Skip to content

S5 · Wardrobe construction

Part II teaches the styling knowledge the prototype reasons over; this chapter explains the structural decisions Wearly makes when it assembles an outfit from a candidate pool.

The concept

After Step 4 has filtered the closet to a candidate pool that matches the occasion and the season, Step 5 has to build one outfit from it. Three structural rules govern construction:

1. Required pieces — what the outfit must contain

Every occasion declares a minimum type-set (see §S3 · Occasion and context). The agent's job is to fill each required slot with one item from the candidate pool. If a slot can't be filled, that's a wardrobe gap handed to Step 6 (see §S7 · Honest gaps).

2. Dress vs. separates — the branching choice

The most consequential branch in outfit construction is dress vs. top + bottom. For formal occasions (and sometimes dinner), Wearly prefers a dress when one exists in the candidate pool — because the styling literature treats dresses as a single-decision solution to formal events, and because rejecting a dress in favor of two separates inflates the chance of mismatch (color, formality, fit).

The branch is rule-based and inspectable — every dress-vs-separates decision cites a named rule in the reasoning trail. If no dress survives Step 4's filter, the agent falls through to top + bottom; if a dress exists but the user has previously rejected dresses for this occasion, the agent also falls through.

3. The accessory cap of two

Wearly's rule pack caps accessories at two per outfit. The cap is a styling convention — "edit the look" — and a UX one: more than two accessories crowds the reasoning trail with notes the user can't act on. The cap can be overridden in a future personalization pass; today it's a hard rule.

The outfit-as-typed-set view

The framing that makes the rules above coherent is outfit as a typed set:

Outfit = {top, bottom, shoes, accessory*}     # separates branch
       | {dress, shoes, accessory*}           # dress branch
       | {top, bottom, shoes, outerwear, accessory*}   # cold-weather variant

Every item carries a type, and the outfit is a set of items whose types satisfy the occasion's REQUIRED_PIECES. This is the same type-aware framing the academic literature uses for outfit compatibility — see References [3] (Vasileva et al. 2018) — only Wearly does it deterministically rather than by learning type embeddings.

Why deterministic branching over scoring

A scored outfit construction would compare every dress-vs-separates candidate by a single number and pick the winner. Wearly does not do this:

  1. Scoring hides the why. A user who asks "why this dress, not that one?" wants to read the rule, not see a difference of 0.04 in a learned compatibility score.
  2. The branching points are the styling decisions. Dress vs. separates is the question; reducing it to a continuous score throws away the explanation surface that makes the agent useful.
  3. The reasoning trail can name the branch. "Picked a dress because formal occasions prefer dresses when one is available" is a sentence; a learned score is a number.

The trade-off, as in §S1 and §S4, is that the agent cannot reason about cases the branches don't cover. A separates outfit that outperforms a dress for a specific formal context is invisible to this version of the agent. That's a roadmap item.

Where this knowledge comes from

  • Type-aware outfit construction. Vasileva et al. (2018) — type-aware embeddings show that conditioning on garment type improves compatibility prediction. Wearly's REQUIRED_PIECES map encodes the same axis without a learned model. ⟶ References [3].
  • Outfit as a set with constraints. Han et al. (2017) — outfit-as-sequence under a learnable distribution. Wearly's branching is the rule-based alternative the paper positions itself against. ⟶ References [2].
  • Recommender-systems framing. Ricci, Rokach & Shapira (2022) — content-based recommendation is one of the named families; Wearly's rule layer is one of its purest forms. ⟶ References [1].

How Wearly applies this

Where What happens
Rule pack occasion-rules.md — R3 (dress-vs-separates), R4 (accessory cap); wardrobe-filtering-rules.md — R3 (occasion-season filter)
Runtime code styling_agent._build_outfit(), wardrobe_tool.filter_items_by_occasion()
Agent step Step 5 (build outfit)
Where it surfaces The outfit grid + the "picked a dress because formal occasions prefer dresses" reasoning line
Part I chapter Chapter 05 · Wardrobe intelligence and Chapter 03 · Agent workflow (Step 5)
MicroSim (none yet — candidate for a future sim that walks the dress / separates / cold-weather branches)

Key terms

  • Required pieces — the minimum type-set for an occasion. See Glossary.
  • Outfit construction — Step 5 of the workflow. See Glossary.
  • Accessory cap — Wearly's rule that an outfit carries at most two accessories.

Self-check

  1. (Remember) What is Wearly's accessory cap per outfit?
  2. (Understand) Why is dress-vs-separates a branch rather than a score?
  3. (Apply) Given a 55°F dinner event for a user with no dress in the closet, list the type-set the agent will build, in order.