Learn with Phoebe Β· System & Product Design

Learn Data Product Heuristics with Phoebe

There are at least 32 published principle sets that bear on a data product - usability heuristics, dashboard pitfalls, data-mesh principles, FAIR, DataOps, the golden signals, Microsoft's human-AI guidelines - holding roughly 215 individual principles between them. Nobody reviews anything against 215 principles. This course does what Nielsen did in 1994: keeps the few that carry the weight and throws the rest away. Nine sessions assemble one small review instrument, layer by layer, and run it over Tideline, a subscription-analytics product carrying twenty real defects. A well-trained usability reviewer finds five of them.

9sessions
32principle sets, mapped by source tier
215principles, before the discard
20real defects, detected not scripted
45min per session

The nine sessions πŸ” Β· one review instrument, assembled across all of them

Each session adds one layer's lens and measures what that lens contributed on its own. Two of the nine add nothing at all, which is the most useful finding in the course.

🧭 Session 1 · start here

The canon and the instrument

Where heuristics come from: 249 observed problems reduced to 10, severity on three factors, and the measured moment a usability lens finds 5 of Tideline's 20 defects.

β–Ά Start here
🎯 Session 2 · foundations

The decision it serves

GQM as a derivation and HEART with its Goals-Signals-Metrics step - plus the four frameworks everyone cites that never published a principle list at all.

the named decision
πŸ“œ Session 3 Β· hands-on

Data as a product, and its contract

FAIR's four principles and all fifteen sub-points, Dehghani's eight characteristics, DAMA's six dimensions - run against a contract that really validates as you type.

πŸ§ͺ live validator
πŸ—οΈ Session 4 Β· systems

The platform underneath

Six Well-Architected pillars, the analytics lens, twelve factors applied to a pipeline, and the PACELC question that governs analytics far more than CAP does.

the tradeoff record
⏱️ Session 5 · systems

Operating the promise

All eighteen DataOps principles, the four golden signals translated to a pipeline, and a freshness SLO with an error budget - so lateness becomes a fact, not an argument.

the two new findings
πŸ“ Session 6 Β· hands-on

Truth on the surface

Graphical integrity with the lie factor computed live at 6.5 to 1, Few's thirteen pitfalls in full, and the standard that adds nothing once two others are on.

6 of the 20 defects
πŸŽ›οΈ Session 7 Β· hands-on

The interaction

Five interaction sets worked hard against filters, long queries and error states - including the nineteen principles almost nobody applies to analytics.

the long query
πŸ€– Session 8 Β· advanced

When the product is intelligent

Microsoft's eighteen human-AI guidelines across their four phases, PAIR's six chapters, and Apple's four dichotomies as the triage that says which guidelines are urgent.

rebuild the AI panel
πŸ”Ž Session 9 Β· capstone

The review

The finished instrument over the whole product: twenty of twenty, severity-ordered, then the exhaustive pass that keeps the catches and destroys the precision.

πŸ”’ the report ships
start here foundations hands-on advanced and capstone

Choose your path πŸ—ΊοΈ

Three doors into the same instrument. Everyone starts at session 1, because the method is what makes the rest defensible.

πŸ’Ό I sign off on data products 1 β†’ 2 β†’ 9 the method, the decision, the report you will receive
πŸ—οΈ I build the platform underneath 1 β†’ 3 β†’ 4 β†’ 5 contract, platform, operating promise - the six defects no interface lens sees
🎯 The whole instrument 1 β†’ 2 β†’ 3 β†’ 4 β†’ 5 β†’ 6 β†’ 7 β†’ 8 β†’ 9 from a usability checklist that finds a quarter of it, to a one-page report that gets two fixes funded

Your passport 🎫

Clear all three quiz questions in a session and it gets stamped. Stamps live in this browser only.

What is real here 🧾

A course about reviewing other people's claims has to be precise about its own.

ClaimStatus
The twenty defects in TidelineReal properties of a screen rendered into the page. Every check queries that DOM - there is no answer key. The truncated axis lie factor of 6.5 to 1 is computed from the drawn bar heights, the 13 distinct fills from resolved computed colours, the 64px below the fold from real overflow
The lens ladder: 5, 10, 11, 11, 14, 14, 17, 19, 20Measured in a browser, one lens at a time, counting only findings no earlier lens produced. IBCS and PAIR genuinely add zero on this product
The exhaustive-pass costTwelve additional findings, each one a true statement about the screen and none of them a defect. Precision 100 to 63 percent is arithmetic on those counts
The review-time figures, 66 and 144 minutesModelled, not measured - a flat per-lens and per-finding cost estimate. Labelled as a model on the widget itself
The contract validatorReally parses what you type against 22 checks drawn from FAIR, DAMA and the data-product attributes. The sample scores 7 of 22, and five added lines take it to 14
Norman, Tufte, Gestalt, Kimball and InmonTaught by concept and consequence, never as numbered lists - none of those authors published one, which is why every secondary source gives a different count
Dehghani's 8, DAMA's 6, ISO 9241-110's 7, the AWS analytics lensReal sets whose primary source is paywalled or members-only, so the wording travels through close secondary sources. Flagged as such on the page that teaches each
The lakehouse, decision intelligence, North Star, the AI hierarchy of needsWidely cited as numbered principle sets. Their primary sources enumerate no such list, so this course gives no counts for them and says why

The knowledge map 🧠

The whole course at a glance - hover a session to spotlight its concepts, click any node to jump in.