SQL WORKBENCH · CACHED STATE · BUILD 2026-08-22

Run the argument, query by query.

Architecture

SQL workbench

Six curated queries walk from raw scale to the study's null result and its one honest counterexample. Every result on this screen is cached from the full local build — the workbench replays it exactly, and says so.

This workbench renders pre-aggregated, cached views of the committed Iceberg snapshot; the full Spark batch results appear in the analysis exhibit below.

01 · Establish the data scaleCACHED
-- How large is the current contract-checked silver data plane?

SELECT
  'silver.interactions' AS dataset,
  count(*) AS row_count,
  count(DISTINCT user_id) AS entity_count
FROM silver_interactions
UNION ALL
SELECT
  'silver.items' AS dataset,
  count(*) AS row_count,
  count(DISTINCT parent_asin) AS entity_count
FROM silver_items
ORDER BY dataset
LIMIT 20

179,901,744 rows scanned · 1922.04 ms · cached · build 2026-08-22

datasetrow_countentity_count
silver.interactions43,365,42418,286,190
silver.items1,610,0121,610,012
2 rowsDownload JSON
ICEBERG
snapshotId
884031112460958161
committedAt
2026-08-05
schema
v2
rowCount
43,365,424
files
50
bytes
1.36 GB
The six queriesa narrative index

NDCG@10 BY HISTORY DEPTH · AMAZON ELECTRONICS + MOVIELENS-32M

One curve never crosses. The other crosses at n*=20.

Same metric, same paired-bootstrap discipline, two catalogs. The gap between a personalized model and recency-weighted popularity either closes or it doesn't — these two panels are that comparison, not a summary of it.

AMAZON ELECTRONICS · POPULARITY VS ALS

No crossover on Amazon

ALS stays below recency-weighted popularity at every observed history depth. The gap narrows from 1–4 interactions to 20+, but never reaches zero.

Amazon NDCG at 10 by history depth0.00000.00450.008901-45-910-1920+
pop-t12m20260805T172047Z-035042bALS (rank 128, seed 20260805)20260806T082441Z-2f2f26d
series01-45-910-1920+
pop-t12m0.00750.00580.00520.00510.0037
ALS (rank 128, seed 20260805)0.00000.00330.00290.00250.0023

MOVIELENS-32M · POPULARITY VS ITEM-KNN

The crossover appears at n*=20

With 6.40% catalog churn, trailing-window item-kNN overtakes popularity from 20 interactions onward on NDCG@10, after Benjamini–Hochberg correction.

MovieLens-32M NDCG at 10 by history depth0.000.260.53n*=2001-45-910-1920+
pop-t12m20260820T221055Z-20d8ff9item-kNN t12m20260820T221701Z-20d8ff9
series01-45-910-1920+
pop-t12m0.44120.36100.28730.24570.0922
item-kNN t12m0.05680.34710.28360.20170.1052

SOURCE · RECEIPTS

Every curve opens
the same six runs.

What was verified
Recorded evaluations retain the Amazon null result and the ML-32M crossover in ranking quality.
Evidence class
Single-machine batch evaluations with frozen splits and individual run receipts; the SQL workbench replays cached results.
Boundary
Recall did not confirm the ML-32M winning buckets. The dataset contrast is non-causal and provides no online or production evidence.
Files, hashes and methods

The Amazon curve resolves to runs 20260805T172047Z-035042b and 20260806T082441Z-2f2f26d; the 41.11% mechanism resolves to 20260817T095926Z-633d454.

The ML-32M curve resolves to runs 20260820T221055Z-20d8ff9 and 20260820T221701Z-20d8ff9. The page projection preserves config hashes, dataset hashes, snapshot IDs, seeds, hardware, and wall-clock time for all six source runs.

20260805T172047Z-035042bpopularity
git SHA
035042bbc4ce5284a4b7afd46f5dca574b5ddee4
Recorded
2026-08-05T17:30:18.009394+00:00
Config
configs/eval_pop_t12m_test.yamlLucisZhang/crossover-study · 035042bbc4ce
config SHA-256
sha256:1e60f2621f29dda053119282bfd8dbc3c824df0af7d95b8d9aeca2f86a1500ed
dataset SHA-256
sha256:4a553c348bfccb05918362192692ed5fe6965e3e2662fe751cf08d3b9f9da849
Frozen split
2026-08-05
Runtime
570.663 s · arm64 · Darwin
20260806T082441Z-2f2f26dals
git SHA
2f2f26d55129eabadfd84012146c8a202d61e1c2
Recorded
2026-08-06T08:46:12.872525+00:00
Config
configs/eval_als_test_seed1.yamlLucisZhang/crossover-study · 2f2f26d55129
config SHA-256
sha256:c50679bf1b89ed1bb82d7914b089a887a6ff91332c228b5ff67724966606d31e
dataset SHA-256
sha256:4a553c348bfccb05918362192692ed5fe6965e3e2662fe751cf08d3b9f9da849
Frozen split
2026-08-05
Runtime
1292.102 s · arm64 · Darwin
20260817T095926Z-633d454regime_map
git SHA
633d4543605b68e03a6e770ade46764f598dd5e9
Recorded
2026-08-17T09:59:26.571790+00:00
Config
configs/regime_map_test.yamlLucisZhang/crossover-study · 633d4543605b
config SHA-256
sha256:32c1c06a860eeb62ae42fe0c1c8e5c7d2b528aeb65d1a0fcf59adffbf0366806
dataset SHA-256
sha256:4a553c348bfccb05918362192692ed5fe6965e3e2662fe751cf08d3b9f9da849
Frozen split
2026-08-05
Runtime
41.002 s · arm64 · Darwin
20260820T134403Z-e2263d2churn_contrast
git SHA
e2263d24e03a0a61d2743314f60cdd21c289dadb
Recorded
2026-08-20T13:44:03.648361+00:00
Config
configs/churn_contrast_ml32m.yamlLucisZhang/crossover-study · e2263d24e03a
config SHA-256
sha256:682d2e89e2c86cd71d46c43b844b70fd54f5f6b3a92ec7cd1285a6599cf3505f
dataset SHA-256
sha256:1f830512534c39f3768e2f8b03a05d950c77fbb927e51b072a9eb541432e98a2
Frozen split
2026-08-19
Runtime
26.124 s · x86_64 · Linux
20260820T221055Z-20d8ff9popularity
git SHA
20d8ff995f4baceda31a9eedfc78effd632099ce
Recorded
2026-08-20T22:11:55.106578+00:00
Config
configs/eval_pop_t12m_ml32m_test.yamlLucisZhang/crossover-study · 20d8ff995f4b
config SHA-256
sha256:0e8c7581986745417dfc534edfc0132a547303c80dbe8d90c9989002d31ee9b0
dataset SHA-256
sha256:1f830512534c39f3768e2f8b03a05d950c77fbb927e51b072a9eb541432e98a2
Frozen split
2026-08-19
Runtime
58.601 s · x86_64 · Linux
20260820T221701Z-20d8ff9item_knn
git SHA
20d8ff995f4baceda31a9eedfc78effd632099ce
Recorded
2026-08-20T22:18:30.876460+00:00
Config
configs/eval_itemknn_t12m_ml32m_test.yamlLucisZhang/crossover-study · 20d8ff995f4b
config SHA-256
sha256:21212ee94a62fb72e684abc3a6d83355c96be6f257603996ca0e40c118d15d9a
dataset SHA-256
sha256:1f830512534c39f3768e2f8b03a05d950c77fbb927e51b072a9eb541432e98a2
Frozen split
2026-08-19
Runtime
88.445 s · x86_64 · Linux

docs/evidence/digits-crossover.md pins every number on this page to one of these six runs or to the cached workbench's own committed JSON files.

Full demo source

public/case-studies/crossover-study/exhibits.jsonLucisZhang/portfolio-site · acf05ae78859

Results & negatives

Amazon produced no crossover at any observed depth. The mechanism measurement found 41.11% catalog churn. Under the same ladder on MovieLens-32M, where churn was 6.40%, item-kNN crossed popularity at n*=20 on NDCG@10.

  1. On Amazon, ALS trails trailing-12-month popularity at every observed history depth. The deficit narrows from −0.00256 at 1–4 interactions to −0.00146 at 20+, but never reaches zero. The fitted policy is n*=∞: build no personalization-routing layer for this regime.

  2. On ML-32M, item-kNN turns positive from 20 interactions onward on NDCG@10 after Benjamini–Hochberg correction. Recall@20 confirms none of the three winning buckets, so the result is ranking quality only; it is not a recall win.

  3. The same ML-32M arm loses −0.3844 at zero history, a bucket containing 43.9% of test users. Popularity still wins globally. The crossover changes the deep-history tail; it does not reverse the deployment default.

Limitations

  1. The Amazon null applies to this catalog, five-core population, temporal split, and observed history depths — not to recommenders in general.

  2. Amazon and ML-32M differ in domain, density, catalog size, and feedback semantics. Their contrast measures a regime association; it does not isolate churn as a cause.

  3. Single-machine batch evaluation only: no serving system, online metric, or A/B test. Raw reviews, per-user arrays, and MiniLM weights are not shipped with this page.