Student guide

This guide is for students learning assay quality control with Analytical Results. The help pages earlier in this manual say where each control is. This guide says what each control does to your data, what you will see, and why you would use it. It covers the four working tabs in depth:

Words in bold are the labels you see on screen. Terms such as CRM, relative difference and check stage are explained in the Glossary.

How the app is put together

The app works on QC results: the extra samples a project sends to the laboratory alongside its real samples, to check the laboratory’s work.

QC type The sample What it tests The question
Precision A duplicate: the same material split in two and assayed twice. Repeatability. Does the lab get the same answer twice?
Accuracy A certified reference material (CRM): a sample whose true value is certified. Bias. Does the lab get the right answer?
Contamination A blank: a sample with none of the element in it. Carry-over between samples. Does the lab find metal where there is none?

Each result arrives in a despatch: a batch of samples sent to the laboratory together and returned together. Most of the app’s checks look at results despatch by despatch, in date order, because that is how a problem at the lab shows itself: it starts in one despatch and continues in the next.

The app has four working parts:

  1. The data source (top right: Demo or Upload) decides where the results come from: the bundled demo data, or your own Excel or CSV export laid out as the templates. Nothing is sent to a database and nothing changes your file.
  2. The Filters (the band under the tabs) choose which results the Charts, Data and Statistics tabs use: the QC type, the laboratory, the analysis suite, the analyte, the check stage or CRM, and the dates or despatches.
  3. The Overview ignores the Filters. It judges every laboratory, analyte and date in the data at once, against a set of rules you can see and change, and sums the laboratory up as a Lab Confidence of High, Moderate or Low.
  4. The Charts, Data and Statistics tabs draw, list and measure the results the Filters pick out. Every finding on the Overview has a button that opens the chart behind it with the Filters already set.

Everything is worked out again as soon as you change something. The only things remembered between visits are your own Overview rules and chart settings, in your browser.

A first session (about 30 minutes)

Work through this with the demo data before using your own. The demo is three years of real-style QC from two laboratories: 46,104 duplicate results and 39,919 CRM and blank results, to November 2025.

  1. Start on the Overview. The app opens there, on the laboratory with the newest data, ALS. Read the top line: Lab Confidence: Low, and the headline under it: Mo_MEICP41s_ppm: sustained negative CRM bias of −1.4 SD since despatch 17301 (16/09/2024). That one sentence is the laboratory’s most serious finding.
  2. Read the four domain cards. Accuracy (CRMs) is Fail: 9 analytes biased, 19 to watch, of 37. Precision (duplicates) is Warn: none over the action limit, 7 to watch, of 171 analytes and stages. Contamination (blanks) is Fail: 7 analytes contaminated in the latest despatch. Despatch coverage is Fail: no duplicate results from Feb 2024 to Jun 2024. The Key Takeaway puts these in order of importance.
  3. Open the Rules. Click Rules at the top of Lab Health. Every number the verdicts depend on is here: CRM bias fails from ±1 SD, pulp duplicates are allowed ±10% relative difference, a blank fails more than 5× the detection limit above its certified value. Close it with Cancel: you have changed nothing. The Overview guide goes through each rule.
  4. Follow a finding to its chart. In Current Risks, the first row is Mo_MEICP41s_ppm: the latest EXP-3 result is 9.0 SD below certified. Click Investigate. The app sets the Filters for you (Accuracy, ALS, Mo_MEICP41s_ppm, CRM EXP-3) and opens a Shewhart chart with a note above it saying where you came from. The results sit well below the certified value of 3.8: the lab has been reporting this CRM low for a year.
  5. Change the chart. In the Charting menu on the left, set Chart to Cusum. The line falls steadily: a cumulative sum drifts when every result is on one side of the target. This is the chart the Overview used to find the bias. Open How this is calculated under the chart title to see the formula with this chart’s own numbers.
  6. Look at duplicates. In the Filters, set QC type to Precision, Sample check stage to Pulverising Stage and Analyte to Cu_MEICP41s_ppm. The Scatter chart plots each original result against its duplicate; good duplicates sit on the X = Y line. Change Chart to RD v Mean Pairs: now each pair is a single point, its relative difference against its concentration. Pairs beyond the 3 SD line are outliers.
  7. Switch to the laboratory with less data. Back on the Overview, click SGS. It has duplicates only, so Accuracy and Contamination show N/A and the headline is about duplicate precision. Notice that Lab Confidence is still Low: the coverage check fails because six months of despatches have no CRM or blank results at all. Missing QC is a finding too.
  8. Check the file itself. Open the Data tab and the Data health bar at the top: Demo data · 46,104 rows · 2 warnings. One warning is 12,697 duplicate results below the detection limit. Every warning says what is odd, why it matters and what to check. The Data guide lists every check.

By the end you will have seen the three QC types, how a verdict is reached from rules you can read, and how every finding leads to the chart and the rows behind it.

Good habits

  • Start on the Overview, not the charts. With 100 analytes, four check stages and two laboratories there are thousands of charts. The Overview has already looked at all of them and ranked what matters.
  • Read the rule before you trust the verdict. A Fail is only as good as the limit behind it. Open Rules, and ask whether a 10% pulp limit or a ±1 SD bias limit is right for your project and your element. Change it if it isn’t; the app remembers your rules in this browser.
  • Mind the threshold. Relative differences blow up near the detection limit: 0.1 against 0.2 is a 67% difference that means nothing. The Overview only counts pairs with a mean of at least 10× the detection limit; on the charts you set the Threshold factor yourself, and it starts at 0.
  • One suite, one stage at a time. Results from different methods or different sample stages have different precision. Mixing them hides the one that is poor. The Filters start on one suite and one stage for that reason; All Suites is there for when you mean it.
  • Check Data health before anything else with your own file. A result typed as text, a date the app can’t read or a CRM with no certified SD can quietly remove results from every chart. The Data tab tells you what was found and shows the rows.
  • Report the settings with the result. The RD formula, the threshold, the control-line method and the date range all change what a chart shows. How this is calculated lists them for the chart you are looking at.

When the app changes

The app is still developing. This guide is checked against the app each time the manual is rebuilt, and changes are listed in What’s new. If something on screen does not match this guide, the screen is right: please tell us through the Feedback tab.