Growth, in focus.

Observed: Event and dress questions appear in first visitor turns across Aug 2025, Jan, Feb, Jun, Aug, and Sep 2026. Examples include a job interview (Aug 18, 2026), professional mixer (Jun 11, 2026), and a golf event fireside chat (Sep 8, 2026). Some first turns pair outfit uncertainty with anxiety, including a Feb 22, 2026 closet scenario.

Boundary: Similar prompts recur and some sessions are tests. We cannot estimate how often the broader audience has this problem or whether advice changes the final outfit.

Next proof: Recruit distinct visitors facing a real event; measure whether they choose an outfit within the session, how long it takes, and whether they wear it.

Observed: On Jun 11, 2026, a visitor described checkered trousers, sneakers, and the only fitted tee available; DOROTA® revised toward an eggshell shirt that the visitor said they would choose. On Sep 4, 2026, a visitor corrected an impractical bag alternative because they owned only one bag. On Aug 18, 2026, a visitor volunteered an existing blazer.

Boundary: This supports wardrobe awareness as a useful input. It does not validate photo upload, closet digitization, automated tagging, or measured rewear.

Next proof: Compare a quick manual item list with a photo based closet flow and no closet context. Measure setup completion, outfit choice, and use of owned items.

Observed: The Sep 8, 2026 golf event request narrowed toward sleeves, no dresses, roomier structure, and executive presence. A Jan 5, 2026 visitor requested an androgynous alternative to a conventional interview look. A Sep 2025 visitor corrected an assumed dress recommendation after specifying male identity. These show both demand for tailoring and the cost of guessing.

Boundary: Different configurations produced different responses. A correction within one chat is not proof of inclusive accuracy across identities, sizes, or skin tones.

Next proof: Test consent based personalization with diverse participants and actual wardrobes; score constraint adherence, respectful language, and whether the final look feels authentic to the person.

Observed: On Jun 11, 2026, a visitor settled on an eggshell shirt; on Aug 18, 2026, a visitor called the suggested interview combination “perfect”; on Sep 2, 2026, a visitor said DOROTA® had been helpful after assembling a game outfit. These are useful moments of immediate task progress.

Boundary: Conversational praise can be polite and is neither independent outcome measurement nor proof of repeat usage. A Sep 4, 2026 exchange also needed several corrections before reaching a usable outfit.

Next proof: Capture a chosen look, confidence before and after, optional outfit photo or follow-up, and a one week return measure. Compare against a simple static style guide.

Counterexample: A Jul 12, 2026 visitor asked for summer office wear, answered successive identity, feeling, context, and wardrobe questions, then said they had been talking too long and asked for the look. In a separate Jul 7, 2026 challenge, repeated calming prompts were described as making the exchange more stressful. Other first turns ask direct binary styling questions that invite an immediate recommendation.

Boundary: Some visitors spontaneously describe distress and appreciate support. The evidence challenges a universal required intake, not the value of empathy. Prompt versions and contexts differ, so a causal penalty is unmeasured.

Next proof: Randomize “quick look first” versus an emotion first intake, with an optional support branch. Measure time to first usable outfit, completed decisions, opt-in to deeper discussion, and satisfaction.

Counterexample: After a detailed office wear exchange on Jul 12, 2026, a visitor asked DOROTA® to show how they would look. DOROTA® said she could not generate a live image. A Jun 13, 2026 visitor separately requested an actual item after receiving general shopping suggestions.

Boundary: These are clear unmet requests, but one direct visual request cannot validate demand for a 3D avatar, AR try-on, or a physical smart mirror.

Next proof: Offer a simple 2D outfit board after a text recommendation, then compare it with text only. Test 3D and try-on only if 2D materially improves decision confidence or completion.

Signal: The literal nervous and anxiety word families each occur in 13 sessions across both years. Some first turns link those feelings to dressing, while later turns may follow an emotion prompt. A Feb 22, 2026 visitor appreciated calming advice and a Jan 12, 2026 visitor asked about first day nerves.

Missing: No before and after measure, control, or follow-up verifies reduced anxiety or greater confidence. The same conversation may contain acute distress unrelated to clothes.

Test: Keep emotional support optional, measure immediate decision confidence and a short follow-up, and compare it with direct styling alone. Assess escalation separately for serious distress.

Signal: Visitors describe pieces they own, and a Sep 8, 2026 conversation discusses closet clutter. Later in that transcript, people present planned app features during a demonstration; that narration is not evidence of adoption.

Missing: No photo ingestion, item accuracy, repeat closet use, rewear, or saved time is observed.

Test: Run a concierge prototype for real wardrobes. Measure setup abandonment, match rate to owned items, decision speed, and rewear against a lightweight manual list.

Signal: The Jul 12, 2026 “show me” request points to visual confirmation. A Sep 2025 3D question closely follows product language and is hard to distinguish from a scripted feature prompt.

Missing: Nobody uses a 3D look or AR try-on in the export; there is no comparison with flat outfit boards, mirror selfies, or real try-on.

Test: Compare 2D collage, own-photo mockup, and 3D only after verifying each rendering accurately reflects body and garments. Measure accuracy, confidence, completion, and production cost.

Signal: A few visitors worry about how other people will see their outfit. This establishes vulnerability around social judgment.

Missing: No visitor posts a look, receives peer feedback, opts into voting, or experiences moderation. The fear of judgment could also make public exposure unattractive.

Test: Offer private DOROTA® feedback, a trusted friend, and a moderated small group as distinct opt-in choices. Measure uptake, perceived safety, constructive feedback, and regret after sharing.

Signal: On Jun 13, 2026, a visitor moved from styling advice to “Where would I buy?” and asked for an actual item with fabric preferences. DOROTA® could only name retailers.

Missing: There are no verified items, links, clicks, purchases, or evidence of broad commerce intent. Someone who asks for a link may still buy elsewhere or not buy.

Test: Provide a handful of real, in-stock choices after an outfit is selected. Measure whether visitors ask for them, click, save, purchase, and return items.

Signal: People do engage in multi-turn conversation. In a Jul 30, 2026 demo someone reacts positively to talking rather than typing.

Missing: The session metadata does not reliably identify input modality. There is no equal text option, random assignment, or measured preference. Demonstration enthusiasm is not a mode comparison.

Test: Offer both modes with identical styling capability; track chosen mode, task completion, setting, accessibility, and preference after trying both.

Signal: Some visitors disclose style preferences and ask DOROTA® about the future app. The same session can contain multiple styling decisions.

Missing: No stable person identifier or reliable cross-session linkage, no persistent memory test, and no measured retention. Session counts cannot be treated as unique people.

Test: With explicit consent, compare a remembered preference card to a fresh session over several weeks. Measure repeated use, correct recall, perceived trust, and opt-out.

Signal: Planned features appear in founder or demo narration. That establishes the concept presented, not customers’ demand for it.

Missing: No hardware prototype use, price tradeoff, placement choice, or behavior that requires hardware over a phone.

Test: First observe unresolved tasks in the phone experience. Then test a functional mirror prototype with a small set of households against a phone and ordinary mirror.

Signal: DOROTA® mentions the waitlist in 24 of the 118 sessions with visitor speech. Some visitors ask about the future app or say a conversation helped.

Missing: No linked click, signup, activation, repeat use, or willingness to pay. The 110 waitlist members cannot be attributed to DOROTA® from this export.

Test: Instrument an explicit opt-in handoff after a completed styling task and link it, with consent, to verified signup, activation, four week retention, and a real pricing choice.