I — Datum
SVEND — Journeyman tooling for quality engineers
Statistical process control, capability analysis, measurement system analysis, hypothesis testing and regression, reliability, and value stream mapping with discrete-event simulation. 174 analyses, calibrated against NIST certified reference datasets.
Svend is Danish for journeyman: the craftsman who has finished the apprenticeship, works the trade daily, and is not yet a master. You earn it by making something and having it judged. This is tooling for the people doing that work, built the same way: in the open, on real lines, and still being judged. Read the colophon →
SVEND is also the consulting practice of Eric Wolters: on-site operational excellence, quality management and strategic planning at $2,000 per day, across South Carolina, North Carolina and north Georgia. Kaizen facilitation, MDI deployment, hoshin kanri, SMED and ISO 9001 deployment. ILSSI accredited, SCMEP Technical Partnership Provider. The practice →
II — Capability Manifest
| Capability | Specification | Status |
|---|---|---|
| Statistical Process Control | Shewhart & Bayesian control charts, Nelson rules, adaptive limits, CUSUM, EWMA, online changepoint detection | ● |
| Process Capability Analysis | Cp, Cpk, Pp, Ppk, Bayesian Cpk with credible intervals, P(Cpk > 1.33), sigma level, DPMO | ● |
| Design of Experiments | Full factorial, fractional factorial, Plackett-Burman, CCD, Box-Behnken, Taguchi, power analysis, response optimization | ● |
| Measurement System Analysis | Gage R&R (crossed), ANOVA method, %GRR, ndc, linearity & bias, ICC, Bland-Altman | ● |
| Statistical Analysis | 174 analyses: ANOVA, regression, hypothesis testing, reliability, survival, Bayesian methods | ● |
| Machine Learning | Classification, regression, clustering, feature importance, model comparison | ● |
| Time Series & Forecasting | ARIMA, SARIMA, decomposition, changepoint and drift detection | ● |
| Value Stream Mapping | Current/future state pairing, simulation, cycle time analysis, OEE, kaizen burst promotion into a vandrebok | ● |
| Vandrebok | A practitioner's book of improvement work: chartered entries promoted from a map, provenance back to the burst, action items, commitments and over-commitment detection | ● |
| Design of Experiments | Full and fractional factorial, Plackett-Burman, definitive screening, CCD, Box-Behnken, Taguchi, with randomized run sheets | ● |
| Managed Datasets | Upload, profile, clean and transform, with a version chain each analysis can cite | Nov 2026 |
| Audit Trail | Tamper-evident hash chain, 21 CFR 11 §11.10(e) construction. Live over gage study readings; analytical records to follow | ● |
Capabilities marked ● are active in production. Dated rows are in build and are commitments rather than aspirations. Tier availability noted in the pricing schedule.
III — Live Examples
Interactive charts rendered by ForgeViz. Hover for values.
Bearing inner diameter, 25.000 ± 0.050 mm, one part, three outputs.
Three worked results. Every number below was computed by the engine that ships rather than written into the page.
| Worked example | Result | What it establishes |
|---|---|---|
| Gage R&R, AIAG shape 10 parts × 3 appraisers × 3 trials = 90 readings |
%GRR 10.78% ndc 13 |
Repeatability 5.89%, reproducibility 9.02% — the appraisers disagree with each other more than the gage disagrees with itself, which makes this a training finding rather than an instrument one. A synthetic study carrying a deliberate appraiser offset, and the variance decomposition puts it exactly where it belongs. How the bench runs it → |
| I-MR control chart shaft diameter, 12 sequential readings |
CL 25.0016 UCL 25.0120 |
Sigma 0.0035 estimated from the moving range, LCL 24.9912, verdict IN CONTROL. Runnable on the supplied data or a substituted dataset — no account, nothing stored. Six CC0 datasets and a live engine → |
| Our own calibration against values generated by R |
114 of 114 CALIBRATED |
4 of 4 packages pass every check, so the certificate reads CALIBRATED. The rule is all-or-nothing, and the verdict is published whichever way it falls. The certificate, including anything that fails → |
IV — Workflow
| 1 | Map current state | VSM, line simulation |
| 2 | Form & test hypotheses | Synara, Bayesian tracking |
| 3 | Prioritize the opportunities | Waste register, proposal review |
| 4 | Charter improvement projects | Kaizen charters, project tracking |
| 5 | Run analysis | 174 analyses, SPC, capability |
| 6 | Verify effectiveness | Before/after, capability re-study |
| 7 | Confirm strategic impact | Dollar rollup, VSM lifecycle |
The cycle repeats. Updated VSM reveals new constraints. New experiments begin.
V — Calibration
| Parameter | Value |
|---|---|
| External certification | NIST Statistical Reference Datasets — 14 certified datasets, values certified to 15 significant digits |
| Control charts & capability | Checked against R qcc — 28 of 28 checks |
| Hypothesis tests & regression | Checked against base R (t.test, aov, cor.test, chisq.test, lm) — 49 of 49 checks |
| Golden-file calibration cases | 149 held in the statistics library alone |
| Tolerance bands | p-value ±0.005, Cpk ±0.05, correlation ±0.02 |
| Platform audit | Daily automated suite against a catalog of 38 SOC 2 controls |
| Audit trail | Hash-chained, tamper-evident — gage readings today |
| Public certificate | svend.ai/calibration — the calibration run's own record, verdict included |
We treat our software like a calibrated instrument. The reference values are not ours: NIST certifies the datasets, R computes the cross-check, and a result that misses either fails the build rather than reaching a customer.
Read the calibration certificate →VI — Pricing Schedule
Every license carries every bench. A license determines seat count and duration: thirty days of evaluation against your own data with no run limit and no seat cap, then $95.00 per seat per year. One Minitab seat is twenty of ours.
| Characteristic | Evaluation Period | Seat License |
|---|---|---|
| Price | $0 for 30 days | $95.00 /seat /year |
| Seats | All nominated users | As many as are bought |
| Analyses | No limit | No limit |
| Benches | All 174 | All 174 |
| On expiry | Read-only; records remain | Renewable annually |
Compliance evidence, record retention and services are priced on the estimate. Licenses are purchased online and take effect on payment.
VII — Questions
How much does a typical quality tool stack cost?
Minitab licenses run approximately $1,900 per seat per year. Add a separate value stream mapping tool and a project-tracking tool and the cost per quality engineer climbs well past that, before any integration work. SVEND puts SPC, capability analysis, Gage R&R, hypothesis testing, regression, reliability, value stream mapping and improvement tracking on one bench at $95.00 per seat per year.
How do I know the statistical results are correct?
The reference values are not ours. Statistical results are calibrated against NIST Statistical Reference Datasets: 14 certified sets, values certified to 15 significant digits. Control charts and capability are cross-checked against R’s qcc package (28 of 28 checks), hypothesis tests and regression against base R (49 of 49). 149 golden cases are held in the statistics library alone, with defined tolerance bands: p-values within ±0.005, Cpk within ±0.05, correlation coefficients within ±0.02. A result that misses a certified value fails the build rather than reaching a customer. Separately, platform configuration is audited daily by an automated suite against a catalog of 38 SOC 2 controls.
Is my data private?
Yes. Your process data is never used to train models, never shared across tenants, and never accessed by SVEND personnel without explicit written authorization. Sensitive fields are encrypted at rest and all traffic is encrypted in transit. Tenants are isolated at the database level by Postgres row-level security, and accounts support multi-factor authentication. A tamper-evident, hash-chained audit trail over analytical records is in build for November 2026.
What is Bayesian SPC?
Bayesian SPC applies Bayesian statistical methods to process monitoring and capability analysis. Instead of producing a single point estimate for Cpk, Bayesian SPC produces a posterior distribution: a credible interval that quantifies uncertainty. This supports statements such as “there is a 95% probability that Cpk exceeds 1.33” rather than relying on a single number that may be unreliable with small samples. A Bayesian control chart incorporates prior knowledge about process behavior, which can make it more sensitive to real shifts and less prone to false alarms than a traditional Shewhart chart when process history is available. SVEND carries both approaches, and is specific about where: Bayesian capability with credible intervals and exceedance probabilities, a Bayesian Cpk trajectory, a Bayesian EWMA chart, and a dedicated Bayesian family covering t-tests, ANOVA, regression, proportions and changepoints. The remaining classical control charts — X-bar R, I-MR, the attribute charts, CUSUM — have no Bayesian counterpart and are not presented as having one.
What is a vandrebok?
A vandrebok is the book a journeyman carried through the travelling years, stamped and signed by every master he worked under: a running attested record of what he had actually done rather than a certificate. In SVEND it is a practitioner’s book of improvement work: chartered entries promoted from a value stream map, each carrying which map and which kaizen burst produced it, the action items inside it, and who has committed to them. It is loosely two existing ideas put together: hoshin kanri’s chartered, cascaded, measured discipline, attached to the Harada method’s self-reliant owner. The difference from hoshin kanri alone is ownership: hoshin deploys strategy downward through a company, while a vandrebok belongs to a person and travels with them between employers. It also does not estimate its own value: an improvement is tied to a measurable on the map, and the map reports what happened. The full explanation is here →
Can I cancel or decline renewal?
Yes. Seat licenses run for a year and renew annually; renewal can be turned off in account settings at any time, and the license runs to the end of the term already paid for. After that the account returns to the free tier, which keeps the workbench and your datasets at five analyses a day. Records remain. On account closure, data stays available for export for 90 days.