Preparing the portfolio shell before the interactive work loads.
Public project · Deterministic synthetic data
This tool turns feed quality into something teams can inspect: day-over-day catalog continuity, exact-clip recurrence, coverage, RFY alignment, quality issues, and a human-review workflow. Every title, profile, image, and recommendation is fictional.
Why evaluation matters
Aggregate engagement cannot show whether one profile sees healthy discovery while another receives the same titles or exact clips repeatedly. Ordered capture makes those differences visible.
The system keeps title continuity and exact-clip recurrence separate. One reveals catalog-level familiarity; the other is the more precise signal for possible clip fatigue.
Computed synthetic findings
These values are calculated from the frozen synthetic run—not hard-coded marketing claims. They demonstrate what the evaluation can detect, not how a production recommender performs.
Profile A — Explorer to Profile C — Familiarity
Profile R1 — Low recurrence to Profile R3 — High recurrence
76% catalog coverage
646 canonical clips observed
Measurement model
For each current appearance, check whether its fictional title appears anywhere in the same profile’s prior synthetic day.
Use it to understand catalog familiarity and how much of the visible title mix changes.
For each controlled run after the baseline, check whether the same canonical clip ID appeared in any earlier run.
Use it to detect precise recurrence that can contribute to clip fatigue.
For each profile’s first 30 recommendations on Synthetic Day 5, check whether the title appears in its separate 30-title synthetic RFY rail.
Use it to understand how much of the visible stream is represented on another recommendation surface.
Reproducible by design
A versioned JSON manifest and fixed seed regenerate the data, artwork, dashboard metrics, and web findings report together.
Public-release boundary
The public source repository has its own clean history and contains only the generator, fictional schema, static evaluation UI, synthetic raster assets, tests, and public documentation.
The built experience is static and offline-capable. It makes no runtime API calls, loads no third-party artwork or metadata, and stores Review Center choices only in the visitor’s browser.
Inspect the system
The interactive tool, native findings report, and public source all resolve to the same deterministic synthetic run.