libfec outputs raw `YYYYMMDD`, FastFEC outputs `YYYY-MM-DD`

[asg017]

Note

Much of this ticket was composed with Claude

Description

libfec's fastfec subcommand passes date values through as raw FEC format (YYYYMMDD), while FastFEC converts them to ISO 8601 format (YYYY-MM-DD). This affects every filing and every form type that contains date fields (coverage dates, contribution dates, disbursement dates, etc.).

Note: FastFEC's date conversion is itself inconsistent - it fails to convert dates on H3 and H4 form types, leaving those as raw YYYYMMDD.

Affected Filing

Filing ID: 1935668 (237KB, Chevron PAC F3XN filing)

Reproduction

libfec fastfec ~/.cache/libfec/cache/1935668.fec /tmp/out_libfec
fastfec --no-stdin ~/.cache/libfec/cache/1935668.fec /tmp/out_fastfec

Observe the Difference

# Cover record date fields:
diff \
  <(sed -n '2p' /tmp/out_libfec/1935668/F3XN.csv | tr ',' '\n' | sed -n '14,15p') \
  <(sed -n '2p' /tmp/out_fastfec/1935668/F3XN.csv | tr ',' '\n' | sed -n '14,15p')
# < 20251201
# < 20251231
# ---
# > 2025-12-01
# > 2025-12-31

# Itemization date fields:
diff \
  <(awk -F, '{print $20}' /tmp/out_libfec/1935668/SA11AI.csv | head -5) \
  <(awk -F, '{print $20}' /tmp/out_fastfec/1935668/SA11AI.csv | head -5)
# < 20251205
# ---
# > 2025-12-05

# Show FastFEC's own inconsistency on H3/H4 (filing 1942420):
libfec fastfec ~/.cache/libfec/cache/1942420.fec /tmp/out_libfec_2
fastfec --no-stdin ~/.cache/libfec/cache/1942420.fec /tmp/out_fastfec_2
awk -F, '{print $8}' /tmp/out_fastfec_2/1942420/H3.csv
# date_of_receipt
# 20251219   <-- FastFEC left this unconverted!
awk -F, '{print $20}' /tmp/out_fastfec_2/1942420/SA11AI.csv | head -2
# contribution_date
# 2025-12-13  <-- But converted this one

Decision

If the goal is exact FastFEC compatibility, add date conversion (YYYYMMDD -> YYYY-MM-DD) in the fastfec writer for all form types. The try_format_fec_date() function already exists in fec-parser/src/lib.rs but is only used by the SQLite export path.

If the goal is correctness over compatibility, raw passthrough is arguably more faithful to the source data.