Workflow template for business systems and integrations

Database table quality check

Once a month, or before a migration, a report or an automation is built on a table, its owner gets one numbered findings sheet: duplicate keys, blank required fields, values off the allowed list, bad dates and numbers, broken links, each with its rows. Nothing is changed.

View this template on GitHub

Put this on your planBuild this in a sprint

Add it to a workflow in the FullyPowered Planner, beside the teams and roles that will run it. Book a fit call

The outcome

The owner of a table knows exactly what is wrong with it, row by row, before anyone builds on it. Every finding names the field, the problem, how many rows it touches and which ones, with each row's key beside its line number, so each one is quick to find. The sheet opens by counting every row in the export against the count the database tool shows, so nobody has to wonder whether part of the table was skipped.

Duplicates come in two kinds: the same key used twice, and records that look like the same thing under different keys, such as one serial number typed two ways. The sheet lists both and the fields where the rows differ, and never says which row is the real one.

Where the owner's rules are silent, the sheet asks instead of guessing. The owner decides every fix, and a person makes it in the database tool. The workflow template changes no record, merges nothing and deletes nothing.

Who starts the work, and when

The operations or systems owner of the table starts it once a month, and always before the table is moved to another tool, before a report is built on it, or before an automation starts reading or writing it. It suits any table the team keeps its records in: customers, assets, projects, stock, suppliers or members.

What they supply

  • The table export from the database tool your team keeps its records in, as a CSV file with the column names on the first line: every record and every field of one table, unfiltered.
  • The table's rules, written once by the owner and kept for every later run: which fields are required, which field is the key, the allowed values for each field, the format of each date and number, and which field links to which other table.
  • The linked table's export, if you want the links checked; without it, the sheet says that link was not checked.
  • A few settings: the table's name, the date, the reason for the check, the record count the database tool shows, and the fields that point to the same real thing when they match.

What comes out

One numbered findings sheet in a fixed order: the rows read, duplicate keys, likely duplicates, blanks in required fields, values outside the allowed list, dates and numbers in the wrong form, and links that point to no record. Each section says how many values it read and prints a plain line when it finds nothing. Then a summary of findings and rows touched, numbered questions where a rule is missing, what the owner decides, and what was not checked.

If a column name in the export does not match the rules, or the export holds a different number of rows from the one the tool shows, the work stops and says why before any finding is written.

An example

Illustrative example with sample data. The company, the people, the assets, the sites and every value below are illustrative. The company is an illustrative facilities-services business, called Company R here. Its operations manager owns the Assets table the team keeps its equipment records in, and wants the table checked before it is moved to a new database tool. First come the inputs the operations manager supplied, in full, then the findings sheet the workflow template returned.

The inputs

1. Settings (typed by the operations manager)

Table name: Assets
Today: Thursday 19 November 2026
For: the operations manager, owner of the Assets table
Reason: before the table is moved to a new database tool
Files: assets.csv, table_rules.csv, sites.csv (all exported 2026-11-18)
Rows in the database tool: 14
Likely-duplicate fields: serial_number
Placeholders: defaults

2. assets.csv: the table export from the database tool your team keeps its records in (header is line 1)

Every record of the Assets table, with all eight fields, exported unfiltered.

asset_id,asset_name,category,site_id,purchase_date,purchase_cost,status,serial_number
A-1001,Van 1,Vehicle,S01,2023-03-14,28500,In use,VN-20931
A-1002,Van 2,Vehicle,S01,2023-03-14,28500,In repair,VN-20932
A-1003,Pressure washer,Tool,S02,2024-06-02,640,In use,PW-7781
A-1004,Laptop 14,IT,S03,2025-01-20,1150.00,In use,LT-55012
A-1005,Laptop 15,IT,S03,20/01/2025,1150,In use,LT-55013
A-1006,Desk set,Furniture,S03,2022-09-30,,In store,
A-1007,Generator,Tool,S04,2024-11-11,3200 est,In use,GN-1180
A-1008,Ladder,tool,S02,2021-05-17,180,In use,LD-0042
A-1009,,Tool,S02,2021-05-17,95,In use,LD-0043
A-1010,Van 3,Vehicle,S01,2024-02-29,31200,Sold,VN-20940
A-1004,Laptop 16,IT,S03,2025-01-20,1150,In use,LT-55014
A-1011,Pressure washer,Tool,S02,2025-07-15,640,In use,pw 7781
A-1012,Floor scrubber,Tool,,2025-08-01,2100,In store,FS-310
A-1013,Monitor,IT,S03,2025-13-02,210,In use,MN-8820

3. table_rules.csv (written once by the operations manager; header is line 1)

field,required,key,allowed_values,format,links_to
asset_id,yes,yes,,text,
asset_name,yes,no,,,
category,yes,no,Vehicle|Tool|IT|Furniture,text,
site_id,yes,no,,text,sites.site_id
purchase_date,no,no,,date YYYY-MM-DD,
purchase_cost,no,no,,number,
status,yes,no,In use|In store|In repair|Retired,text,
serial_number,,no,,,

4. sites.csv: the linked table's export (header is line 1)

site_id,site_name
S01,Depot North
S02,Depot South
S03,Head office

The findings sheet

TABLE QUALITY FINDINGS: Assets

For: the operations manager, owner of the Assets table. Prepared Thursday 19 November 2026 from assets.csv, table_rules.csv and sites.csv (all exported 2026-11-18). Reason: before the table is moved to a new database tool. Nothing in the table was changed, merged or deleted.

1. ROWS READ
Header check: the export's 8 columns match the rules one for one.
assets.csv: 15 lines; header on line 1; 14 data rows, lines 2 to 15; every row has 8 fields.
Rows in the database tool: 14, matches.
Linked exports: sites.csv: 3 data rows.

2. DUPLICATE KEYS (14 rows read, key asset_id)
F1. asset_id "A-1004": 2 rows, lines 5 (A-1004) and 12 (A-1004). They differ in asset_name (Laptop 14 / Laptop 16), purchase_cost (1150.00 / 1150) and serial_number (LT-55012 / LT-55014); they agree in category, site_id, purchase_date and status.
Blank keys: none.

3. LIKELY DUPLICATES (14 rows read; compared on the key and serial_number)
F2. serial_number: 2 rows, lines 4 (A-1003, "PW-7781") and 13 (A-1011, "pw 7781") read the same once capitals, spaces and punctuation are ignored. They differ in purchase_date (2024-06-02 / 2025-07-15); they agree in asset_name, category, site_id, purchase_cost and status.

4. BLANKS IN REQUIRED FIELDS (5 required fields x 14 rows = 70 values; 68 filled, 2 blank, 0 placeholder)
F3. asset_name: 1 blank, line 10 (A-1009).
F4. site_id: 1 blank, line 14 (A-1012).

5. VALUES OUTSIDE THE ALLOWED LIST (2 fields with a list; 28 filled values read)
F5. category (allowed: Vehicle|Tool|IT|Furniture): 1 row. Line 9 (A-1008) "tool", differs only in capitals or spaces.
F6. status (allowed: In use|In store|In repair|Retired): 1 row. Line 11 (A-1010) "Sold".

6. DATES AND NUMBERS IN THE WRONG FORM (2 fields; 27 filled values read)
F7. purchase_date (date YYYY-MM-DD): 2 of 14 filled values. Line 6 (A-1005) "20/01/2025": written day/month/year, not year-month-day. Line 15 (A-1013) "2025-13-02": there is no month 13.
F8. purchase_cost (number): 1 of 13 filled values. Line 8 (A-1007) "3200 est": holds a word after the number.

7. LINKS THAT POINT TO NO RECORD (1 field with a link; 13 filled values read)
F9. site_id -> sites.site_id (13 filled values looked up in 3 rows): 1 not found. Line 8 (A-1007) "S04".

8. SUMMARY
Findings: 9 (1 duplicate keys, 1 likely duplicates, 2 blanks, 2 outside the list, 2 wrong form, 1 broken links).
Row references: 12. Rows with at least one finding: 11 of 14. Rows with none: 3.

9. QUESTIONS FOR THE OWNER
Q1. asset_name (table_rules.csv line 3): the format is blank, so the field was read as text. Is text the right format?
Q2. serial_number (table_rules.csv line 9): required and format are both blank, so it was checked as not required and read as text. Is it required, and in what format?

10. FOR THE OWNER TO DECIDE
Each finding: correct the record in the tool, leave it, or ask whoever entered it.
No finding waits on a question.
Every fix is the owner's decision and is made by a person.

11. NOT CHECKED
Whether the values are true (that each asset exists, sits at its site and cost what is recorded); assets missing from the table; the site_name field of sites.csv and the quality of sites.csv itself, which have no rules here; the line 7 blank purchase_cost and serial_number, which the rules allow. No record was changed and no tool was connected.

Line numbers count from line 1 of assets.csv as exported on 18 November 2026; the key value of each row is in brackets.

How the figures were reached

  • Rows: 15 lines in assets.csv, less the header, is 14 data rows, the same as the 14 the database tool shows.
  • Required values: asset_id, asset_name, category, site_id and status are required, so 5 fields across 14 rows is 70 values; the blanks are line 10's asset_name and line 14's site_id, leaving 68 filled.
  • Allowed lists: category and status are filled on all 14 rows each, 28 values. Category reads Vehicle 3, Tool 5, tool 1, IT 4, Furniture 1; status reads In use 10, In repair 1, In store 2, Sold 1.
  • Formats: purchase_date is filled on all 14 rows; purchase_cost on 13, because line 7 is blank; 27 in all. Line 11's 2024-02-29 is a real date, since 2024 is a leap year, and line 5's 1150.00 is a number.
  • Links: site_id is filled on 13 rows (S01 3, S02 4, S03 5, S04 1); only S04 is missing from the 3 sites.
  • Rows with a finding: lines 4, 5, 6, 8, 9, 10, 11, 12, 13, 14 and 15, which is 11; line 8 carries two findings (F8 and F9), so the row references come to 12. Lines 2, 3 and 7 have none.

Where it fits

It checks a table against rules the owner has written. To decide what fields an app needs in the first place, use app data requirements. For an analyst's first look at an unfamiliar extract, use dataset profile; this one is the owner's recurring check. To explain why two reports give different numbers, use metric definition reconciliation.

What has to change for your company

Write the rules file once per table and keep it beside the table; it is the part that makes later runs quick. Export the whole table, not a filtered view, and use the column names exactly as the tool writes them, since a mismatch stops the work on purpose. Decide which fields point to the same real thing for each table, such as a serial number for assets or a supplier code for stock, and list the words your team types instead of leaving a field blank. Name who receives the sheet, who decides each fix and who has edit rights to make it. It runs in ChatGPT, Claude, Copilot or Gemini from files you supply, needs no connection to your database tool, and never asks for a login.

What has been checked

FullyPowered AI wrote this workflow template. It was editorially reviewed on 19 November 2026. It was tried once on its example on 19 November 2026 in Claude; the record is published with it, and lists the small edits made afterward that have not been run again. Only the first of its eleven acceptance tests has been run. It has not yet been tested in a client environment. It has not yet been tested as a packaged version in ChatGPT, Claude, Copilot or Gemini.

Testing in your environment means running it on a copy of a table your owner already knows well, with faults planted and written down in advance: a key used twice, the same thing entered under two keys, a required field left blank, a value off the list, a date in a second format, a link to a record that does not exist, and a renamed column that should stop the work. The owner confirms each finding against the table before anyone relies on the sheet. Pilot acceptance-test preparation sets those cases up, and what we test before we call an AI workflow ready describes the record we keep.

Who decides

People decide

Your people keep

  • Each fix, made by a person in the database tool
  • Which of two look-alike rows is right
  • What the missing rules should say
By design

It does not

  • Changes no record
  • Merges or deletes nothing
  • Connects to no database tool

Questions people ask

Who is this for?

The operations or systems owner of a table the team keeps its records in. They run it monthly, or before a migration, a report or an automation is built on the table.

What do I need to supply?

An export of the whole table, the table's rules written once by the owner, and optionally the export of a linked table.

Does it change my records?

No. A person fixes the table in the database tool, after the owner decides.

What if a column name does not match the rules?

It stops and lists every difference before writing any finding, rather than guessing which column is which.

What has been checked?

It was editorially reviewed and tried once on an example. It has not yet been tested in a client environment.

What it connects to

The business systems this workflow template works with, what it reads from each and what it writes to it, from a reviewed reading of its own text.

Writes to it: nothing.

What it connects to
How it works with the systemReads from it
Works from an export from a database, spreadsheet or low-code app builder.Reads from itEvery record and field of one table, unfiltered.

The full workflow template

Everything above is the outline. The complete workflow template is the working version: the roles, the inputs, the procedure, the exceptions and stop conditions, the output format, the reviewer checklist, the acceptance tests, the instructions for an AI assistant and what to customize for your company. It comes with a download and the link to its folder in FullyPowered’s public repository.

View the full template

Free with an account. Sign in by email: no password and no card.

Next step

Put this workflow template to work for your team: on your plan, or built with you in a sprint.

  1. 01

    The full workflow template

    Free with an account. Sign in by email: no password and no card.

  2. 02

    FullyPowered Planner

    Add it to a workflow in the FullyPowered Planner, beside the teams and roles that will run it.

  3. 03

    Build this in a sprint

    In a sprint, FullyPowered adapts this workflow template to your rules, proves it on your own work and puts it in the hands of the roles that own it.

These workflow templates are examples and a starting point, not a complete list. We can build any workflow for your company, including one that runs several steps, in a Workflow Sprint or a scoped implementation.

Rather talk it through first? Book a fit callDiscuss adapting this

FullyPowered Planner. A card to start, everything unlocked, nothing charged for seven days.Security & your informationPlanner termsRefunds