Skip to content
allanninal.dev/build
Collaborate

Part 5 of 7 · Packaging data reporter series ~5 min read

The submission, and the seven years behind it

The submission itself is small enough to read on one screen, which is why every business treats it as the deliverable. The deliverable is the part nobody submits: the trail from any row in that file back to a weighing record made three years ago.

Key takeaways

  • Freeze the inputs at period close. A submission built on moving data cannot be reproduced.
  • Reconcile against the previous period before filing, not after somebody queries it.
  • A correction is a new version with a reason, never an edit to the old one.
  • Every submitted row traces to sales lines, weights, methods and documents.
  • Seven years, and the people who did the work will have left.

Closing a reporting period

A reporting period closed, reconciled and submittedA vertical chain inside an AWS account container, entered from a box on the left labelled Period close, the reporting window has ended. Freeze the inputs pins sales, weights and rules. Aggregate groups by material, split and nation. Reconcile compares against the previous period line by line, with a side exit reading greater than ten per cent: Investigate, because an unexplained move is a data error. Produce the file writes the rows that go to the regulator. Write the pack records every row and what is behind it. A note says the reconciliation step is the only one that catches errors before they are filed, and it is the one every manual process skips because there is nothing to compare against.AWS ACCOUNTPeriod closethe reporting windowhas endedFreeze the inputssales, weights, rules,all pinnedAggregateby material, split,and nationReconcileagainst the previousperiod, line by lineInvestigatean unexplained moveis a data error> 10%Produce the filethe rows that goto the regulatorWrite the packevery row, and whatis behind itThe reconciliation step is the only one that catches errors before they are filed, and it is the one every manual process skips because there is nothing to compare against.
Fig 1. Five steps and one branch. The interesting one is the third, and it only works because the previous period was frozen rather than overwritten.
  • Security & identity
  • Management
  • Analytics

Why freezing matters more here than elsewhere

Packaging data has an unusual property: every input to it keeps changing after the period it describes. Suppliers send spec sheets late, which corrects a weight. Sales get credited, which changes a volume. Somebody reclassifies a customer, which moves a tonne across the household line.

A system that recomputes the current answer from current data therefore produces a different figure for last year every time it is run, and none of those figures is the one that was submitted. Freezing at close, and treating later information as a new version with a reason, is what makes the question what did we file, and why answerable at all.

What an audit actually walks

The five step trace from a submitted tonnage back to a source documentA horizontal row of five boxes joined by arrows. A row in the file, for example two hundred and fourteen tonnes of household plastic. The sales lines that made it up. The per-unit weight in grams, by component and material. The method, being supplier specification, weighed, or sampled. And the document, being the spec sheet or the sample record. A note says an auditor picks one row and walks this, and if any step is missing the whole submission stops being evidence and becomes an assertion.FIVE STEPS FROM A SUBMITTED NUMBER TO A PIECE OF PAPERA row in the filee.g. 214 tonnesof plastic, householdThe sales lineswhich shipmentsmade it upThe per-unit weightgrams, by componentand materialThe methodsupplier spec,weighed, or sampledThe documentthe spec sheet orthe sample recordAn auditor picks one row and walks this. If any step is missing the whole submission stops being evidence and becomes an assertion.
Fig 2. The trace, in the order it gets walked. Every arrow here is a stored link rather than somebody’s memory of how the spreadsheet worked.

The threat is turnover, not dishonesty

Nothing in the seven-year retention requirement assumes anybody is trying to cheat. It assumes that the person who weighed forty cases in a stockroom in March will not be there in four years, and that their spreadsheet will have been through two migrations and a rename.

Which is why the evidence is written at the time the number is produced rather than assembled when it is requested. Assembling it later is not merely inconvenient; for a sampled weight it is impossible, because the forty cases are long gone and the only record that they were ever weighed is the record you did or did not make.

The sentence the evidence has to be able to write

  • This row reports 214 tonnes of household plastic for the second half of the year.
  • It comes from 41,900 sales lines across 96 product codes.
  • The largest, code 7714, contributed 38 tonnes from 1.9 million units.
  • Its per-unit plastic weight is 20 grams, from the supplier specification dated 14 March.
  • That specification is attached, and it superseded a sampled figure of 23 grams.
  • The household classification came from the sales channel, which is a consumer retail one.

Corrections are versions

A submission that turns out to be wrong is not edited. A new version is produced, carrying the same period, a reason, and the specific rows that moved. Both versions stay, and any report can be run as at either of them.

This is the same append-only discipline the other compliance systems in this series use, and it earns its keep for the same reason. The question that gets asked in year four is never what is the right number. It is why does this year’s figure differ from last year’s, and that question is unanswerable if last year’s figure was quietly improved.

Where this ends

The system does not make a business compliant, because most of what compliance requires here is measurement that somebody has to actually do. What it removes is the category of failure where the measurement was done and then lost: the weighing session whose results went into a tab, the classification decision nobody wrote a reason for, the year where the whole thing was rebuilt from scratch because the previous version could not be understood.

It also turns a compliance cost into an operational input, which is the part worth copying. A ranked list of suppliers by estimated tonnage is a purchasing document. A number for your own shipping cartons is a packaging design brief. Neither of those exists in a business that files the return and closes the file.

The next post prices it and the one after gives the service names, the tables and the IAM. The interesting thing about the cost is that the model reads specifications, which arrive when the range changes, while the tonnage arithmetic runs monthly over millions of units for essentially nothing.

All posts