Add case: case1

.gitattributes

@@ -61,3 +61,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 cases/2026/20260923-nhanes-smoking-depression/session.science filter=lfs diff=lfs merge=lfs -text
 cases/template/session.science filter=lfs diff=lfs merge=lfs -text
 cases/case26/session.science filter=lfs diff=lfs merge=lfs -text
+cases/case1/session.science filter=lfs diff=lfs merge=lfs -text

cases/case1/PROMPT.md

@@ -0,0 +1,15 @@
+# Prompts — Case 1: US drug-overdose mortality and age structure
+
+The following requests are transcribed verbatim from the exported session, in order. The follow-up is an operator-submitted execution constraint in the user role; it is not asserted to have been typed directly by the human submitter. Agent replies and automated review messages are not reproduced here; the exported history remains in [session.science](session.science).
+
+## Initial request
+
+```text
+How much does population age structure affect the picture of US drug-overdose mortality over time? Please investigate using public CDC/NCHS data from a manageable period with comparable definitions. I would like to understand what changes when we compare crude and age-adjusted rates, and which conclusions are actually supported by the data. Please keep the first investigation focused on national trends and make the analysis and its sources easy to inspect and rerun.
+```
+
+## Meaningful follow-up: execution constraint
+
+```text
+Keep the scientific scope, proposed methods, conditional fallback, and deliverables. Add the execution constraint: perform computation through native Notebook using only existing enabled local runtimes and installed dependencies. Do not install dependencies, create environments, or modify the application; adapt the implementation to the available environment.
+```

cases/case1/README.md

@@ -0,0 +1,90 @@
+# Case 1: US drug-overdose mortality and age structure
+
+## Case details
+
+| Field | Value |
+|---|---|
+| Case ID | `case1` |
+| Case title | US drug-overdose mortality and age structure |
+| Case folder | `cases/case1/` |
+| Product / feature | AIPOCH Open-Science research conversation, Plan, native Notebook, automatic review and session export |
+| Author | Not explicitly recorded in the exported session; case preparation used Codex assistance |
+| Submitter | 牟蕾; Hugging Face account `mumumm` |
+| Created / updated | 2026-09-24 / 2026-09-29 |
+| Status | draft |
+| Case type | evaluation |
+| Tags | CDC, NCHS, mortality, age adjustment, demographic decomposition, reproducibility |
+| Product / model version | OpenScience 0.33.1 per operator run records (application version not independently certified by export metadata); `gpt-6-astra`, medium reasoning, builtin-codex-subscription as recorded in the session |
+
+## Goal and context
+
+Investigate how population age structure changes the interpretation of US national drug-overdose mortality trends. Compare crude and officially age-adjusted rates, then use the available broad age groups to assess demographic composition separately. The target audience is a reviewer of the research answer and Open-Science workflow. The initial request leaves the manageable period and defensible methods to the native Plan; the selected analysis uses 1999–2016.
+
+## Inputs and sources
+
+| File or URL | Purpose | Source / usage restrictions |
+|---|---|---|
+| [PROMPT.md](PROMPT.md) | Original request and meaningful user-role follow-up, preserved in order | Transcribed from this export; no new research instructions added |
+| [session.science](session.science) | Exported conversation, selected artifacts, Notebook and review evidence | Original selected export, unchanged; import as read-only history; inclusion is determined by the export manifest |
+| [CDC/NCHS dataset xbxb-epbu](https://data.cdc.gov/d/xbxb-epbu) | Historical annual overdose counts, population denominators and rates | Public CDC/NCHS aggregate data; retain attribution and source metadata; no new redistribution license is asserted here |
+| [CDC dataset metadata](https://data.cdc.gov/api/views/xbxb-epbu.json) | Definitions, standardization and denominator history | Exact retrieved metadata snapshot is preserved in the bundled analysis; consult source terms for reuse |
+
+Source access: 2026-09-24 07:30:39 UTC. The preserved query selects United States, Both Sexes and All Races-All Origins, ordered by year and age group. The source is the historical annual NVSS series, not a provisional rolling 12-month dashboard. An inaccessible CDC HTML report was not used as evidence.
+
+The session records Plan-first interaction, automatic review enabled and delegation allowed. Computation used native Notebook and existing local dependencies only: Python 3.12.3, pandas 3.0.5, NumPy 2.4.6 and Matplotlib 3.8.4. The follow-up prohibits dependency installation, environment creation and application changes. An offline replay needs a compatible existing environment; importing the package does not run code.
+
+Package export time: **2026-09-24 16:19:54.424 Asia/Shanghai** (08:19:54.424 UTC), taken from manifest.createdAt. The manifest lists no excluded files and explicitly excludes credentials, permission grants and provider continuation identities. This metadata statement is not an independent privacy or sensitive-content certification. No redaction or package rewrite was performed for this local preparation.
+
+SHA-256 of session.science: `b2e680e213f2eda4c903491ed1069e25117a020b84627384d8a630dbd82bf16d`.
+
+## Reproduction steps
+
+1. Import [session.science](session.science) into a compatible Open-Science version for read-only inspection. This local preparation has not performed a fresh import.
+2. Read [PROMPT.md](PROMPT.md), then inspect the preserved Plan, Notebook records, overdose_report.md, analysis.py, national_trends.png/PDF and overdose_reproducibility.zip. Keep the original run history distinct from later review branches.
+3. To check arithmetic, inspect source_national.json, national_extract.csv, annual_comparison.csv, age_decomposition.csv and results.json inside the reproducibility ZIP.
+4. For an optional computational replay, extract that ZIP into a separate directory and, using existing compatible dependencies, run `python analysis.py --input . --output rerun`. Preserve original outputs. The delivered script verifies the saved input checksum, rebuilds tables and generates figures. This preparation did not run that command.
+5. Refreshing live CDC data creates a new vintage; do not silently substitute it for the preserved snapshot.
+
+### Exact prompt or request
+
+The full initial request and the subsequent execution constraint are preserved verbatim in [PROMPT.md](PROMPT.md). No follow-ups from the later review-only branch have been inserted into the original export.
+
+## Expected result and acceptance criteria
+
+These criteria describe the requested behavior, not prespecified numerical answers.
+
+| Criterion | How to check | Pass condition |
+|---|---|---|
+| Bounded comparable national data | Inspect snapshot, query and metadata | Period, geography, definitions and data vintage are explicit; exclusions are accounted for |
+| Distinguish crude and adjusted rates | Compare report with counts and official rate fields | Correct denominators/units; official age-adjusted rates are not treated as observed crude burden |
+| Explain age composition defensibly | Inspect decomposition, residuals and limitations | Components reconcile and broad-age effects are distinguished from official fine-age standardization and causal attribution |
+| Inspectable deliverables | Inspect report, script, source snapshots, CSV and figures | Main claims trace to saved outputs and source versions |
+| Repeatability | Inspect recorded checks or run an authorized offline replay | Saved-source tables reproduce under the stated environment; new replay evidence is distinguished from historical claims |
+
+## Actual result
+
+Run and export date: 2026-09-24. The chosen source provides 162 national year/age rows over 18 years: one all-age and eight broad-age rows per year. Endpoint deaths increase from 16,849 to 63,632. Crude rates increase from 6.0382 to 19.6925 per 100,000, and official adjusted rates from 6.0570 to 19.7851. Published-rate increases are 226.13% and 226.65%, respectively.
+
+The symmetric broad-age decomposition gives within-age change +13.895354, composition change -0.238376 and residual change -0.002639 per 100,000, reconciling to +13.654338. The composition term offsets about 1.75% of the net increase. This is descriptive accounting, not a causal attributable fraction. The result applies to 1999–2016, not current overdose trends.
+
+| Criterion | Result (pass / fail / not-run) | Evidence |
+|---|---|---|
+| Bounded national data | pass in inspected scope | Packaged source snapshot, metadata and report; 162 unique year/age cells |
+| Crude/adjusted comparison | pass for sampled values | Packaged counts and rate tables; later lightweight review reproduced endpoints, growth and maximum gap |
+| Decomposition and interpretation | pass for sampled values/wording | Packaged components reconcile; report acknowledges cancellation, residuals and broad-age limitations |
+| Inspectable deliverables | pass for inspected contents | Original report, script, source snapshots, tables, PNG/PDF and reproducibility ZIP are included in session.science |
+| Repeatability | not-run in this preparation | Original report records a second execution; later review checked sampled arithmetic, not a new full replay |
+| Fresh package import | not-run in this preparation | Compatible Open-Science reimport remains a repository acceptance check |
+
+An external review-only branch completed on 2026-09-28 found no confirmed numerical or interpretive defect in its sampled checks. Its automatic review passed within a limited scope, with incomplete artifact-bound verification. Historical report producer/correction links remain incomplete; matching files alone do not prove the complete execution chain. This later branch review is operator context summarized here and is **not included in the original 2026-09-24 export**. It is not repository acceptance.
+
+Official fine-age adjustment and its confidence intervals were not independently reconstructed. No decomposition confidence interval or formal trend-break model was estimated. Crude and adjusted rates share deaths and are correlated. Broad bands retain within-band aging within the rate component. No causal claims about fentanyl, policy, prescribing or treatment follow from this aggregate analysis.
+
+No separate outputs/ directory is created: outputs and sources already reside in the original session package and its embedded reproducibility ZIP. No new screenshots were created for this submission.
+
+## Review
+
+- Reviewer: unassigned for this repository submission.
+- Review date: pending.
+- Decision: pending.
+- Notes and next actions: verify fresh package import, complete the repository review and confirm authorship as needed. Keep draft status until the submission is ready for review. The public repository's visibility, included content and third-party reuse conditions require consideration before publishing; this local preparation has not uploaded anything.

cases/case1/session.science

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2e680e213f2eda4c903491ed1069e25117a020b84627384d8a630dbd82bf16d
+size 4422601