README.md

---
license: other
language:
- en
tags:
- seo
- content-performance
- data-warehouse
- tabular
- education
- flyrank-internship
pretty_name: FlyRank Internship — Warehouse Star Schema (Pseudonymized, Gated)
size_categories:
- 10M<n<100M
extra_gated_prompt: >-
  By requesting access you agree to the FlyRank Internship Data Use Terms:
  anonymized research and education use only; no attempt to re-identify clients,
  domains, queries, keywords, or content; no redistribution of the raw data; and
  no client-identifying data in any public output (case study, repo, chart, or demo).
extra_gated_fields:
  Name: text
  Email: text
  Affiliation or cohort: text
  I agree to the FlyRank data-use terms: checkbox
configs:
- config_name: dim_clients
  data_files: dim_clients.parquet
- config_name: dim_content
  data_files: dim_content.parquet
- config_name: fact_content_daily_performance
  data_files: fact_content_daily_performance/**/*.parquet
- config_name: fact_content_query_90d
  data_files: fact_content_query_90d.parquet
---

# FlyRank Internship — Pseudonymized Warehouse Release (v20260703)

The open-ended, warehouse-shaped dataset (~**81.8M** rows; daily fact
78,835,655 rows) for advanced capstone work. Star schema with salted, namespaced,
fingerprinted hash keys. Built from warehouse **v2 full history** (frozen snapshot,
export date 2026-07-03): an unbalanced panel — per-client history depth differs;
see `dim_clients.gsc_data_start` / `ga4_data_start`.

| Table | Rows | Grain |
|---|---:|---|
| `dim_clients` | 104 | one row per pseudonymized client |
| `dim_content` | 519,606 | one row per pseudonymized content item |
| `fact_content_daily_performance` | 78,835,655 | one row per report date, pseudonymized client, and pseudonymized content item |
| `fact_content_query_90d` | 2,414,248 | one row per pseudonymized client, content item, and query hash over the fixed 90-day window |

`fact_content_daily_performance` is partitioned by `month=YYYY-MM`. The `_sample`
file is the latest full month — start there.

`fact_content_query_90d` is query-level: salted query hashes over a fixed 90-day
window with last-30/prev-30 sub-windows; the rare tail (< 10 impressions) and
Google-anonymized impressions are preserved as per-content aggregate shares.

## Load

```python
from datasets import load_dataset
ds = load_dataset("FlyRank/internship-warehouse", "fact_content_daily_performance", streaming=True, split="train")
```

Or DuckDB (works in Colab; no full download needed):

```python
import duckdb
con = duckdb.connect()
con.execute("CREATE SECRET (TYPE huggingface, TOKEN 'hf_your_read_token')")  # accept the gate in-browser first, then paste a READ token
rel = "hf://datasets/FlyRank/internship-warehouse"
con.sql(f"SELECT COUNT(*) FROM read_parquet('{rel}/fact_content_daily_performance/**/*.parquet')")
```

## Terms

Anonymized research and education use only. No re-identification attempts, no
redistribution, no client-identifying output. Hash keys are salted and namespaced;
the salt never ships.