[MC-1] NACHOS automated scoring engine - initial code migration

.gitignore

@@ -12,17 +12,17 @@ dist/
 build/
 
 # Runtime data — raw fetches, enriched outputs, spine manifests are regenerable
-data/raw/
-data/out/
-data/cache/
-data/inline/
-data/explore/
-data/bootstrap/
-data/spine/*
+apps/pipeline/data/raw/
+apps/pipeline/data/out/
+apps/pipeline/data/cache/
+apps/pipeline/data/inline/
+apps/pipeline/data/explore/
+apps/pipeline/data/bootstrap/
+apps/pipeline/data/spine/*
 # …except the Ed-Fi domain map — canonical snapshot committed so fresh clones
 # can build WI/MN spines without re-fetching AZ's swagger first. Regenerate
 # via `mc spine domain-map --source AZ`.
-!data/spine/edfi_domain_map.json
+!apps/pipeline/data/spine/edfi_domain_map.json
 
 # Excel lockfiles
 ~$*.xlsx
@@ -44,10 +44,10 @@ docs/human-scored-files/
 # regime (ADR 0010, issue #213 item 4). Untracked WITHOUT history rewrite;
 # prior revisions remain recoverable from git history. Exact paths only —
 # the methodology spec + data/bootstrap/ source docs stay tracked.
-data/bootstrap/NACHOS[ ]Arizona_20260310.xlsx
-data/bootstrap/NACHOS_Arizona_No_scoring.xlsx
-data/bootstrap/SourceData_Agent_training_20pct_sample[ ]-[ ]origin.xlsx
-data/bootstrap/SourceData_Agent_training_file.xlsx
+apps/pipeline/data/bootstrap/NACHOS[ ]Arizona_20260310.xlsx
+apps/pipeline/data/bootstrap/NACHOS_Arizona_No_scoring.xlsx
+apps/pipeline/data/bootstrap/SourceData_Agent_training_20pct_sample[ ]-[ ]origin.xlsx
+apps/pipeline/data/bootstrap/SourceData_Agent_training_file.xlsx
 
 # Calibration JSON join artifacts — regenerated by scripts/issue108_az_compare.py
 # and similar diagnostic scripts. The writeup .md next to them stays committed.

apps/pipeline/.env.example

@@ -0,0 +1,28 @@
+# Copy to `.env` (gitignored) and fill in real values before running
+# commands that need them. Loaded at CLI entry via python-dotenv —
+# `mc ...` picks it up automatically; `python -c "..."` one-offs still
+# need `export ...` or `source .env`. Exported shell vars always win
+# over .env (override=False).
+
+# Anthropic API key — required for `mc score extract` without --dry-run.
+# Get one at https://console.anthropic.com/settings/keys.
+ANTHROPIC_API_KEY=
+
+# Fill in only one provider block at a time. The runtime uses the first
+# matching configured lane, so keeping a single block populated avoids
+# ambiguity.
+
+# Ed-Fi Azure Anthropic-compatible route for the main scoring pipeline.
+# When both of these are set, src/score/azure_client.py selects the
+# Azure Anthropic Haiku client instead of the default Anthropic client.
+ANTHROPIC_API_HAIKU_ENDPOINT=
+ANTHROPIC_API_HAIKU_KEY=
+
+# Optional Sonnet 4.6 Azure Anthropic-compatible route.
+# If both Sonnet vars are set, they take precedence over the Haiku pair.
+ANTHROPIC_API_SONNET_ENDPOINT=
+ANTHROPIC_API_SONNET_KEY=
+
+# Ed-Fi Azure OpenAI Responses route (parity / alternate client use).
+OPENAPI_API_GPT54=
+OPENAPI_API_GPT54_KEY=