Skip to content
Threadwise.
Home Comparison

Document 01 · Product Source of Truth

The why and the what,
in one place.

The frozen strategy and research behind Threadwise: the problem, the user, the insights, the point of view, the principles, the MVP scope, the flow, and the decisions — the spine every hi-fi screen must pass.

Visual conversion of the Master Product Document · Threadwise v0.1 · July 2026

01At a glance

A shopper reads the fiber-content label into Threadwise — by typing it or photographing it — and gets a fast, honest, shame-free verdict on whether the garment is worth buying, judged on its material. It works exactly where brand-rating apps go blind: the unbranded and thrifted item.

What

A fabric-label decision tool that returns a calm verdict at the moment of purchase.

Who

The sustainability-curious, non-expert shopper who already reads labels.

Problem

People want to choose better but feel powerless, distrust brand claims, and can’t read the one honest signal — the fiber label.

Decision moment

Garment in hand, label visible — a yes-or-no on this one item.

Differentiator

Item-level material judgment, sustainability-led, uniquely useful on unbranded and thrifted items.

Stance

Honest broker: states what the fabric is and gets out of the way. No guilt, no upsell.

02Context & vision

A portfolio case study, not a production product — built to demonstrate senior product thinking and real fluency with AI as a design material.

AI in the back

The expensive AI work runs once, offline, to build a frozen reference database. The shipped product makes no AI calls at runtime and the user never talks to an AI. The intelligence is absorbed at build time so the front stays instant, private, and calm.

Build-time versus runtime architecture for the fabric app Build time uses AI and a person to research and produce a frozen database, done once. That database is the handoff. Run time is the user entering or scanning fabric percentages and getting an instant lookup with no AI. Build time · you + AI · done once Research sources AI drafts data You verify it Frozen database scored once, no live AI Run time · the user · no AI Enter or scan % Look up + blend Show verdict Photo path uses on-device OCR — free, local, no cloud AI
AI in the back: build-time AI, checked by a person, produces a frozen database — the single handoff. At runtime the user gets an instant, on-device lookup with no AI calls.

Concept vs production

Settled physical facts about fibers are real and free, so they are used as real. A certified sustainability score is proprietary and expensive, so the composite is a transparent, self-authored model on top of those facts, openly labeled illustrative. The moment a task starts to feel like running a real company is the edge of the frame — and the response is to stop or simplify.

The ambition

Success is a better purchase, and eventually a shopper who has learned enough to read labels themselves and need the tool less. A good teacher works toward being unnecessary. The vision is bounded on purpose: Threadwise does one honest thing well, and does not try to solve sustainability, rate every brand, or manage a wardrobe.

03The problem

The real problem is a human decision problem, not a missing feature. The barrier is not a lack of facts — it is powerlessness and the fear of getting it wrong.

Powerlessness, not ignorance

Guilt and overwhelm — not a lack of knowledge — are what stop the behavior.

The label is hard to read

“80% viscose, 20% elastane” is the trusted signal shoppers can least interpret.

Claims can’t be trusted

Sustainability info is scattered, jargon-heavy, and greenwashed — and never present at the decision.

Loneliest when unbranded

The decision is fastest and hardest on the thrifted item brand tools structurally cannot judge.

04Target user

Built for one person, not everyone. Narrowing to this user is what keeps the product sharp.

User type

Sustainability-curious, non-expert. Reads food and skincare labels; wants the same for clothing.

Context

Fitting room or aisle, or online. Standing, seconds to spare, one hand free, sometimes offline. Increasingly secondhand.

Goals

Choose better quickly, feel confident, avoid the guilt of getting it wrong.

Frustrations

Labels they can’t decode, claims they can’t trust, help that is never there when it counts.

Behaviour

Fast and physical. Wants one glanceable answer, distrusts anything like a sales pitch or a lecture.

Not for

Supply-chain researchers, brand analysts, or people who want a wardrobe manager.

05User decision journey

The arc Threadwise serves: from doubt, through a calm verdict, to a confident decision the shopper can act on and move past.

Before · doubt

Powerless at the shelf

Garment in hand, label unreadable, claims untrusted. Wants to choose well but fears getting it wrong — especially on an unbranded or thrifted piece.

Feeling: overwhelmed

During · the read

One honest signal

Types or scans the fiber label. Threadwise returns a calm verdict, a score with an “illustrative” cue, and one plain reason — in one breath, no scroll.

Feeling: oriented

After · capability

A decision, then out

Buys or skips with confidence, having learned a little about the material. Nothing saved, nothing uploaded — the tool steps back.

Feeling: capable
Doubt → orientation → capability. The product’s job is to make the user feel effective, not to educate them or keep them engaged.

06Key insights

Four insights from the research, each written as a design directive. Every design choice should walk back up to one of these, or it is decoration.

Insight 1

The barrier is powerlessness, not ignorance

People fail to shop sustainably because they feel their choice doesn’t matter and fear getting it wrong — not because they lack facts.

Implication: the first job is a confident, shame-free verdict that makes the user feel effective, not an education.

Insight 2

They trust the label because brands lie to them

Greenwashing has trained shoppers to distrust brand claims, leaving the legally required content label as the one signal they believe. Trust is scarce.

Implication: read the label, and openly state the score is illustrative, not certified. Honesty is the moat.

Insight 3

The decision is fast, physical, and loneliest where every tool goes blind

It happens standing up, in seconds, most acute on the unbranded or thrifted garment brand tools can’t judge.

Implication: design for the shelf — speed, glanceability, offline — and own the moment no incumbent can follow.

Insight 4

A single score is both what they want and what can hurt them

People crave one glanceable answer, and that same simplicity can mislead, shame, and flatten nuance.

Implication: deliver the verdict, but calibrate severity honestly, show the facts beneath, never shame, gently correct myths.

07Product point of view

A practical, non-judgmental, honest broker for real shopping moments.

Practical, not idealistic

Helps decide about the item in hand. It does not demand a lifestyle or moralize.

Non-judgmental

Every verdict is factual and calm. “Better to skip,” never “bad.”

Tradeoff-based

Surfaces real tradeoffs in plain words — great for the planet, weak on durability.

Decision-support

States a verdict and steps back. It never tells the user what to buy instead.

For the moment of purchase

Fast, glanceable, offline, one-handed. The whole product lives in one moment.

Not trying to be perfect

Openly an illustrative model on real facts. That honesty is the source of trust.

08Design principles

Six principles resting on a low-tech, secure-attachment philosophy: teach through use, orient rather than trap, return the user to the world, measure capability not engagement.

#PrincipleMeaningHow it shows up
1Answer in one breathA verdict in seconds, or it has failed the shelf.The peek gives verdict, score, and reason with zero scroll.
2Anchor on the honest signal, and be honest yourselfRead the label; admit the score is illustrative.The “illustrative, meant to guide” cue sits next to the score and on About.
3Guide, never shameAn honest verdict in calm words; never leave the user judged.Calm Skip, “better to skip” not “bad,” a factual reason line. Non-shaming now rests entirely on tone.
4Honest about nuance, never preachySurface real tradeoffs briefly, only when they help.Blend notes, natural-vs-synthetic reality, durability and care, in plain words.
5Quiet by default, on the user’s deviceNo accounts, no cloud, no AI in the face, no mechanics.Runs on device, nothing uploaded, works offline, no engagement loops.
6Built for the messy real worldMisreads, blends, unknown fibers, missing labels are first-class.Confirm-the-read, partial and unknown states, multi-component handling.

The refusals — hard lines, rejected on sight

Notifications / push Streaks & badges Infinite scroll / feed Urgency & scarcity Accounts & logins Ads AI chat in your face

09MVP scope

Text-first, single-garment, sustainability-led, with a practical durability-and-care layer. Text exercises the entire data-to-verdict logic at the least technical risk; the photo path is additive realism on a working core.

❄ Frozen · MVP decision The MVP is text-first, single-garment, sustainability-led, with durability and care as a practical qualitative layer, multi-component (lined) garments handled in v1, and all the unhappy paths specified. This scope is frozen for hi-fi.
✓ In the MVP (v1 core)
  • Single-garment input — type or scan the label
  • Forgiving parser + fiber-name normalization + frozen fiber database
  • The verdict: scoring model, blend rules, verdict bands
  • Material explanation: what it is, end of life, microplastic flag
  • Durability & care notes (qualitative)
  • Multi-component handling (lined coats, blazers)
  • Unhappy paths: partial, unknown, low-confidence, offline, error
  • About: method, honesty cue, privacy · plus a clean closure moment
→ Not in the MVP
  • Compare 2–5 garments Designed & ready as Phase 2; seams reserved
  • Wardrobe-level insights Needs persistence + account
  • Brand-level analysis Would cede the wedge
  • Purchase history Needs accounts + storage
  • Certified lifecycle scoring Proprietary data, out of frame
  • Community / reviews Pure engagement machinery
  • EU Digital Product Passport ingestion Depends on a maturing data layer
In-scope items are built for v1; out-of-scope items are deferred deliberately, each with a reason. The MVP reserves the seams for comparison without shipping it.

10Main user flow

Single garment. The result is a card that rises over a dimmed Home and expands on tap — the resolved flow direction that serves the rapid re-check wedge.

Step 1Open HomeValue line, input with example, scan option
Step 2Enter labelType composition, or tap Scan
Step 3Result peekVerdict + score + one-line reason, no scroll
Step 4 · optionalExpandBreakdown, decomposition, flags, durability, “why”
Step 5Decide → out“Check another” or leave — nothing saved
Peek-to-expand is a tap, not a drag — it removes fragile gesture physics and degrades cleanly to a full page on desktop. Imperfect input routes to honest partial / low-confidence states before the verdict.
⤳ Conflict resolved The Develop phase chose a two-screen structure and held the bottom sheet in reserve. Later work judged rapid re-checking central to the thrift-and-aisle wedge, so the card-over-Home pattern is now canonical.

11Future comparison flow

Designed and wireframed, but not in the MVP. Useful when a shopper is choosing between several similar garments — most naturally on a thrift rack or a search page.

◷ Parked for v2 · Phase 2 Opt-in “Add to compare” drops a verdict into a session-only tray (max 5). “Compare” opens a full-page view: a pinned dimension rail with one column per garment, the best value marked per row, and a gentle “best of the set” line. The tray is invisible until opted into, so the single-item flow is untouched — which is why it stays out of v1.
See also The comparison ranking logic is visualized in Document 02 · Fabric Composition Score, and the compare flow is diagrammed in Document 03 · Hi-Fi Design Spec.

12Scoring, in brief

The verdict is a calm, illustrative signal — a large number with its verdict word and an “illustrative, meant to guide” cue, never a precise gauge. The full engine lives in its own document.

Good to buy · 70–100 Okay, with tradeoffs · 45–69 Better to skip · 0–44

End of life is weighted highest because it is the product’s clearest differentiator and the least understood by shoppers. Blends graduate the end-of-life penalty rather than hard-capping to the lowest fiber — so a 97% cotton, 3% elastane tee does not read as “skip.” Multi-component (lined) garments carry one headline verdict anchored on the shell, with the lining as an honest modifier. Durability and care are surfaced as qualitative notes, not folded into the score.

❄ Frozen · MVP decision Three calm bands, chosen over blunter labels: “Better to skip,” never “Avoid.” “Avoid” reads as a command and edges toward shaming, which the point of view forbids.
Full engine The MVP scoring engine — fiber base scores, composition rules, bands, and eight worked examples — is specified in Document 02 · Fabric Composition Score.

13Content & tone

Voice: calm, plain, warm. Short sentences, common words. A knowledgeable friend who shops with you — never a lecturer or a salesperson.

Guiding metaphor

Fabric is ingredients — for your skin and for the planet. Read the label as you would on food.

Working tagline

“Read the fabric, not the marketing.” Supporting: “fabric is ingredients, read the label.”

Terminology

“Breaks down,” not “biodegrades.” “Sheds tiny plastic fibers,” not “microplastic pollution.” “Better to skip,” never “bad.”

Honesty cues (central)

“Scores are illustrative, meant to guide, not certify.” “Everything stays on your device.”

14Information architecture

Deliberately flat. Three places, no hierarchy, no menus to browse.

PrimaryHomeValue line, input + example, scan, quiet About link
PrimaryResultVerdict card over Home — peek, then expand
PeerAboutMethod, honesty cue, privacy
Home → Result via submit; Result → Home via “Check another.” The bottom-of-Home zone is the reserved seam where the Phase 2 compare tray docks. Scan-and-confirm is a secondary screen on the photo path.

Progressive disclosure is the mechanism that lets one screen serve both a two-second glance and a curious drill-down: the peek shows verdict, score, and reason; everything else waits for the expand; the five score dimensions wait behind the “why this score” accordion.

15Decision log

The confirmed decisions that govern hi-fi. Use this to refer back rather than re-litigate.

AreaDecisionWhyStatus
Target userThe sustainability-curious, label-reading non-expert — not everyoneSharpness beats breadthConfirmed (persona untested)
Concept vs productionConcept case study, real fiber facts + illustrative compositeCertified data out of frame; honesty is itself the artifactConfirmed
AI architectureBuild-time AI to a frozen database, zero runtime AISpeed, trust, privacy, offline, zero per-use costConfirmed
MVP scopeText-first, single garment, sustainability-led + durability/careExercises full logic at least riskConfirmed
Single-garment flowCard over a dimmed Home, tap to expandGlance + context-keeping + focus; serves rapid re-checkConfirmed · test re-check
Future comparisonSession-only tray, 2–5, Phase 2Opt-in, additive, honors no-accountsConfirmed for Phase 2
Scoring modelWeighted composite, end of life highest, calm bandsEnd of life is the clearest differentiatorConfirmed · numbers need validation
Blend / trace-syntheticGraduate the end-of-life penalty, don’t hard-capA 97% cotton tee must not read as SkipConfirmed
Multi-componentOne verdict anchored on the shell, lining as modifierLined garments common in the wedgeConfirmed
SuggestionsRemove all purchase suggestions; state the verdict and stopKeeps it a pure honest brokerConfirmed
Design systemExpressive heritage “festival” system, as shipped, is canonicalDistinctive, real POV, disciplined by calm base / loud heroesConfirmed · final
Product nameKeep ThreadwiseGood, and used throughoutConfirmed

16Gaps & next steps before hi-fi

Direct, as a senior reviewer would put it. The design system is not a blocker — it is final and ready to build against.

⚑ Known gaps & risks
  • Personas & the four insights are hypotheses No primary user research yet
  • Scoring numbers are drafted, unverified A few values need a one-line rationale
  • Card-over-Home rests on the re-check assumption Untested; collapse to a full page if not valued
  • Central scoring risk: misleading precision Illustrative cue + qualitative layer mitigate
  • No real user reactions yet Even five sessions would ground hi-fi
✓ Must fix before hi-fi
  • Verify the scoring numbers (confirm the trace-synthetic fix)
  • Fully spec the confirm-the-read screen (permission-denied, low-light)
  • Name the result surface consistently as the card over Home
  • Lock the product name and retire the working-name note
↑ Back to top