Edge cases

The edge cases that break migrations

Bespoke systems and old data produce problems nobody plans for. These are the ones we meet most, and the ones we enjoy. If yours isn't here, we would like to hear about it.

foundrate=650 unit=hour

fixed

Meaning

Rates that changed meaning

What is going on: The same column holds daily rates in one system and hourly rates in another.
What we do: Units are checked against totals before anything loads. In one migration this stopped project costs reading eight times too high.

foundnote="renews 14/2/27, call Sue"

fixed

Meaning

The renewal date lives in the notes field

What is going on: Users worked around a missing field by typing into a free-text one, for years.
What we do: Rules pull dates, phone numbers and emails out into real columns, and every value can be traced back to the note it came from.

foundÂ£1,200.00

fixed

Formats

£ turns up as Â£

What is going on: UTF-8 text read as Windows-1252 somewhere along the way and saved again, so every £ gains a stray character.
What we do: We detect the encoding column by column, repair the damaged values, and count every repair on the sign-off sheet.

foundhours=26.5 cap=24

fixed

Access and rules

Records the new system refuses

What is going on: Archived items it treats as read-only, entries over its limits, values outside its picklists.
What we do: Every record is tested against the target's rules in SQL first. Then we decide with you what to fix, split or hold back.

found03/04/05

fixed

Formats

Three date formats in one column

What is going on: Free-typed dates, where 03/04/05 could be three different days.
What we do: We work out the format per source and per period. Anything still ambiguous goes on a list for a person to decide, never a guess.

found0.1 + 0.2 = 0.30000000000000004

fixed

Formats

Money stored as floating point

What is going on: Pennies drift when currency sits in a float column, and the totals stop agreeing.
What we do: Values are rebuilt as exact decimals and reconciled to the penny against the source's own totals.

foundACME Ltd | Acme Limited | A.C.M.E.

fixed

Structure

Two systems, one customer, no shared key

What is going on: The same company spelt four different ways across two CRMs.
What we do: Matching on normalised names, addresses, email domains and phone numbers, with every merge listed for review before it happens.

found0x255044462D312E34...

fixed

Formats

Attachments stored inside the database

What is going on: Files held as binary columns with no file name or type recorded.
What we do: We identify each file type from its contents, rebuild sensible names, and check every file's size and hash after the move.

foundstatus today: Closed

fixed

Structure

History that only exists as an audit log

What is going on: The system stores today's values; the history you need is buried in a change log.
What we do: We replay the log to rebuild the state at any date, then reconcile the replay against today's values.

foundCC-12 rolls up to CC-12

fixed

Structure

Hierarchies with orphans and loops

What is going on: Tasks with no parent, cost centres that roll up into themselves.
What we do: Every tree is walked and checked before load. Breaks are fixed or listed, never dropped quietly.

founddeleted=1, still on invoices

fixed

Meaning

Deleted, but only on some screens

What is going on: A deleted flag that some parts of the system respect and others ignore.
What we do: We find which records each part of the system really treats as live, and agree the rule with you before anything is held back.

foundno API, no DB, no export

fixed

Access and rules

A system with no export at all

What is going on: No API, no database access, no report worth the name.
What we do: Then we read the screens, or the calls the screens make to their own back end. Slowly, logged, and with the data owner's permission.

Got a strange one?

Describe the system and the problem. The odder it is, the more we want to look at it.

Describe it[email protected]