S6 · Rotation and freshness

Part II teaches the styling knowledge the prototype reasons over; this chapter explains how Wearly keeps a wardrobe in rotation without ever exiling a well-loved item.

The concept

A working wardrobe is one whose pieces are worn. Two failure modes break that:

• Stagnation. The same five items get worn every week; everything else sits unused. The closet behaves like a small wardrobe even though the inventory is large.
• Exile. A favorite shirt gets recommended so many times that an optimization metric — "vary the recommendations" — pushes it permanently off the candidate list. The wearer loses access to a piece they actually love.

Wearly's rotation rule is built to address the first without causing the second.

The freshness formula

freshness = 1.0
          − 0.25 × min(worn_count, 4) / 4    # frequency penalty
          − 0.25 if days_since_last_worn < 3 # recency penalty
freshness = clamp(freshness, 0.40, 1.00)

Two signals, both capped:

• Frequency penalty scales linearly from 0 wears to 4 wears and then stops. A piece worn 100 times receives the same frequency penalty as a piece worn 4 times.
• Recency penalty is binary: it applies inside a 3-day window and vanishes the moment the window passes.

And the floor is the rule that prevents exile:

• Freshness ≥ 0.40, always. Even at maximum penalty, the score never drops below 0.40. A much-loved piece is never pushed out of rotation.

The role in selection

Freshness is a tie-breaker, never a filter. At Step 5, after the candidate pool is set, items within each type are sorted by descending freshness. Higher freshness = picked sooner when the two candidates are otherwise equivalent. Its contribution to the selection score is 0.4 × freshness, which is small enough never to override a formality mismatch but large enough to break ties between two equally-eligible items.

A separate reasoning line surfaces when an item has been resting: "you haven't worn this in 42 days — bringing it back today" fires when worn_count ≥ 1 and days_since_last ≥ 30. That's an explanation surface, not a separate scoring axis.

Why a tie-breaker, not a filter

A filter applied earlier in the pipeline would exclude items below some freshness threshold from Step 4's candidate pool. The agent would never see them; the user would never get to wear them.

Wearly's choice is the opposite. Every eligible item stays in the pool. Freshness only matters when the rest of the rules return a tie — same occasion, same formality, same color tier. That preserves the 0.40 floor as a real guarantee rather than a number that happens to not crash the math.

Where this knowledge comes from

• Explicit vs. implicit feedback framing. Hu, Koren & Volinsky (2008) is the canonical text on the distinction between user signals that are given (a wear-confirmation, a reject) and signals that are observed (clicks, dwell time). Wearly uses only explicit feedback — worn_count increments when the user taps "wear this today," never when they merely view it. ⟶ References [11].
• Wardrobe-utilization framing. Sustainable-fashion literature (often appearing under terms like Project 333 or cost-per-wear) treats rotation as a wardrobe-design problem, not a recommender-system problem. Wearly's frequency cap at four wears is a design-informed approximation of that framing; no specific peer-reviewed source has been verified yet — see docs/evidence-and-references § 3.7.

What this rule deliberately does not model

• Wear context. A blouse worn five times to work is treated the same as a blouse worn five times to dinner, even though their rotation patterns may be different.
• Laundry state. Wearly does not know if a piece is currently in the laundry; the rotation rule assumes availability.
• Seasonal hibernation. A winter coat worn six times last December has a high worn_count but is effectively absent from the May candidate pool because Step 4's seasonal filter excludes it. The frequency penalty doesn't reset between seasons.

These are deliberate scope choices. The rotation rule is focused: one freshness score, one floor, one tie-breaker — every decision visible in the reasoning trail.

How Wearly applies this

Where	What happens
Rule pack	wear-history-rules.md — R2 (freshness as tie-breaker with 0.4 floor), R4 (rest-period reasoning note)
Runtime code	history_tool.get_freshness(), history_tool.days_since_last_worn(), styling_agent._score_main_piece()
Agent step	Step 5 (build outfit), tie-breaker only
Where it surfaces	The freshness pill on each card; the "haven't worn in 42 days" reasoning line
Part I chapter	Chapter 05 · Wardrobe intelligence
MicroSim	Wear-history freshness — the floor is visible as the slider refuses to drop below 0.40

---

Key terms

• Wear history — per-item worn_count and last_worn_date, written only on explicit confirmation. See Glossary.
• Freshness score — a 0.40–1.00 value combining frequency + recency. See Glossary.

Self-check

1. (Remember) What is the floor on the freshness score, and at which worn_count value does the frequency penalty stop growing?
2. (Understand) Why is freshness a tie-breaker rather than a filter — what would change if it were applied earlier in the pipeline?
3. (Apply) A user wore their navy blazer once, five days ago. Compute the freshness by hand, naming each penalty that does and does not apply.