ci: pin the measurement_id hash with golden vectors in-repo

.github/workflows/ci.yml

@@ -136,6 +136,22 @@ jobs:
           uv run --all-packages make html
         working-directory: docs/
 
+  # Pins the measurement_id hash that scripts/post-ingest.py uses as the primary key of every
+  # benchmarks-database fact row. The golden vectors are a frozen artifact (see the note inside
+  # the JSON); this job fails if a change to scripts/_measurement_id.py drifts the hash, which
+  # would break ON CONFLICT idempotency against the rows already in production.
+  measurement-id-golden:
+    name: "measurement_id golden vectors"
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    steps:
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0  # v7
+      - uses: spiraldb/actions/.github/actions/setup-uv@a746510eafaa926484c354541cfc49b2ec06cc63  # 0.18.6
+      - name: Pytest - measurement_id golden vectors
+        run: |
+          uv run --no-project --with pytest --with xxhash \
+            pytest scripts/tests/test_measurement_id.py
+
   python-cuda-test:
     name: "Python CUDA (test)"
     if: github.repository == 'vortex-data/vortex'

_typos.toml

@@ -8,7 +8,10 @@ extend-ignore-re = [
 ]
 
 [files]
-extend-exclude = ["/vortex-bench/**", "/docs/references.bib", "benchmarks/**", "vortex-sqllogictest/slt/**", "encodings/fsst/src/dfa/tests.rs", "encodings/fsst/src/dfa/flat_contains.rs", "benchmarks-website/server/static/**", "benchmarks-website/server/tests/snapshots/**"]
+# scripts/measurement_id_golden.json is a frozen hash-pin artifact whose vectors include
+# deliberate Unicode edge cases (JSON-escaped, so "café" reads as the token "caf"); its bytes
+# must never change, so it is excluded rather than "fixed".
+extend-exclude = ["/vortex-bench/**", "/docs/references.bib", "benchmarks/**", "vortex-sqllogictest/slt/**", "encodings/fsst/src/dfa/tests.rs", "encodings/fsst/src/dfa/flat_contains.rs", "benchmarks-website/server/static/**", "benchmarks-website/server/tests/snapshots/**", "scripts/measurement_id_golden.json"]
 
 [type.py]
 extend-ignore-identifiers-re = [

scripts/_measurement_id.py

@@ -1,20 +1,20 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright the Vortex contributors
 
-"""Python port of the server-internal `measurement_id` xxhash64 functions.
-
-This is a byte-for-byte port of the `vortex-data/benchmarks-website` repo's
-`server/src/db.rs` (`measurement_id_*`, `hasher_for`,
-`write_str`, `write_opt_str`, `write_i32`, `write_f64`, `finish`). The v4 ingest
-writer computes the same `measurement_id` the v3 Rust server computes, so
-re-ingesting an existing `(commit, dim-tuple)` upserts the existing row via
-`ON CONFLICT (measurement_id) DO UPDATE` instead of inserting a duplicate.
-
-The equivalence is NOT assumed -- it is pinned by golden vectors generated from
-the Rust source of truth (the `measurement_id_golden_vectors` test in the
-`vortex-data/benchmarks-website` repo's `server/src/db.rs`). The Python port is
-checked against those vectors; any drift in either implementation fails that
-check.
+"""The `measurement_id` xxhash64 functions -- the primary-key hash of the benchmarks database.
+
+This began as a byte-for-byte port of the Rust implementation in the `vortex-data/
+benchmarks-website` repo's `server/src/db.rs` (`measurement_id_*`, `hasher_for`, `write_str`,
+`write_opt_str`, `write_i32`, `write_f64`, `finish`); with that server crate retired, this module
+is now the only implementation. Re-ingesting an existing `(commit, dim-tuple)` upserts the
+existing row via `ON CONFLICT (measurement_id) DO UPDATE` instead of inserting a duplicate, so
+the hash is frozen forever: every row already in the production Postgres carries an id computed
+this way.
+
+The freeze is NOT assumed -- it is pinned by `scripts/measurement_id_golden.json`, whose vectors
+were generated by the original Rust source of truth and must never be regenerated.
+`scripts/tests/test_measurement_id.py` asserts this module reproduces every vector; any drift
+fails that check.
 
 ## The hash, precisely