Add trimmed manifest type and candidates scaffold

.gitignore

@@ -6,5 +6,8 @@ __pycache__/
 .venv/
 .ruff_cache/
 .pytest_cache/
-cache/
+
+# Aggregator: the full ~50 MB Reservoir manifest is fetched on demand
+# and kept locally only. The trimmed snapshot beside it is committed.
+/aggregator/candidates/raw_data/manifest.json
 

aggregator/candidates/README.md

@@ -0,0 +1,24 @@
+# Candidates
+
+Tracking which Lean packages we are considering for inclusion in
+`lean-pool`, along with the reasoning behind each decision.
+
+## Layout
+
+- `packages/` — one file per candidate package, holding the curated
+  metadata and review status. The per-package files are the source of
+  truth for everything in this directory.
+- `raw_data/` — Reservoir manifest snapshots used as input.
+  - `manifest_trimmed.json` is committed: a slim derivation of
+    Reservoir's bundle (~3 MB) keeping only the fields needed for
+    candidate selection.
+  - `manifest.json` (gitignored) is the full ~50 MB Reservoir bundle,
+    fetched on demand by the aggregator CLI when `--raw-output` points
+    here.
+
+The summary table below is auto-generated from `packages/` and should
+not be edited by hand.
+
+<!-- BEGIN AUTOGENERATED TABLE -->
+<!-- The aggregator will populate this section in a follow-up PR. -->
+<!-- END AUTOGENERATED TABLE -->