refactor(027)!: rename ohbm2026→abstractatlas + ohbmcli→aacli (US1, data preserved)

Mechanical, behavior-preserving rename (US1 / the shippable MVP slice):

- `git mv src/ohbm2026 → src/abstractatlas`; surgically rewrite imports +
  `ohbm2026.<module>` refs across src/, tests/, scripts/, .github/, docs/
  (README + CLAUDE.md pre-SPECKIT). Two passes: `(from|import) ohbm2026`
  and `ohbm2026.<X>` EXCEPT `ohbm2026.parquet`.
- pyproject: package name→abstractatlas; `aacli = abstractatlas.cli:main`;
  legacy per-stage `ohbm-*` scripts dropped (most pointed at Stage-5-moved
  modules; functions remain importable).
- Deprecation shims (FR-003/SC-007): `ohbmcli` entry point warns then
  delegates to aacli; `src/ohbm2026/__init__.py` meta-path shim redirects
  `import ohbm2026[.*]` → `abstractatlas[.*]` with a DeprecationWarning.
  Both labeled with removal follow-up.

DATA PRESERVED (FR-004) — NOT renamed: `ohbm2026.parquet`, `/ohbm2026/`
route, `SITE_MODE='ohbm'`, `OHBM2026_*` CI vars, conference `kind`,
`ohbm2026.ui.*`/`analytics` localStorage keys, `--ohbm2026-parquet` flag,
`ohbm2026_state_key`, on-disk paths/state-keys. No data regen, no
site re-publish, byte-identical. Site (TypeScript) untouched — it never
imports the Python package (T007 verified no-op).

Verified: full suite `Ran 1034 — OK (skipped=1)`, identical to baseline
(SC-002); import smoke test OK; constitution-check --full exit 0; residual
`git grep 'ohbm2026|ohbmcli'` = only data-identity, shims, historical
specs. US2 (ingestor arch) + US3 (LinkML schema) follow in later slices.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: satra

.github/scripts/resolve-data-channel.sh

@@ -45,7 +45,7 @@
 # Dropbox shared links OR Cloudflare R2 (e.g. https://aadata.cirrusscience.org/
 # <sha256>/<file>.parquet). This script and the site loader are unchanged — the
 # URL is opaque to both; loader.ts only rewrites Dropbox hosts, so R2 URLs pass
-# through verbatim. `ohbmcli upload-atlas-package` emits a ready-to-paste
+# through verbatim. `aacli upload-atlas-package` emits a ready-to-paste
 # channel entry in exactly the shape above.
 
 set -euo pipefail

.github/workflows/deploy-ui.yml

@@ -30,7 +30,7 @@ on:
       - 'site/**'
       - '.github/workflows/deploy-ui.yml'
       - 'specs/008-ui-rewrite/contracts/references.yaml'
-      - 'src/ohbm2026/ui_data/link_check.py'
+      - 'src/abstractatlas/ui_data/link_check.py'
   workflow_dispatch:
     inputs:
       target:
@@ -191,7 +191,7 @@ jobs:
         # the deploy when a citation 4xx's or times out (T086 / T088).
         # Runs on both targets — a broken reference shouldn't reach
         # sandbox either.
-        run: PYTHONPATH=src python -m ohbm2026.ui_data.link_check specs/008-ui-rewrite/contracts/references.yaml
+        run: PYTHONPATH=src python -m abstractatlas.ui_data.link_check specs/008-ui-rewrite/contracts/references.yaml
 
       # Resolve the parquet URLs from the keyed registry variable, using the
       # channel this ref declares in site/data-channel.json. The same

.github/workflows/pr-preview.yml

@@ -14,7 +14,7 @@ on:
       - 'site/**'
       - '.github/workflows/pr-preview.yml'
       - 'specs/008-ui-rewrite/contracts/references.yaml'
-      - 'src/ohbm2026/ui_data/link_check.py'
+      - 'src/abstractatlas/ui_data/link_check.py'
   workflow_dispatch:
 
 permissions:
@@ -71,7 +71,7 @@ jobs:
         # Same check that gates the production deploy in deploy-ui.yml —
         # mirrored here so PR previews surface a broken reference BEFORE
         # the merge instead of failing at deploy time.
-        run: PYTHONPATH=src python -m ohbm2026.ui_data.link_check specs/008-ui-rewrite/contracts/references.yaml
+        run: PYTHONPATH=src python -m abstractatlas.ui_data.link_check specs/008-ui-rewrite/contracts/references.yaml
 
       # Resolve the parquet URLs from the keyed registry variable, using the
       # channel this PR branch declares in site/data-channel.json. A PR can

CLAUDE.md

@@ -12,7 +12,7 @@ the per-abstract pipeline introduced in Stage 11.1 — each abstract
 pandoc-renders in parallel with content+toolchain caching, then a
 two-pass assembly attaches a page-numbered author index.
 
-DOCX export was **retired in Stage 11.1 US3** (`ohbmcli book --format docx`
+DOCX export was **retired in Stage 11.1 US3** (`aacli book --format docx`
 exits 2 with a pointer at `--format md` and `--format pdf`). The
 real-corpus DOCX hit 2.8 GB even after figure resizing — too large
 for Word to open — and the markdown bundle + PDF cover every use
@@ -39,7 +39,7 @@ live under `specs/011-abstracts-book/`:
   optional Tufte styling).
 - `data-model.md` — three layers (Stage-1 inputs → markdown-bearing
   in-memory model → outputs with figure-filename contract).
-- `contracts/cli.md` — `ohbmcli book` flags, error path table,
+- `contracts/cli.md` — `aacli book` flags, error path table,
   known limitations.
 - `quickstart.md` — operator runbook for install + first run.
 
@@ -53,13 +53,13 @@ uv pip install --python .venv/bin/python ".[abstracts_book]"
 brew install pandoc tectonic
 
 # Markdown-only first run (no system deps needed):
-PYTHONPATH=src .venv/bin/python -m ohbm2026.cli book --format md
+PYTHONPATH=src .venv/bin/python -m abstractatlas.cli book --format md
 
 # Full PDF run:
-PYTHONPATH=src .venv/bin/python -m ohbm2026.cli book --format pdf --sort poster_id
+PYTHONPATH=src .venv/bin/python -m abstractatlas.cli book --format pdf --sort poster_id
 
 # All three formats + Tufte typography for the PDF:
-PYTHONPATH=src .venv/bin/python -m ohbm2026.cli book --format all --sort poster_id --style tufte
+PYTHONPATH=src .venv/bin/python -m abstractatlas.cli book --format all --sort poster_id --style tufte
 ```
 
 Outputs land at `data/outputs/book/book__<state-key>/`. Figure
@@ -108,7 +108,7 @@ Operator flags:
 Single-abstract debug (re-render one chunk in isolation, populating
 the cache):
 ```bash
-PYTHONPATH=src .venv/bin/python -m ohbm2026.book.render_per_abstract \
+PYTHONPATH=src .venv/bin/python -m abstractatlas.book.render_per_abstract \
     --poster-id 0042 --style plain
 ```
 

README.md

@@ -25,8 +25,8 @@ experience overwhelming.
 
 ## Execution
 
-1. Run `ohbmcli analyze-figures` first if you want cached figure notes included in
+1. Run `aacli analyze-figures` first if you want cached figure notes included in
    the claim manuscript.
-2. Run `ohbmcli extract-claims` to populate or resume the claim cache.
-3. Run `ohbmcli enrich` to merge claim extraction into the enriched abstract data.
-4. Run `scripts/build_ui_data.py` (Stage 6) so the SvelteKit site's data package picks up the new fields. (The legacy `ohbmcli export-ui` / `build-ui` commands have been removed; this plan predates Stage 2.1 + Stage 6 and is kept for historical reference only.)
+2. Run `aacli extract-claims` to populate or resume the claim cache.
+3. Run `aacli enrich` to merge claim extraction into the enriched abstract data.
+4. Run `scripts/build_ui_data.py` (Stage 6) so the SvelteKit site's data package picks up the new fields. (The legacy `aacli export-ui` / `build-ui` commands have been removed; this plan predates Stage 2.1 + Stage 6 and is kept for historical reference only.)

docs/abstracts-book-plan.md

@@ -102,7 +102,7 @@ Runtime checks on real data:
 
 ## Status
 
-- [x] Added `ohbmcli cluster-benchmark`
+- [x] Added `aacli cluster-benchmark`
 - [x] Implemented method sweep across `kmeans`, `agglomerative-ward`, `agglomerative-average`, `gaussian-mixture`, and `birch`
 - [x] Added label-independent ranking using silhouette, Davies-Bouldin, Calinski-Harabasz, inter/intracluster distance ratio, and cluster-balance terms
 - [x] Added test coverage for method execution, metric computation, ranking, CLI wiring, and output writing

docs/cllm-claims-plan.md

@@ -37,36 +37,36 @@ This is the operational planning document for the repository.
 
 ## Architecture
 
-- `src/ohbm2026/graphql_api.py`
+- `src/abstractatlas/graphql_api.py`
   - GraphQL requests, batching, retries, env loading, author/submission fetches
-- `src/ohbm2026/assets.py`
+- `src/abstractatlas/assets.py`
   - local asset indexing, figure URL extraction, reuse-aware downloads, JSON-only asset refresh, and ingest CLI
-- `src/ohbm2026/enrichment.py`
+- `src/abstractatlas/enrichment.py`
   - author export, HTML-to-markdown conversion, section mapping, local Qwen figure analysis, enriched abstract assembly, and phase-2 CLI
-- `src/ohbm2026/neuroscape.py`
+- `src/abstractatlas/neuroscape.py`
   - stage-one embeddings, semantic graph generation, clustering, and NeuroScape handoff manifest
-- `src/ohbm2026/cli.py`
-  - unified `ohbmcli` command
+- `src/abstractatlas/cli.py`
+  - unified `aacli` command
 
 ## Unified CLI
 
-- `ohbmcli ingest`
+- `aacli ingest`
   - fetch abstracts and figure assets from the API
-- `ohbmcli refresh-assets`
+- `aacli refresh-assets`
   - rebuild `local_assets` from the local JSON database without rerunning abstract extraction
-- `ohbmcli authors`
+- `aacli authors`
   - export author metadata
-- `ohbmcli enrich`
+- `aacli enrich`
   - build enriched abstracts from local databases
-- `ohbmcli analyze-figures`
+- `aacli analyze-figures`
   - analyze local figures with Ollama
-- `ohbmcli embed-minilm`
+- `aacli embed-minilm`
   - generate local MiniLM embeddings
-- `ohbmcli embed-voyage`
+- `aacli embed-voyage`
   - generate Voyage embeddings
-- `ohbmcli semantic-analysis`
+- `aacli semantic-analysis`
   - build the semantic graph and semantic cluster summaries from any local embedding bundle
-- `ohbmcli write-manifest`
+- `aacli write-manifest`
   - write the NeuroScape handoff manifest
 
 ## Execution Order
@@ -226,7 +226,7 @@ Checks:
 - [x] Create a local MiniLM embedding pipeline and neighbor database.
 - [x] Create a purely semantic graph-and-cluster pipeline for the single-year corpus.
 - [x] Split the codebase into API, asset, enrichment, and embedding modules.
-- [x] Add a unified `ohbmcli` entrypoint.
+- [x] Add a unified `aacli` entrypoint.
 - [ ] Complete a full cached figure-analysis sweep with local `qwen3.5:35b`.
 - [ ] Validate the NeuroScape retraining path against the PR-based workflow and artifact requirements.
 

docs/clustering-analysis-plan.md

@@ -3,8 +3,8 @@
 Every pipeline stage in this repo (Stage 1 fetch, Stage 2 enrich,
 Stage 3+ analyses, …) satisfies the same six contracts. This doc
 defines them and points at two canonical reference implementations
-— **Stage 1 (`src/ohbm2026/fetch_stage.py`)** for a single-fetch
-stage, and **Stage 2 (`src/ohbm2026/enrich_stage.py`)** for a
+— **Stage 1 (`src/abstractatlas/fetch_stage.py`)** for a single-fetch
+stage, and **Stage 2 (`src/abstractatlas/enrich_stage.py`)** for a
 multi-component stage — for each.
 
 A new stage author should be able to read this page plus
@@ -111,7 +111,7 @@ Stage 2 reference:
 What failures the stage surfaces loudly, with what typed cause,
 and at what exit code.
 
-- Typed exception hierarchy in `ohbm2026.exceptions` rooted at a
+- Typed exception hierarchy in `abstractatlas.exceptions` rooted at a
   cross-stage `OhbmStageError(RuntimeError)`. Stage 1 subclass tree:
   `Stage1Error` → `SchemaContractError`, `CheckpointError`,
   `FigureFailureError`. Stage 2 subclass tree: `Stage2Error` →
@@ -219,36 +219,36 @@ Stage 2 shipped the per-stage pattern scaffolding with
 (`specs/004-enrich-production-wiring/`) wires production runners
 through those seams and keeps the orchestrator surface unchanged:
 
-- **Figures**: `src/ohbm2026/stage2_figures.py` —
+- **Figures**: `src/abstractatlas/stage2_figures.py` —
   `run_figure_component(abstract, model_id, flex_enabled, client, cwd)`
   groups all figures of one abstract into a single OpenAI Responses
   API call with manuscript text as context. In-memory JPEG-q85
   compression at 1024 px (`compress_image`) + a four-field local
-  quality probe (`src/ohbm2026/image_quality.py`:
+  quality probe (`src/abstractatlas/image_quality.py`:
   `laplacian_variance`, `mean_brightness`, `native_max_dim`,
   `compression_ratio`) populate `local_quality_estimate` on every
   record. Pydantic `FigureInterpretationResponse` + a fixed
   `model_quality_estimate` enum validate the response server-side.
-- **Claims**: `src/ohbm2026/stage2_claims.py` —
+- **Claims**: `src/abstractatlas/stage2_claims.py` —
   `run_claims_component(abstract, ..., figure_interpretations)`
   issues a single agentic Responses API call with three function
   tools registered (`verify_source_quote`, `lookup_eco_code`,
   `dedupe_check`). The model orchestrates extract → verify →
   annotate → dedupe internally. Post-response validation drops
   claims whose source quote isn't a substring of the manuscript
   OR whose `evidence_eco_codes` include off-vocabulary ECO IDs.
-- **References**: `src/ohbm2026/stage2_references.py` is a thin
+- **References**: `src/abstractatlas/stage2_references.py` is a thin
   adapter to the existing `openalex.collect_reference_metadata`
   async pipeline.
-- **Flex tier**: `src/ohbm2026/flex_tier.py` —
+- **Flex tier**: `src/abstractatlas/flex_tier.py` —
   `call_with_flex_fallback(...)` wraps each LLM call with a 1-flex +
   1-standard retry budget; defaults to flex ON; per-component
   disable flags `--no-flex-figures` / `--no-flex-claims`.
 - **Concurrency**: `enrich_stage._run_component_concurrent`
   fans out per-abstract work across a `ThreadPoolExecutor`
   (default 30 in flight). Per-component caps via
   `--concurrency-figures` / `--concurrency-claims`.
-- **ECO v1 vocabulary** lives at `src/ohbm2026/data/eco_top_codes.json`
+- **ECO v1 vocabulary** lives at `src/abstractatlas/data/eco_top_codes.json`
   (committed source). The claims component's `lookup_eco_code`
   tool draws from it without network access.
 - **Provenance extensions**: per-component `flex_tier_enabled`,
@@ -269,7 +269,7 @@ When you write Stage N:
 2. **Author the test file first** (Principle IV). One test per
    contract element, organized into a class per contract.
 3. **Implement the orchestrator** in
-   `src/ohbm2026/<stage_name>.py`. Use Stage 1's `fetch_stage.py`
+   `src/abstractatlas/<stage_name>.py`. Use Stage 1's `fetch_stage.py`
    as the layout reference. Share helpers via `artifacts.py`,
    `exceptions.py`, and `schema_diff.py` where possible.
 4. **Add a CLI subcommand** in `cli.py` that delegates to your
@@ -280,12 +280,12 @@ When you write Stage N:
 
 ## Common Helpers
 
-- `ohbm2026.artifacts.build_*_path(state_key)` — path derivation
+- `abstractatlas.artifacts.build_*_path(state_key)` — path derivation
   under gitignored roots.
-- `ohbm2026.artifacts.build_dependency_basis` + `build_state_key` —
+- `abstractatlas.artifacts.build_dependency_basis` + `build_state_key` —
   deterministic state key from input fingerprint.
-- `ohbm2026.exceptions.*` — typed exception hierarchy.
-- `ohbm2026.schema_diff.*` — schema-diff / discovery primitives
+- `abstractatlas.exceptions.*` — typed exception hierarchy.
+- `abstractatlas.schema_diff.*` — schema-diff / discovery primitives
   (reusable across stages that talk to GraphQL).
 - The constitution lint at
   `.specify/scripts/bash/constitution-check.sh` catches the