---
name: flint-record
description: Prepare a November 6, 2015 Flint public-record advisory from the three supplied dated sources, with bounded claims and reproducible checks.
---

# Dated record workflow

1. Set the as-of date to November 6, 2015. Read `SOURCES.csv` for source IDs, document dates, public-by dates, and publication evidence; then read `report.txt` (F1) and `task-force.txt` (F2). Use only these supplied sources. Do not browse, read `later/` or other folders for facts, or use information published after the as-of date. F1 is dated October 21, transmitted November 4, publicly available November 6; F2 is an October 16 public release.
2. Extract F1's dated sample measurements exactly: February 18: 104 µg/L; March 3: 397 µg/L; April 2: 707 µg/L* (F1, printed page 11, Appendix B). Keep the asterisk if quoting the row, and do not invent its meaning. Those three results are from one residence; F1 concerns three residences overall. Cite printed page numbers, not PDF page numbers. Attribute all substantive claims to F1 or F2 with their dates and specific pages or excerpts.
3. Separate observations, recommendations, and institutional commitments. F1 describes service-line risks and recommends notification about disturbances (printed pages 6-7). F2 says EPA's task force will provide technical assistance for reconnection and optimizing corrosion control (October 16 release), not that those steps achieved systemwide safety. State that prevalence outside the described residences and the effect of planned actions remain unestablished by this record. Never identify households, streets, addresses, map locations, or children.
4. Run `analyze.py` to generate the analysis from the supplied sources; inspect the resulting claims and citations against the dated record. Run `check.py` and `check.py --self-test` before release. Resolve any failures without weakening evidence or privacy bounds. Do not present checks as run if they have not been run.
5. Append a new dated, source-grounded decision entry to `memory/decisions.jsonl`; never replace or rewrite earlier entries. Record uncertainties and rationale without asserting a role, deadline, or publishing authority that is not established. Keep `open-questions.md` limited to precise unresolved questions.
6. Produce a short, plain-language public answer that reports the measured values and their narrow scope, explains relevant risk and the announced response, names what is not known, and cites F1 and F2. Do not declare citywide prevalence, completed systemwide safety, or permission to publish.