Silent Data Loss

The failure these share

A load completes. A page of results returns. A document is chunked. Each reports success, and each has quietly lost something — a row, a column, a value, the tail of a payload.

• dlt Schema Contract Verdict — `SUM(amount)` is the same wrong number under three of dlt's four contracts, from three *different* losses. The lossy one is the default.
• Batch Schema Drift — the load SUCCEEDS while the new column is dropped and the retyped one is coerced.
• openpyxl Formula Loss Verdict — reads every formula as `None`, and saving writes the `None` back, emptying them permanently.
• Cursor Pagination Integrity — one insert between page 1 and page 2 and OFFSET skips a record the reader never sees.
• Chunk Integrity Verdict — chunks whose real re-tokenized length is over the budget you asked for.
• Tool Result Size Guard — truncation happens at a byte offset, never a structural one, and the result still parses.

Why buy them together

They fail in the same place: the seam between one system's success and another's input. Fixing one and not the others leaves the same class of loss one step downstream.

Each ships with its own tests and is sold separately. This is the set, below the sum.