Audit an uploaded dataset in Code Interpreter before it produces a single chart

You are auditing a dataset uploaded here, using Code Interpreter, before doing any actual analysis, chart, or summary statistic on it. Run this audit first and show the results before touching the actual question — do not skip ahead to answering it on unverified data.

DATASET
orders_export.csv — 14 months of e-commerce order records exported from Shopify, roughly 40 columns.

WHAT I ULTIMATELY WANT TO ANALYZE
Monthly revenue trend by product category, to spot which categories are actually growing versus flat.

KNOWN QUIRKS IN THE SOURCE DATA
Refunded orders sometimes show up as a second row with a negative amount instead of modifying the original row — need to confirm whether that's still true in this export.

EXPECTED SCALE
Should be somewhere around 60,000-80,000 order rows based on known order volume for this period.

AUDIT STEPS
Run and show the actual code for each of these checks, not just a prose summary of what the data probably looks like — column names, inferred dtypes, and whether each dtype actually matches what the column should logically be, such as a date column read in as a string, a numeric ID column read in as a float that silently drops a leading zero, or a currency column that's a string because of a stray dollar sign or comma. Row count and a check against the expected scale given — if the actual row count is off by an order of magnitude from what was expected, stop and flag it before proceeding, since that's usually a sign the whole file loaded wrong, not that the data itself is just smaller than expected. Null counts per column, and for any column with unexpected nulls, show a few actual example rows rather than just the count, to distinguish a real gap from a load artifact. Duplicate rows, and separately, duplicate values in whatever column should logically be a unique identifier — these are different problems and need to be reported separately, since a duplicate full row is often a genuine data issue while a duplicate ID with different row contents usually means a join or ingestion bug upstream. Obvious outliers in any numeric column central to the analysis goal — values off by orders of magnitude from the rest of the column, more often a unit mismatch or a data-entry error than a genuine extreme value. If known data quirks were given, verify explicitly whether each one is still present in this specific file, rather than assuming it applies uniformly — a quirk mentioned as historical may already be fixed in this particular upload.

OUTPUT FORMAT
A findings table: Check | Result | Row/column examples if relevant | Severity (blocks analysis, should fix first, informational only). After the table, one paragraph naming anything that should stop the actual analysis until resolved, or an explicit confirmation that nothing found blocks moving forward — do not proceed to the actual analysis goal in the same response unless nothing blocks it.