Update test structure for new role

[santiaro90]

The exercise now carries two challenges, and which one a candidate runs is theirs to pick — the interviewer hands over a project, not a task. The project ships with a staging layer over raw and nothing above it — no intermediate models, no marts, no whitelabel models. What the platform should expose, at what grain, under what names, is the candidate's call.

Challenge 1 — onboard Meridian Live. A MARI portfolio ticketing company lands in raw with multi-currency formatted amounts, order-level grain against TodayTix's payment-level grain, naive local timestamps plus an offset, and no identity merge log. Modeling judgment against messy source data. The grain bullet describes both shapes rather than naming them — TodayTix splitting checkout from payment, Meridian carrying the money on the order row with line items underneath — because naming them alone read as if a Meridian order simply were a transaction.

Challenge 2 — whitelabel storefronts that onboard themselves. Brands write page tracking into their own wl_<brand> schemas in the warehouse. They aren't declared in sources.yml and nothing reads them. The candidate builds the page-load path — and the question is whether they build the version that costs a pull request per brand launch or the version that doesn't. This is the exercise version of page_loads_pxt_schemas / page_loads_pxt_leg in analytics-models.

The sources are built so that "add a name to a list" can't pass:

Source	Why it's there
wl_arcadia.pages	The full standard column set
wl_northgate.pages	Older tracker — showtime_id and utm_* don't exist on the table
wl_lumen.pages	Standard set plus two columns that mean nothing to the platform
wl_sandbox.sessions	Prefixed schema with no pages relation — discovery has to filter on relation name too
partner_orpheum.page_views	Pre-standard: no prefix, own column names, page intent only readable off the URL path. Folding it into the generic path is the wrong answer

A brand missing columns and a brand with extra columns break select * in both directions, so each leg has to be rendered from the brand's actual column set.

How it's graded live. bin/add-partner <brand> creates a new wl_* schema directly in the warehouse — nothing lands in the repo. The interviewer runs it mid-session against the candidate's untouched working tree, then bin/dbt build.

Verified. ./scripts/init.sh builds the 7 staging models clean and loads all five brand schemas unread. With a reference solution applied: bin/add-partner zephyr (reduced columns) and bin/add-partner halloway --full both appear after a plain bin/dbt build, with absent columns null-filled rather than erroring; bin/add-partner zephyr --drop removes it again; wl_sandbox stays out. ./scripts/reset.sh drops brands added mid-session, so the exercise resets between candidates.

Loader fix. Recreating one relation dropped the whole brand schema, so a brand with two relations lost the first one and rerunning add-partner could wipe siblings. It now creates the schema if absent and replaces only the named table — confirmed by adding a table to wl_arcadia, rerunning bin/add-partner arcadia, and seeing both relations with their rows intact. mp_7 also carried a +60 offset for a Bath venue in February, when the UK was on GMT and the same event a week later already read 0.

Rejected alternative. A manifest file of brand slugs in the repo would have been simpler to build and to grade, but it leaves the pull request in place — which is the whole problem the candidate is meant to see.

Rubric and reference solution live in the private ttg repo at _private/drafts/tt-data-exercise-challenge-2-rubric.md, not in this one.

🤖 This was posted by Claude

[Copilot]

Pull request overview

Reworks the exercise around onboarding Meridian Live into the existing dbt platform.

Changes:

• Adds Meridian source fixtures, FX rates, and ingestion support.
• Adds the existing TTG staging, intermediate, and mart baseline.
• Documents the revised architecture-focused exercise.

Reviewed changes

Copilot reviewed 26 out of 31 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
README.md	Defines the Meridian onboarding exercise.
scripts/load_initial_source_data.py	Loads Meridian source tables.
scripts/ingest.py	Ingests incremental Meridian data.
dbt/seeds/fx_rates.csv	Provides currency conversion rates.
dbt/seeds/.gitkeep	Seed-directory placeholder.
dbt/models/sources.yml	Declares TTG and Meridian sources.
dbt/models/staging/stg_accounts.sql	Adds account staging baseline.
dbt/models/staging/stg_events.sql	Adds event staging baseline.
dbt/models/staging/stg_showtimes.sql	Adds showtime staging baseline.
dbt/models/staging/stg_orders.sql	Adds order staging baseline.
dbt/models/staging/stg_transactions.sql	Adds transaction staging baseline.
dbt/models/staging/stg_pages.sql	Adds normalized page staging.
dbt/models/staging/stg_identity_merges.sql	Adds identity-merge staging.
dbt/models/staging/.gitkeep	Directory placeholder.
dbt/models/staging/.gitkeep_meridian	Meridian work placeholder.
dbt/models/intermediate/int_pages_identity_resolved.sql	Adds TTG identity resolution.
dbt/models/intermediate/.gitkeep_meridian	Meridian work placeholder.
dbt/models/mart/dim_customers.sql	Adds customer dimension baseline.
dbt/models/mart/dim_events.sql	Adds event dimension baseline.
dbt/models/mart/fct_transactions.sql	Adds certified transaction fact baseline.
dbt/models/mart/.gitkeep_meridian	Meridian work placeholder.
data/initial/meridian_customers.csv	Supplies customer fixtures.
data/initial/meridian_venues.csv	Supplies venue fixtures.
data/initial/meridian_events.csv	Supplies event fixtures.
data/initial/meridian_performances.csv	Supplies performance fixtures.
data/initial/meridian_orders.csv	Supplies order fixtures.
data/initial/meridian_order_items.csv	Supplies line-item fixtures.
data/initial/meridian_web_sessions.csv	Supplies web-session fixtures.
data/incremental/meridian_orders/batch_001.csv	Adds incremental orders.
data/incremental/meridian_order_items/batch_001.csv	Adds incremental line items.
data/incremental/meridian_web_sessions/batch_001.csv	Adds incremental sessions.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 40 out of 47 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 40 out of 44 changed files in this pull request and generated no new comments.

Suppressed comments (2)

scripts/whitelabel_sources.py:33

• create_relation is relation-scoped by name, but it drops the entire schema. Loading two CSVs for one brand—or running bin/add-partner sandbox against the included wl_sandbox.sessions fixture—deletes the brand's other relations. Recreate only the requested table so unrelated warehouse data survives.
  README.md:20
• This sentence says the listed certified models need not change immediately after proposing that candidates extend those same models. That gives conflicting acceptance criteria for whether Meridian must appear in the existing marts; state explicitly that candidates may change the relevant marts while preserving their contracts.

- **Challenge 1 builds:** the staging → intermediate → mart path for Meridian's raw tables, converging into the existing marts (extending `fct_transactions`/`dim_customers`/`dim_events`, adding new models, or some combination — your call, but be explicit about why) so Meridian's activity shows up alongside TodayTix's in a way stakeholders and downstream consumers can trust. Nothing in the list above needs to change for this.

[Copilot]

Pull request overview

Copilot reviewed 32 out of 36 changed files in this pull request and generated no new comments.

Suppressed comments (2)

scripts/whitelabel_sources.py:33

• create_relation drops the entire schema for every CSV. Consequently, if data/whitelabel/ ever contains two relations for one brand (the module explicitly supports <schema>__<table>.csv files), loading the second file deletes the first relation. Recreate only the target table while preserving other relations in the schema.
  dbt/models/sources.yml:14
• This says the TodayTix sources are already modeled through intermediate and mart, but this PR intentionally leaves those layers empty and README.md:19-20 states that only staging exists. Update the comment so candidates are not given contradictory project-state guidance.

      # ---- Existing TodayTix (TTG) platform — already modeled in staging/intermediate/mart ----

[Copilot]

Pull request overview

Copilot reviewed 32 out of 36 changed files in this pull request and generated no new comments.

Suppressed comments (3)

scripts/whitelabel_sources.py:33

• This helper drops the entire brand schema when recreating one relation. Because load_dir supports one CSV per relation, adding a second CSV for the same brand will cause the later file to erase the relation loaded from the earlier file; add-partner can likewise delete unrelated relations already present in that brand schema. Create the schema if needed and replace only the target table.
  data/initial/meridian_performances.csv:8
• mp_7 and mp_8 are performances of the same Bath event one week apart in February 2025, when the UK was UTC+0 throughout. Giving only mp_7 a +60-minute offset makes its normalized instant one hour early and introduces a DST change that did not occur.

mp_7,me_comedy_night,2025-02-01 20:30:00,60,2025-02-01 19:45:00

dbt/models/sources.yml:14

• This says the existing platform is already modeled through intermediate and mart layers, but this PR intentionally leaves both layers empty and the README states that only staging is built. The source documentation should not tell candidates that those higher layers already exist.

      # ---- Existing TodayTix (TTG) platform — already modeled in staging/intermediate/mart ----

[Copilot]

Pull request overview

Copilot reviewed 32 out of 36 changed files in this pull request and generated no new comments.