Skip to content

S8 · Body-positive framing as a discipline

Part II teaches the styling knowledge the prototype reasons over; this chapter explains the language contract that runs across every other rule and why the discipline is enforced in code, not only in documentation.

The concept

Most popular styling literature is written in corrective vocabulary — flaw, fix, hide, minimize, slimming, problem area, flatter, shameful. The words frame styling as a project of changing the wearer to fit an external ideal. Wearly explicitly refuses that framing.

The replacement is body-positive framing as a discipline. Two sentences capture the rule:

  1. Wearly honors what the wearer has shared about their preferences. It never describes a feature of the wearer as something to correct.
  2. Wearly's vocabulary is calibrated — "tends to work well," "a clean line," "a balanced look" — never deterministic judgments like "is most flattering" or "hides this."

The discipline is enforced at two levels.

Level 1 — runtime forbidden tokens

The function fit_tool.check_reasoning_for_forbidden_language() inspects every reasoning line the agent emits and refuses any line that contains a forbidden token. The contract is organized around root concepts — the words and phrases the styling literature uses to frame body features as flaws — with spelling and dialect variants of each one also banned so a contributor can't slip past the check with British spelling or a synonym.

The root concepts (with their currently-enforced variants in parentheses):

flaw           (flaw, flaws)
fix            (fix)
hide           (hide)
minimize       (minimize, minimise)
correct        (correct, corrective)
problem area   (problem area, problem-area, problem_area, problem zone)
trouble area   (trouble area, trouble-area)
slimming       (slimming, slimmer)
shameful       (shameful)

That's 17 actual strings the runtime scans for, across the nine families above. A reasoning line that trips the check fails CI; a developer cannot ship copy that violates the contract without removing it from fit_tool.FORBIDDEN_TOKENS deliberately. The list grows when a reviewer surfaces a new corrective phrase and the team agrees.

Level 2 — static documentation check

tests/test_documentation_integrity.py walks every Markdown file in the book, the Skill pack, and the references doc and asserts none of the forbidden tokens appears. The discipline is not just for the runtime output; the documentation about the agent holds itself to the same standard.

If a contributor writes "hides the waist" in a chapter, the test suite fails. Body-positive framing is not a stylistic preference; it is a structural property of the project.

Why this is a methodology, not a style choice

A style choice would be a recommendation in a contributor guide. Wearly's body-positive framing is structural for three reasons:

  1. The audience is mostly women. The corrective vocabulary in styling literature does identifiable harm; the SEIS 666 project's evidence-informed framing requires that the project not perpetuate it.
  2. The contract is testable. A vocabulary rule that can fail CI is one a future contributor cannot quietly drop. The FORBIDDEN_TOKENS set is a contract surface.
  3. The framing affects the rule design, not just the copy. Because the agent can't say "hides the waist," it has to find a different framing — "this silhouette honors the wearer's preferred waist definition." That different framing requires a different rule shape (preference-aware, not correction-aware), which ultimately is a different agent.

Edge cases and their resolutions

  • "Balance" is allowed; "correct" is not. Balance is a styling vocabulary about composition; correction is about the wearer.
  • "Highlight" is allowed and encouraged. Highlighting a feature the wearer has chosen to share is body-positive; the choice came from the user.
  • Numbers (sizes, measurements) are never used as judgment. A measurement in the fit profile is a coordinate, not a rating; the rule pack documents this explicitly.
  • "Recommendation" carries no judgment by itself; the agent recommends outfits. The rejected verbs are the ones that describe the wearer rather than the garment.

Where this knowledge comes from

  • Hokka (2024) documents body diversity as a known challenge for inclusive design and frames small-set categorization as a limitation rather than a truth. Wearly's framing — every body-shape field optional, body-shape labels as user-declared preferences — is the direct expression of that framing. ⟶ References [7].
  • Miller (2019) on explanations in AI argues that explanations should be contrastive, selective, and socially appropriate. Body-positive framing is the socially appropriate axis applied to styling reasoning lines. ⟶ References [8].
  • Amershi et al. (2019) on Human-AI Interaction guidelines includes "Match relevant social norms" and "Mitigate social biases." Wearly's forbidden-token contract is one operational expression of both. ⟶ References [9].

How Wearly applies this

Where What happens
Rule pack fit-silhouette-rules.md — R1 (body-positive contract)
Runtime code fit_tool.FORBIDDEN_TOKENS, fit_tool.check_reasoning_for_forbidden_language()
Static check tests/test_documentation_integrity.py
Where it surfaces Implicitly in every reasoning line; explicitly in Chapter 06 and Chapter 12
Part I chapter Chapter 06 · Fit-profile logic
MicroSim (none — this rule is cross-cutting, not a single-rule sim)

Key terms

  • Body-positive language contract — the runtime + documentation rule that forbids corrective vocabulary. See Glossary.
  • Honesty contract — the four rules that govern every claim Wearly makes; body-positive language is one of them. See Glossary.
  • Forbidden tokens — the 17 runtime strings (across nine root concept families: flaw, fix, hide, minimize, correct, problem area, trouble area, slimming, shameful) that fail CI when they appear in a reasoning line.

Self-check

  1. (Remember) List four forbidden tokens from the body-positive language contract.
  2. (Understand) Why is body-positive framing called a discipline rather than a style choice?
  3. (Apply) You're writing a new reasoning line. You want to say "this dress hides the waist." Rewrite it so it passes the contract and means something coherent.