check-csvs

name	check-csvs
description	Use when the user wants a data-quality review of the generated CSV files in public/, especially after fetch-data or stitching changes. Audit the current CSV set for duplicates, date/order issues, suspicious jumps, stitch-boundary continuity, and cross-file consistency using DataFetch.md and scripts/fetch-data.ts as the source of truth.

Check CSVs

Use this skill to sanity-check the generated CSV files in public/.

Current files to audit:

• etf-upro.csv
• etf-tqqq.csv
• etf-sso.csv
• etf-qld.csv
• index-sp.csv
• index-nq.csv
• risk-sgov.csv
• risk-vgsh.csv
• risk-gldm.csv
• risk-brka.csv
• rate-borrow.csv
• inflation.csv

Do not expect rate-tbill-1m.csv or rate-treasury-2y.csv; those were removed.

Source of truth

Before reviewing anomalies, read:

• DataFetch.md
• relevant sections of scripts/fetch-data.ts

If the prompt, docs, and code disagree, treat the code as the source of truth.

Checks

1. Structural sanity

   • No duplicate dates.
   • Dates strictly increase.
   • Expected columns exist for each file type.
   • Values are finite and positive where required.

2. Gap sanity

   • Daily files: gaps should be plausible market/weekend/holiday gaps.
   • Inflation: monthly cadence should be consistent.
   • Flag long unexplained gaps.

3. Stitch-boundary continuity

   • Focus on source/name transitions, not normal movement inside one segment.
   • Check that boundary prices are continuous after scaling.
   • Use name and source changes plus fetch-data.ts boundary logic to identify expected stitch points.

4. Suspicious jumps

   • ETFs / indices: flag implausible one-day jumps.
   • SGOV / VGSH: very small daily moves expected.
   • GLDM: moderate commodity moves only.
   • Borrow rate / inflation: gradual changes expected.
   • Use common sense; review context before calling a move wrong.

5. Cross-file consistency

   • Leveraged ETFs should broadly move with their underlying index direction.
   • SGOV and VGSH should look treasury-like.
   • Stitched proxy behavior should make sense relative to the live asset after transitions.

Review method

• Start with lightweight checks:
  • header
  • first/last rows
  • duplicate dates
  • largest daily moves
  • source/name transitions
• Then inspect only the suspicious windows in detail.
• Prefer explaining whether an anomaly is:
  • expected by stitch design
  • a likely data bug
  • or uncertain and needs manual verification

Output

For each issue found, report:

1. file
2. issue type
3. exact date/rows involved
4. why it is suspicious
5. likely fix or next verification step

If no issues are found, say that explicitly and mention what you checked.