# RESEARCH BRIEFS — six bounded evidence bundles (P1) Authored 2026-07-30 [F]. Executors: Flash (`[FL]`) for collection; judgment items listed at the bottom route to F/O/S only. ## Universal source contract (applies to every bundle) 1. Output file: `01_RESEARCH/evidence/B-.md`. One file per bundle, plus rows appended to `01_RESEARCH/evidence/SOURCE_REGISTER.md` (`id | url | title | access date | bundle | note`). 2. **Every factual claim**: source URL + access date + verbatim quote or extracted datum. No URL → the claim is written as `UNVERIFIED:` or omitted. 3. Nothing from model memory presented as evidence. If a fact is "known" but unsourced, it goes to the bundle's `## Open questions` section instead. 4. Conflicting sources: record both, flag `CONFLICT`, do not resolve (resolution is judgment work). 5. Polish-language sources are first-class; quote in Polish, add a one-line English gloss. 6. Per-bundle bounds below are hard: stop at the cap, list what was cut in `## Not covered`. A truncated honest bundle beats a padded one. 7. Read-only research. No sign-ups, no messaging anyone, no scraping behind logins, nothing sent to sprawdzonaopiekunka.pl. --- ## B1 — Poland: direct platforms & informal alternatives (childcare seed) **Question:** Through what channels do Polish families find childcare today, and who operates there? **In scope:** Polish babysitting/nanny platforms and agencies (e.g. niania.pl, Sitly PL, local agencies), classifieds (OLX etc.), Facebook-group/word-of-mouth channels as documented in public posts/articles, plus `sprawdzonaopiekunka.pl` itself (public surface only). Per operator: model type, pricing, geographic coverage, claimed supply size, verification claims (claims only — mechanics belong to B3). **Out of scope:** non-Polish platforms (B2), trust mechanics detail (B3), law (B4), rates/economics (B5). **Bounds:** 10–18 operators/channels; ≤3 pages per operator. Sources: operator sites, press, app-store listings. ## B2 — Mature-market comparators & business-model taxonomy **Question:** What distinct care-platform models exist abroad, and what does each monetize? **In scope:** 8–14 non-Polish platforms across childcare, elder, pet, home care (e.g. Care.com, Babysits, Yoopies, Rover, Honor/Papa-class managed models, care-benefit players). For each: classify as directory / marketplace / managed matching / subscription / staffing / benefit (charter taxonomy), revenue mechanism, category breadth (single vs multi-care), stated scale. **Out of scope:** Poland (B1), visual/UX capture (B6), trust mechanics detail (B3). **Bounds:** ≤4 sources per platform; prefer official pricing/help pages and filings/press over blogs. ## B3 — Trust & safety mechanics inventory **Question:** Concretely, what verification and safety machinery do care platforms operate, at which step, at whose cost? **In scope:** For ~8 platforms drawn from B1+B2 lists: identity checks, background/criminal-record checks (which registry, where lawful), references, certifications, insurance, payments-as-trust, safeguarding features, emergency escalation, ratings, dispute handling. Record mechanism, mandatory-vs-optional, who pays, evidence URL. **Out of scope:** whether these are lawful in Poland (B4), whether they work (judgment). **Bounds:** feature-table format; ≤5 sources per platform (help centers, T&Cs, trust pages). ## B4 — Poland: legal, regulatory & compliance sources (collection only) **Question:** Which primary legal texts and official guidance govern paid care work and platform operation in Poland? **In scope:** Collect and excerpt primary sources: KRK (criminal record registry) access rules for employers/parents; the "legalna niania" / umowa uaktywniająca scheme (ZUS); child/elder care professional regulation; RODO/GDPR constraints on vetting workers; employment-form rules (umowa zlecenie etc.); pet-care/animal law basics; platform-liability basics (DSA). Official portals (gov.pl, zus.gov.pl, EUR-Lex) strongly preferred. **Out of scope:** interpretation, risk assessment, "what we may do" — **all legal interpretation is judgment work** (see below). Flash copies the text and the citation; it never paraphrases law into advice. **Bounds:** 12–20 primary sources; each entry = source, article/section, verbatim excerpt, access date. ## B5 — Poland: demand, supply & economics **Question:** What do supply, demand, prices, and channels look like, in numbers? **In scope:** Published rates for nannies/babysitters/elder carers/pet sitters in PL (portals' own stats, salary surveys); GUS/Eurostat demographics (births, aging, pet ownership, female labor participation); childcare-place availability stats; documented acquisition channels and partner types (employers, nurseries, NGOs) as reported publicly. **Out of scope:** willingness-to-pay conclusions, market-size modelling (judgment work — Flash records inputs only). **Bounds:** 15–25 data points; every number with unit, year, and source; mark estimates `EST` with the source's own methodology note if given. ## B6 — Visual & interaction competitive capture **Question:** What do the key screens/flows of 6–8 comparators actually look like? **In scope:** For 6–8 platforms chosen from B1/B2: public landing, search/browse, carer profile, trust signals placement, onboarding entry, pricing page. Deliver: screenshot files under `01_RESEARCH/evidence/B6-shots//`, plus an index file listing each shot with URL + capture date + one factual caption (what is on screen, not whether it is good). **Out of scope:** aesthetic judgment, design-direction lessons (judgment work, feeds P4). Never reproduce protected assets in our own designs — capture is for analysis only. **Bounds:** ≤8 shots per platform. **Box constraint:** browser capture is heavyweight — run only on boxguard GREEN, one platform at a time; this bundle may lag the others. **Note [F]:** if run via webget/webget-stealth text extraction instead of a browser, capture DOM text + structure and mark visual claims `UNVERIFIED` until screenshots exist. --- ## Judgment work — NOT for Flash (route to Fable/Opus/Sonnet) - Legal interpretation and compliance posture from B4 sources (what vetting is lawful, employment risk, RODO limits). - Cross-bundle synthesis: competitive matrix, white-space analysis, trust-model comparison, child/elder/pet overlap assessment. - Willingness-to-pay and market-size reasoning from B5 inputs; resolving any `CONFLICT` rows. - Visual/interaction lessons from B6 and their translation into an original design direction. - Anything scoring, ranking, or recommending — including which alternatives to discard. These become P1-7 papers in `01_RESEARCH/` (signed), and feed the P2 council.