Student guide: Lab files

The other tabs check QC results that are already in a database.

The other tabs check QC results that are already in a database. The Lab files tab works one step earlier, on the laboratory’s certificate itself, before anything is loaded. A batch that passes can go into the database. One that fails goes back to the lab, with a report that says why.

This page walks through the two example certificates and says what each finding means. They are made up, so you can try everything without real data.

What a certificate is

A laboratory reports a batch of results as a certificate. This is a SIF file, either comma-separated or in fixed-width columns. From the top, it holds:

  • The certificate number, and whether it is Finalized yet.
  • Header rows with the client, the number of samples, the dates received and completed, and the PO number.
  • Column rows with each column’s method, element, unit and detection limit.
  • One row per sample, plus the lab’s own check rows.

The lab names its check rows with a row prefix and the sample. Ch:T0001 is a lab duplicate of sample T0001, St01:GA-CRM-1 is a lab standard, and Blank01 is a lab blank. An extra column of the same element, method and unit, such as Au then Au again, or Au-Rp1000, is a repeat. It is the same pulp analysed again.

A first look (about 15 minutes)

  1. Open the Lab files tab and click Or try the example certificates, which are made up. The app reads two certificates, with example CRM values and an example QC sheet.
  2. Under Batch checks, EXAMPLE00000001 is a Fail, with 2 findings that fail and 8 queries. EXAMPLE00000002 is a Query, with 6 queries.
  3. Above them, Before these can be accepted asks what you haven’t told the app yet. It asks what the result code X means, which appears 4 times, and what the lab’s Blank, Ch, Dup and St rows are. The app doesn’t guess, because X means below detection at one site and not assayed at another.
  4. Set X to below. Leave Ch as a pulverising stage duplicate, and make Dup a Crushing Stage duplicate. Click Use these. The box closes, and nothing is left to confirm.
  5. Read EXAMPLE00000001’s report, below the table.

Reading the report

The report opens with the verdict and one line on what to do. Each finding then gives its level, what is odd, why it matters and what to check.

Level Meaning
Fail Send the batch back to the lab.
Query Ask the lab before loading.
To confirm Confirm what a code or prefix means. The batch can’t be accepted until you do.
Note For information. There is nothing to ask the lab.

EXAMPLE00000001’s findings, and why each matters:

  • A lab blank over 5× the detection limit (Fail). Blank01 reads 15 ppm Cu against a limit of 1 ppm. A blank should read near nothing, so a high one means contamination in the lab’s preparation or analysis, and the samples beside it are in doubt.
  • A lab standard beyond ±3 SD (Fail). GA-CRM-1 reads 790 ppm Cu against its certified 709 ppm with an SD of 21, which is +3.9 SD. The lab’s run was out of control for copper.
  • 12 samples declared, 11 on the certificate (Query). The missing one is on another certificate, still at the lab, or lost.
  • Lab duplicates outside their limit (Query). Ch:T0001 differs from T0001 by 26% for Zn, against ±10% for a pulp duplicate. Dup:T0002 differs by 22% for Cu, outside even the ±20% crushing stage limit you set for Dup. One batch rarely has the 10 pairs needed to check a rate, so the report lists each pair outside.
  • A repeat outside ±10% (Query). T0009’s gold reads 0.50 and then 0.80 ppm. For gold, consider the nugget effect before blaming the lab.
  • Values that aren’t results (Query). T0005 had insufficient sample for gold, and T0006 wasn’t received. T0004’s Zn is above the method’s upper limit with no ore-grade result, so its grade is unknown. T0008 has no Zn result, although the other samples do.
  • A standard not checked (Note). GA-CRM-9 isn’t in the CRM values, so the report can only list it. Add its certified values, and it is checked from then on.
  • No detection limit for Ag (Note). Without it, silver’s duplicates and blanks can’t be checked.

Below detection is never a finding. It is left out, as everywhere in the app, and a blank below detection is a clean blank.

Your own checks

The lab’s checks are named on the certificate. Yours carry ordinary sample numbers, so you say which they are. EXAMPLE00000002’s QC sheet says A1005 is a coarse duplicate of A1004, A1010 is GA-CRM-1, and A1028 is a blank.

  • A1005 against A1004 (Query, one per element). They differ by 26 to 40%, outside ±20% for a coarse duplicate. Either the sample isn’t repeatable at the crusher, or the duplicate came from the wrong sample.
  • A1028 sent but not returned (Query). It falls among the certificate’s sample numbers but isn’t on it.
  • A1010 reads within its certified values, so there is no finding. What was checked, at the foot of the report, still lists it under Your CRMs.

The card These look like the QC samples you inserted then offers three more, from their values alone. A1015 looks like a blank, because it is near detection for every element, where samples usually aren’t. A1020 looks like GA-CRM-2, with all 3 of its certified elements within limits. A1025 looks like a field duplicate of A1024, within 2% where neighbouring samples differ by 49%.

Values only ever suggest. A failed CRM won’t look like its CRM, and that is exactly what QC must catch. Tick the ones that are yours and click Check the ticked ones as my QC.

There are four ways to mark your QC samples: the QC sheet, ticket rules, these suggestions, and by hand. Ticket rules suit ticket books and bar codes, for example a rule that numbers ending 00 are CRMs. Signed in, the app saves ticket rules and the QC sheet with the project.

Taking the report away

Print or save as PDF prints the report as it is shown. HTML and Excel download it. All as Excel puts every certificate and every finding in one workbook, with one row per certificate on its first sheet. The foot of the report states the rules it applied, so whoever reads it knows what “fail” meant.

Keeping certificates (signed in)

Nothing is stored when you select Check only. The app reads the files for the session and then drops them. Keep in a project saves them to a project’s holding area, with the laboratory, despatch number and date sent.

Each certificate waits there, checked again, until you Accept it or Return to the lab. Accepted results open with the project under Projects, on every other tab. A returned certificate stays in the project with its results left out.

The same file can’t be kept twice. A certificate number already in the project is treated as the lab’s re-issue, and it replaces the old one if you agree.

Your set-up saves the CRM library, clients and rules with your account. A consultancy can give each client its own projects, CRM library and rules. For example, one client fails a CRM beyond ±3 SD, and another at 10% from certified. Its Codes and prefixes tab lists the result codes and row prefixes you have saved. There you change a code’s meaning or a duplicate’s stage, or delete one so the app asks about it again.

Questions to try

  1. Before you confirm anything, why does the app refuse to guess what X means? What would go wrong if it took X as below detection at a site where it means not assayed?
  2. Change Dup to a Pulverising Stage duplicate. Which finding changes, and why?
  3. EXAMPLE00000001 fails on one blank and one standard. Would you send the whole batch back, or ask the lab to re-assay part of it? What would you need to know?
  4. Why is a duplicate pair with a mean below 10 times the detection limit not checked?