All videos
Antvia Data · Episode 3 of 8

Files people upload, and the columns they carry

Antvia Data treats the spreadsheets people upload by hand as governed data, with an owner and a habit rather than a schedule. This episode reads the upload register, opens a file of every kind, from the orphaned targets file to the one nobody reads, and then moves to the column register: what each column holds, which personal and sensitive columns have nothing protecting them, a policy that exists and is switched off, and three columns that changed under the pipeline without failing it.

7:297 chapters927 wordsSilent, with captions

Also on YouTube, with the rest of the Antvia Data series. The whole episode is written out below, with the timing, if you would rather read it.

Chapters

What happens, in order.

0:00
Introduction
Title card.
0:07
Files with a person, not a schedule
The upload register: why a file is different from a connector, the orphaned targets file at the top, and the columns that describe each file.
1:35
One of each
Upload panels for the orphaned targets file, the distributor region map, the scheme terms that are ageing, the engineer roster kept current, the unused price file and the appointment dates that are old and correct.
3:33
Every governed column
The column register: the four classifications, the summary of unprotected and drifted columns, and the default order that puts exposed personal data first.
4:39
Open to anyone
The distributor owner contact columns with no policy, the customer comments whose policy is switched off, and the distributor credit limit that is sensitive and uncovered.
5:46
Changed under the pipeline
The three drifted columns: a type change that turns every sum to zero, a new column that identifies households, and a column that stopped arriving when its connector broke.
7:12
Close
Closing card.
Transcript

The whole episode, in words.

The videos are silent, with the captions burned in. This is the same text, with the timing, so you can find the part you need without scrubbing for it.

Introduction

0:00
Antvia Data. Files people upload, and the columns they carry. The spreadsheets that run a business, and the personal data hiding in its tables.

Files with a person, not a schedule

0:07
Files with a person, not a schedule. Why uploads get their own register.
0:12
UploadsEvery business of this size runs on spreadsheets: targets, distributor mappings, scheme terms. Antvia Data keeps them in a register of their own, for Acme Home Appliances, a demo company.
0:25
A person and a habitA connector has an owning team, a schedule and somebody who is paged when it stops. An upload has a person and a habit, and nothing fires when the habit lapses.
0:38
The one that mattersSo the page opens on the risk that matters: a file nobody can replace, still being read by three dashboards and models.
0:48
Nobody maintainingOne uploader has left the company.
0:52
Depended onSix of the seven files are read by something, eighteen readers in all.
0:59
Kept in caseAnd one is read by nothing at all.
1:04
A personMaintained by names a person, not a team. That is the point of the column: when the person leaves, the file has nobody.
1:14
The habitMeant to be replaced records how often somebody is supposed to refresh it. Freshness is judged against that, not against a fixed number of days.
1:25
Read byRead by counts the dashboards and models that depend on each file, which is what turns an old file into a risk.

One of each

1:35
One of each. Orphaned, current, ageing, stale and unused.
1:40
OrphanedThe regional targets file. Nikhil Arora uploaded it and was deactivated in June, so nobody has been given the file, and three things still read it.
1:51
Old targets, current labelThe second-quarter revision was due at the start of July and never happened. Three dashboards report percent to plan against first-quarter targets and call it current.
2:02
Where it goesIt was promoted into the region scorecard mart, which is how a spreadsheet ends up behind a leadership number.
2:11
The fixThe fix is not technical. Assign an owner, then replace the file.
2:20
Small and criticalThe distributor region map is 470 rows, and six things read it. A single mistyped region silently moves a distributor between rollups.
2:30
On realignmentIt changes on realignment, about twice a year, so a file last replaced in July is current.
2:39
AgeingThe trade scheme terms are meant to be replaced monthly. August schemes were announced on 1 August and have not been uploaded, and nothing schedules this.
2:53
The healthy caseThe engineer roster was replaced four days ago by the head of service. Somebody whose job includes it, doing it.
3:04
UnusedCompetitor prices collected by hand in June. Nothing reads the file, so it is storage and a governance surface for no benefit.
3:14
Offer to deleteAn unused file is the only kind the panel offers to delete.
3:23
Old and correctAppointment dates, nine months old. Nobody has been appointed since November, so that is correct. Age alone is not a problem.

Every governed column

3:33
Every governed column. What each column holds, and who may read it.
3:43
The column registerDesign a source is where new sources are shaped from a sample file, and where every governed column already lives, 35 of them across six tables.
3:54
Needing protectionEleven columns hold personal or sensitive data. Six of them are covered by a masking policy.
4:02
UnprotectedFive are not. Anyone who can query those tables reads them in full, including the answer engine, which puts them in a grid somebody exports.
4:12
Four kindsHolds is the classification: public, internal, personal or sensitive. Personal identifies a person; sensitive is financial and needs an audit trail too.
4:22
PolicyPolicy is read from the policy register, not typed on the column, so editing a rule updates every column it covers.
4:31
Worst firstThe list opens with exposed personal data, then drift, then everything else by table.

Open to anyone

4:39
Open to anyone. Classified, and left unprotected.
4:43
Distributor ownersDistributor owners' email addresses were added in 2024 for order alerts. Classified as personal; nothing covers them.
4:52
Readable in fullTheir mobile numbers too. Anybody who can query dim_distributor reads them, and the panel says so in those words.
5:02
Switched offCustomer comments are a different failure. A redaction policy covers this column, and it is switched off.
5:10
Worse than noneA rule that exists and does nothing is worse than no rule. Anybody checking whether comments are handled will find that they are.
5:22
SensitiveDistributor credit limits are sensitive, and no policy covers them. A regional head who asked the answer engine was refused by the surface, but a direct query would have worked.
5:36
CoveredFor contrast, customer mobile numbers are covered: masked to the first two and last three digits for anyone without an exemption.

Changed under the pipeline

5:46
Changed under the pipeline. Three kinds of drift, none of which failed a run.
5:51
DriftThree columns changed under the pipeline. None of them failed a run.
5:59
Type changedWarranty claim amounts were numeric until 3 August. The partner portal now exports them with a rupee sign and Indian digit grouping, so they land as text.
6:11
Silently zeroEvery sum on the column now returns zero, and the warranty dashboard shows it without an error.
6:20
AppearedA serial number column appeared on the same day. Nobody asked for it, nothing reads it, and joined to an installation it identifies a household.
6:33
VanishedAnd a column stopped arriving on 30 July, when the channel stock connector broke. It was not removed; it is simply absent, which reads the same to everything downstream.
6:50
One tableSearch narrows to one table. The warranty claims staging table has both a type change and a new personal column, from one portal redesign.
7:03
By classificationOr to a classification. Five sensitive columns: margin, discounts, amounts due, credit limits and the cash discount rate.

Close

7:12
Spreadsheets and columns, governed like everything else. Every upload with a named person and an expected cadence. Files nobody can replace, surfaced before a number goes wrong. Personal and sensitive columns checked against real policies. Drift caught even when every run succeeds.
Next step

See it on your own data.

Thirty minutes on the half you came for. Bring the report that breaks every month, or the decision you never settled.