Prominent data-quality caveat: independent review found missing meetings and inconsistencies; coverage claims in these docs are UNVERIFIED — audit before use
Browse files- DATA-QUALITY-WARNING.md +31 -0
DATA-QUALITY-WARNING.md
ADDED
|
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# ⚠️ DATA-QUALITY WARNING — READ BEFORE TRUSTING ANY COVERAGE CLAIM IN THESE DOCS
|
| 2 |
+
|
| 3 |
+
An independent review of this archive (2026-10-03) identified **missing meetings
|
| 4 |
+
and inconsistencies**. The reviewer's findings were not shared in detail, so the
|
| 5 |
+
specifics are unrecorded here — but the general failure modes below are real and
|
| 6 |
+
were either under-weighted or not independently verified in the phase write-ups:
|
| 7 |
+
|
| 8 |
+
1. **Catalog enumeration was keyword-pattern-based.** The Tucson catalog (75
|
| 9 |
+
meetings) was selected from 1,561 archive.org items by title regex — variant
|
| 10 |
+
titles may have been missed. No independent second-pass enumeration was done.
|
| 11 |
+
2. **Suspiciously small transcripts exist** in the boards corpus (e.g.
|
| 12 |
+
`tucson-boards-49-2026-08-07.jsonl` at 2.4 KB) — possibly dropped audio
|
| 13 |
+
parts or mis-grouped meetings, never investigated.
|
| 14 |
+
3. **The Pima corpus starts Dec 2025.** Older Pima Board meetings exist
|
| 15 |
+
somewhere (OnBase / county site); this archive does not have them.
|
| 16 |
+
4. **Speaker naming is heuristic** (roll-call/hand-off voting, ~68–82%);
|
| 17 |
+
`speaker_name` fields are not authoritative.
|
| 18 |
+
5. **Tucson vote records are absent** (OnBase meeting-level docs are agendas;
|
| 19 |
+
item-level attachments never exercised). Pima votes were pulled but not
|
| 20 |
+
reconciled against minutes.
|
| 21 |
+
6. **No transcript was validated against source audio/video** — WER is
|
| 22 |
+
unmeasured everywhere; the "quality gate" only checks punctuation
|
| 23 |
+
heuristics.
|
| 24 |
+
|
| 25 |
+
**Treat the coverage numbers in the READMEs and runbook (75/75, 53/53, 621/624)
|
| 26 |
+
as pipeline-output counts, not verified completeness.**
|
| 27 |
+
|
| 28 |
+
The pipeline itself is idempotent and its scripts + provenance are on the Hub
|
| 29 |
+
(see `pima-tucson-pipeline-runbook.md` and `RESUME.md` in
|
| 30 |
+
`tucson-boards-commissions`), so a proper audit-and-patch pass is incremental,
|
| 31 |
+
not a rebuild. But that audit has not happened.
|