OpsFulfillment IA research
Same data · same stylesheet · different architecture

The same Monday morning, three ways

96 paid fills across 15 preparations. 2 catalog exceptions. A draft never sent, an order never acknowledged, an order 4 vials short. Every column below is showing exactly that.

A · Queue-first

What needs a person, ranked. Open full size

B · Order-first

The ledger of purchase documents. Open full size

C · Exception-first

Only what has stalled. Open full size

What differs, on five axes

AxisA · queue-firstB · order-firstC · exception-first
Primary object A work item — an action owed A supplier order — a document A stall — something that stopped
Home screen Buckets of pending action The orders index Only what is broken or ageing
Unit of action Batch — many items, one act Document — advance one order Single item — resolve one thing
Screens 7–9, two modes 4 4
Characteristic failure Work outside a bucket is invisible Demand that never reached an order has no home On a good day the home screen is empty

How the ten scenarios landed

3 · 2 · 3, with 2 unwinnable
ScenarioABC
S1 · Cold start Mondaywins2 clickshides the real work
S2 · 100 fills, 15 preparationswinsclosemust leave the concept
S3 · Where is Rx #482?failsfailsfails
S4 · Pins conflict1 clickhomelesswins
S5 · No active pin1 clickhomelesswins
S6 · 4-day-old draftcloseneeds a sortwins
S7 · Sent, never acknowledgedno datano datano data
S8 · 8 of 12 arrivedthroughput onlywinsnowhere to record it
S9 · Undo before sendingloses the trailwinsnot a stall
S10 · On a phone at the benchwinsworst mobile caseanswers a narrower question

The scenarios do not partition by concept. They partition by phase of work. A wins origination, C wins blockage, B wins the document. S3 and S7 are unwinnable by any architecture — they are missing capabilities (cross-object search, and an acknowledgement state) rather than IA choices.

Full reasoning and the 30-cell rubric are in 04-concepts.md.

All prototypes