Runs 2–5 internal QA findings

Date: 2026-06-22
Scope: Internal consistency only (no new features).
Answer key: Not read during this QA pass.

Summary

Area	Result
Missing deliverable files	Pass — all planned artefacts present
Markdown / repo links	Pass — no broken relative links in warehouse-demo or examples
HTML hrefs (agent-operating-model)	Pass — 15 warehouse-related targets exist
JSON pipeline specs	Pass — 4/4 valid JSON
`site/reports/` modified	Pass — not modified
`draft-reports.html` modified	Pass — not modified
Answer-key leakage (agent outputs)	Pass — no planted counts in profiling report; forbidden lists only elsewhere

Issues found and disposition

Fixed (low-risk)

ID	Issue	Fix applied
QA-01	Corrected HTML brief showed Mar operational cases 220 but `demo_provider_month_measures.csv` has 197 for same filter	Updated `reports/urgent-care-provider-month-brief.html`
QA-02	Jan operational cases showed em dash; mart CSV has 274	Same HTML table row updated
QA-03	`qa.ExpiredLocalOpsMappingInUse` references `stg.LocalOpsUserMapping` — table missing from DDL	Added to `sql/staging/01_stg_tables.sql`
QA-04	Loader/ADF/EXPECTED counts reference `raw.CareCase`, `raw.LegendaryReferral`, etc. — only `raw.CareCallContact` had DDL	Added `sql/staging/02_raw_other_sources.sql`; README run order updated
QA-05	HTML caveat cited `linkage_resolution_strategy.md` without working relative path	Linked to `../design/linkage_resolution_strategy.md`
QA-06	Agent operating model section still titled "in progress" after Run 5	Title updated to "Warehouse design demo"

Documented — no fix (by design or out of scope)

ID	Issue	Notes
QA-07	`qa.MonthlyContactCaseTrend` uses `FULL OUTER JOIN` on month only — can mis-aggregate cases vs contacts	Demo QA view; profiling uses Python `volume_trends.csv` instead
QA-08	SQL DDL incomplete vs design (`staging_model.md` lists many stg/raw tables)	Artefact spec only; not all sources have CREATE TABLE scripts
QA-09	ADF JSON references stored procedures (`stg.usp_`, `mart.usp_`) not defined in SQL folder	Expected for pipeline specs; procedures not in demo scope
QA-10	Flawed brief uses invented open-referral figures (4,200 / 4,350)	Intentional QA trap — not sourced from mart
QA-11	Flawed brief claims "42% surge" for March	Intentional error for Report QA demo
QA-12	`suggested_next_agent_task.md` still describes Run 2 profiling	Superseded by `suggested_run3_agent_task.md`; retained as historical Run 1 handoff
QA-13	`2026-06` partial month in volume/mart outputs	Source data edge rows; profiling includes with caveat

Answer-key references (expected)

References to human-only reviewer checklist (not published on site).md appear only in: forbidden-source lists, human-only checkpoints, comparison template, README reviewer note. No leakage into source_profiling_report.md or agent conversations (except intentional flawed draft errors).

Cross-artifact consistency checks

Check	Status
Mar 2026 cases opened 542	Consistent: volume_trends, mart, profiling report, HTML
Mar IUCS 1,272	Consistent across profiler and mart
Mar cases without SourceContactId 322	Consistent: volume_trends, DQ002, profiling report
Feb operational cases 402	Consistent: mart CSV and corrected HTML (after QA-01/02)
Agency spend Feb ~£61,356	Consistent: volume_trends, mart, HTML
Table names SQL ↔ ADF ↔ design	Core names align (`raw.CareCallContact`, `stg.CareCase`, `mart.ProviderMonthUrgentCare`)
Checkpoint immutability	`run2`–`run4` checkpoints not rewritten during this QA

Re-verification commands

python3 -m json.tool site/warehouse-demo/pipelines/pl_ingest_source_csv.json > /dev/null
cd site/warehouse-demo/profiler && .venv/bin/python profile_sources.py
cd site/warehouse-demo/sql/load && python3 load_from_csv.py --dry-run

QA status: Pass with documented limitations. Low-risk fixes applied.