. You pass d2. This package does not look up d2 for you except to point at the handbook when you copy a table.
+
+Assumptions. Rational subgroups, normal model for the d2 relationship. NIST/SEMATECH 6.3.2.1: the mean of the range is d2 * sigma. https://www.itl.nist.gov/div898/handbook/pmc/section3/pmc321.htm
+
+Worked example. Test vector: R-bar = 1.128, d2 = 1.128. sigma hat = 1.128 / 1.128 = 1. The value 1.128 is d2 for a span of 2, stated on NIST/SEMATECH 6.3.2.2. https://www.itl.nist.gov/div898/handbook/pmc/section3/pmc322.htm
+
+## X-bar and R
+
+Definition. For subgroup size n, using factors you take from the table above or pass in yourself:
+
+- Upper X-bar = X double bar + A2 * R-bar
+- Lower X-bar = X double bar - A2 * R-bar
+- Center X-bar = X double bar
+- Upper R = D4 * R-bar
+- Lower R = D3 * R-bar
+- Center R = R-bar
+
+Leave a negative X-bar lower limit as that negative number. Floor a negative R lower limit at 0 and say the formula went below 0. The worked example above stays positive, so it does not show that floor. A negative center stays negative. X double bar = -5, R-bar = 2, n = 5, A2 = 0.577 gives upper X-bar = -3.846 and lower X-bar = -6.154. X double bar = -1 with the same R-bar and A2 gives lower X-bar = -2.154. Do not floor those lower limits.
+
+Assumptions. Shewhart three-sigma limits. Normal model for the factor derivation. Rational subgroups. NIST/SEMATECH 6.3.2.1. https://www.itl.nist.gov/div898/handbook/pmc/section3/pmc321.htm
+
+Worked example. Test vector: n = 5, A2 = 0.577, D4 = 2.115, D3 = 0, X double bar = 50, R-bar = 4. Upper X-bar = 50 + 0.577 * 4 = 50 + 2.308 = 52.308. Lower X-bar = 50 - 2.308 = 47.692. Upper R = 2.115 * 4 = 8.46. Lower R = 0.
+
+## Individuals and moving range
+
+Definition. Moving range is the absolute difference of successive points. d2 for span 2 is 1.128.
+
+- Upper individuals = x-bar + 3 * MR-bar / 1.128
+- Lower individuals = x-bar - 3 * MR-bar / 1.128
+- Center = x-bar
+- Upper moving range = D4 * MR-bar with D4 = 3.267 from the n = 2 row
+- Lower moving range = 0
+
+Leave a negative individuals lower limit as that negative number. The moving-range lower limit stays 0. x-bar = -1 and MR-bar = 0.5 give lower individuals = -1 - 1.5/1.128, about -2.330. Do not floor it.
+
+Assumptions. NIST/SEMATECH 6.3.2.2. The moving range uses successive pairs. https://www.itl.nist.gov/div898/handbook/pmc/section3/pmc322.htm
+
+Worked example. Test vector: x-bar = 10, MR-bar = 2. Upper = 10 + 6/1.128. Lower = 10 - 6/1.128. Upper moving range = 3.267 * 2 = 6.534. Lower moving range = 0. In a spreadsheet, 6/1.128 is about 5.319149, so the individuals limits are about 15.319149 and 4.680851. Keep the fraction 6/1.128 as the exact form.
+
+## p chart
+
+Definition. For equal subgroup size n:
+
+- Upper = p-bar + 3 * square root of (p-bar * (1 - p-bar) / n)
+- Lower = p-bar - 3 * square root of (p-bar * (1 - p-bar) / n), then 0 if that result is negative
+- Center = p-bar
+
+Assumptions. Binomial model, independent units, constant n. NIST/SEMATECH 6.3.3.2. https://www.itl.nist.gov/div898/handbook/pmc/section3/pmc332.htm
+
+Worked example. Test vector: p-bar = 0.1, n = 100. Square root of (0.1 * 0.9 / 100) = square root of 0.0009 = 0.03. Upper = 0.1 + 0.09 = 0.19. Lower = 0.1 - 0.09 = 0.01.
+
+## np chart
+
+Definition. Center = n * p-bar. Upper = center + 3 * square root of (n * p-bar * (1 - p-bar)). Lower = center - that same quantity, then 0 if negative.
+
+Assumptions. Same binomial model as the p chart. The mean of the count is n * p and the variance is n * p * (1 - p), as stated in NIST/SEMATECH 6.3.3.2. https://www.itl.nist.gov/div898/handbook/pmc/section3/pmc332.htm
+
+Worked example. Test vector: n = 100, p-bar = 0.1. Center = 10. Square root of (100 * 0.1 * 0.9) = square root of 9 = 3. Upper = 10 + 9 = 19. Lower = 10 - 9 = 1.
+
+## c chart
+
+Definition. Upper = c-bar + 3 * square root of c-bar. Lower = c-bar - 3 * square root of c-bar, then 0 if negative. Center = c-bar.
+
+Assumptions. Poisson counts, equal area of opportunity. The normal approximation is the usual basis for these limits. NIST/SEMATECH 6.3.3.1. The handbook example is this vector. https://www.itl.nist.gov/div898/handbook/pmc/section3/pmc331.htm
+
+Worked example. Test vector: c-bar = 16. Square root of 16 = 4. Upper = 16 + 12 = 28. Lower = 16 - 12 = 4. Clamp vector: c-bar = 1. Square root of 1 = 1. Upper = 1 + 3 = 4. The formula lower limit is 1 - 3 = -2, so the reported lower limit is 0.
+
+## u chart
+
+Definition. Upper = u-bar + 3 * square root of (u-bar / n). Lower = u-bar - 3 * square root of (u-bar / n), then 0 if negative. Center = u-bar. Here n is the opportunity size for this limit. One n per calculation. A different subgroup size needs another calculation.
+
+Assumptions. Poisson counts per unit of opportunity. Same three-sigma Poisson form as NIST/SEMATECH 6.3.3.1, per unit. https://www.itl.nist.gov/div898/handbook/pmc/section3/pmc331.htm
+
+Worked example. Test vector: u-bar = 4, n = 4. Square root of (4 / 4) = 1. Upper = 4 + 3 = 7. Lower = 4 - 3 = 1.
+
+## Sample size for a proportion
+
+Definition. n = z squared * p * (1 - p) / E squared, then the smallest integer greater than or equal to that value.
+
+Assumptions. Normal approximation for a margin of error. Write "margin of error, not power". E must be greater than 0. p is between 0 and 1.
+
+Worked example. Test vector: z = 2, p = 0.5, E = 0.1. n = 4 * 0.25 / 0.01 = 1 / 0.01 = 100.
+
+Reference. Classical normal-approximation sample size in the ASQ Certified Six Sigma Black Belt body of knowledge. https://www.asq.org/cert/six-sigma-black-belt
+
+## Sample size for a mean
+
+Definition. n = (z * sigma / E) squared, then the same ceiling.
+
+Assumptions. Sigma treated as known. Margin of error, not a power calculation. E and sigma greater than 0.
+
+Worked example. Test vector: z = 2, sigma = 5, E = 1. n = (10 / 1) squared = 100.
+
+Reference. Same margin-of-error algebra as the proportion sample size, in the ASQ Certified Six Sigma Black Belt body of knowledge. https://www.asq.org/cert/six-sigma-black-belt
+
+## Known-sigma confidence interval
+
+Definition. Interval from x-bar - z * (sigma / square root of n) to x-bar + z * (sigma / square root of n).
+
+Assumptions. Sigma known. Normal model, or a large sample if you say that is the basis. You supply z. This package does not choose 1.96 for you.
+
+Worked example. Test vector: x-bar = 10, sigma = 2, n = 16, z = 2. Standard error = 2 / 4 = 0.5. Margin = 2 * 0.5 = 1. Interval = 9 to 11.
+
+Reference. Known-sigma interval in the ASQ Certified Six Sigma Black Belt body of knowledge. https://www.asq.org/cert/six-sigma-black-belt
+
+## Confidence interval with t
+
+Definition. Interval from x-bar - t * (s / square root of n) to x-bar + t * (s / square root of n). t is the critical value you supply.
+
+Assumptions. Sigma unknown. You bring t from a table you trust. This package does not look up t.
+
+Worked example. Test vector: x-bar = 10, s = 2, n = 16, t = 2. Standard error = 0.5. Margin = 1. Interval = 9 to 11. The 2 is your input, not a table lookup.
+
+Reference. Interval with a caller-supplied t, in the ASQ Certified Six Sigma Black Belt body of knowledge. https://www.asq.org/cert/six-sigma-black-belt
+
+## One-sample t
+
+Definition. t = (x-bar - mu0) / (s / square root of n). Degrees of freedom = n - 1.
+
+Assumptions. Independent observations. Normal model, or you say it was not checked. Sigma estimated from this sample. No p-value in this version. If you also supply a critical value, say whether the absolute statistic exceeds it. Do not look up the critical value here.
+
+Worked example. Test vector: x-bar = 10, mu0 = 8, s = 2, n = 16. Denominator = 2 / 4 = 0.5. t = (10 - 8) / 0.5 = 4. Degrees of freedom = 15.
+
+Reference. One-sample t in the ASQ Certified Six Sigma Black Belt body of knowledge. https://www.asq.org/cert/six-sigma-black-belt
+
+## Chi-square goodness of fit
+
+Definition. Chi-square = the sum of (observed - expected) squared / expected. Degrees of freedom = k - 1 - m, where k is the number of classes and m is the number of parameters you estimated from the data. You pass m. Each expected value must be greater than 0.
+
+Assumptions. Independent counts. Expected counts large enough for the approximation you intend. If an expected value is 0, do not report the statistic.
+
+Worked example. Test vector: observed 10, 20, 30. Expected 20, 20, 20. m = 0. Terms: (10 - 20) squared / 20 = 5, (20 - 20) squared / 20 = 0, (30 - 20) squared / 20 = 5. Chi-square = 10. Degrees of freedom = 3 - 1 - 0 = 2.
+
+Reference. Chi-square goodness of fit in the ASQ Certified Six Sigma Black Belt body of knowledge. https://www.asq.org/cert/six-sigma-black-belt
+
+## One-way ANOVA
+
+Definition. Grand mean, group means, between sum of squares (SSB), within sum of squares (SSW), mean squares, and F = MSB / MSW. Degrees of freedom between = groups minus 1. Degrees of freedom within = observations minus groups.
+
+Assumptions. Independent groups. Normality not tested by this arithmetic. Equal variance not tested by this arithmetic. No p-value in this version.
+
+Worked example. Test vector: groups {1, 2, 3} and {4, 5, 6}. Grand mean = 21 / 6 = 3.5. Group means = 2 and 5. SSB = 3 * (2 - 3.5) squared + 3 * (5 - 3.5) squared = 3 * 2.25 + 3 * 2.25 = 13.5. SSW = (1 - 2) squared + (2 - 2) squared + (3 - 2) squared + (4 - 5) squared + (5 - 5) squared + (6 - 5) squared = 1 + 0 + 1 + 1 + 0 + 1 = 4. MSB = 13.5 / 1 = 13.5. MSW = 4 / 4 = 1. F = 13.5. Degrees of freedom between = 1. Degrees of freedom within = 4.
+
+Reference. One-way ANOVA in the ASQ Certified Six Sigma Black Belt body of knowledge. https://www.asq.org/cert/six-sigma-black-belt
+
+## Simple linear regression
+
+Definition. Slope = sum of (x - x-bar) * (y - y-bar) / sum of (x - x-bar) squared. Intercept = y-bar - slope * x-bar. R squared = 1 - (residual sum of squares / total sum of squares).
+
+Assumptions. A straight line is your choice. If every x is equal, do not report a slope. If n is under 3 and the x values are not all equal, you may still report the fit and you flag small n. No p-value in this version.
+
+Worked example. Test vector: (1, 2), (2, 4), (3, 6). x-bar = 2. y-bar = 4. Numerator = (-1) * (-2) + 0 * 0 + 1 * 2 = 4. Denominator = 1 + 0 + 1 = 2. Slope = 2. Intercept = 4 - 2 * 2 = 0. Fitted values match y, so residual sum of squares = 0 and R squared = 1.
+
+Reference. Simple linear regression in the ASQ Certified Six Sigma Black Belt body of knowledge. https://www.asq.org/cert/six-sigma-black-belt
+
+## Takt time
+
+Definition. Takt = available time / customer demand, for the same period.
+
+Assumptions. Both numbers cover one period. Demand is the customer's demand, not the rate you wish you could run.
+
+Worked example. Test vector: 480 minutes available, 240 units demanded. Takt = 480 / 240 = 2 minutes per unit.
+
+Reference. Takt time in the ASQ Certified Six Sigma Black Belt body of knowledge (Lean). https://www.asq.org/cert/six-sigma-black-belt
+
+## OEE
+
+Definition. OEE = availability * performance * quality, each a ratio from 0 to 1.
+
+- Availability = run time / planned time
+- Performance = (ideal cycle time * total count) / run time
+- Quality = good count / total count
+
+A factor outside 0 to 1 is not clamped. Stop and check the inputs.
+
+Assumptions. The three ratios describe the same period.
+
+Worked example. Test vector: 0.9, 0.8, and 0.75. OEE = 0.9 * 0.8 * 0.75 = 0.54.
+
+Reference. OEE in the ASQ Certified Six Sigma Black Belt body of knowledge (Lean). https://www.asq.org/cert/six-sigma-black-belt
+
+## Little's Law
+
+Definition. Average inventory = average throughput * average flow time, in consistent units.
+
+Assumptions. The averages describe the same stable period. Units of throughput and flow time must match (per day with days, not per day with hours, unless you convert).
+
+Worked example. Test vector: 20 units per day, flow time 2 days. Inventory = 20 * 2 = 40.
+
+Reference. Little's Law in the ASQ Certified Six Sigma Black Belt body of knowledge (Lean). https://www.asq.org/cert/six-sigma-black-belt
+
+## RPN
+
+Definition. RPN = severity * occurrence * detection, only when all three integers are present. Each score is an integer you enter from 1 to 10.
+
+Assumptions. The scale is the one your FMEA sheet uses. A blank score means a blank RPN. Do not average a missing score into a 1 or a 10.
+
+Worked example. Test vector: severity 8, occurrence 4, detection 5. RPN = 8 * 4 * 5 = 160.
+
+Reference. FMEA RPN in the ASQ Certified Six Sigma Black Belt body of knowledge. https://www.asq.org/cert/six-sigma-black-belt