Where the work actually stands
Poland map closed; comparator slices collected; four bundles remain
Baby Nurt is an evidence-led exploration of a trust-first care wedge in Poland. Nothing here is a product. The record below states what has been collected, what each collection can carry, and what is still missing — in the same terms the underlying artifacts use.
Bundle ledger
-
B1 — Poland, direct platforms Complete · bounded
A formal-accessible-channel map of the Polish childcare market: 16 URLs attempted across 15 operators and channels, 12 pages retrieved across 11 of them, 4 unreachable and recorded as receipts. Covers consumer platforms, the OLX classifieds channel, and five recruitment agencies.
Complete as a supply-side operator map only. It is not an answer to how Polish families actually find care: informal channels — parent groups, word of mouth, nursery noticeboards, grandparent care — were in scope and zero sources were collected on them.
01_RESEARCH/evidence/B1-poland-direct-platforms.md · limits in 01_RESEARCH/P1-B1-QA.md -
B2-A — Care.com, Babysits Collected
Six captures, five content-bearing, one 404 receipt. Two degraded sources are recorded rather than corrected: the Care.com terms capture is truncated at 20,000 characters, cutting off the vetting and payment sections, and the requested Babysits “global” home resolved to a Malay-language Malaysian localisation.
01_RESEARCH/evidence/B2-A-global-marketplaces.md · raw: reports/evidence/B2-A-raw/ -
B2-B — Yoopies, Rover, UrbanSitter Collected · null yield
Nine captures produced one page title. Six responses were anti-bot refusals (HTTP 403), two returned an unrelated static-hosting 404 body under a comparator hostname, one returned an empty body. 0 of 12 requested taxonomy fields were established; all three comparators are recorded as unclassified.
The slice is kept because the failure is itself documented and bounded — not because it characterises these three operators. It does not.
01_RESEARCH/evidence/B2-B-global-marketplaces.md · raw: reports/evidence/B2-B-raw/ -
B2-C — Sittercity, Papa, Honor, Home Instead Collected
Twelve captures, five content-bearing, seven operator-template 404s. Yields a four-way model contrast — self-service marketplace, benefit-funded managed care, technology-and-network parent, franchised agency — and one structural correction: Honor and Home Instead state on their own sites that they are one group, so they must not be counted as two comparators.
No monetary amount of any kind appears in the twelve receipts. Model shape was captured; economics were not.
01_RESEARCH/evidence/B2-C-managed-and-specialists.md · raw: reports/evidence/B2-C-raw/ -
B2-S — comparator merge Technical hold
The final B2 synthesis — merging slices A, B and C into a single comparator taxonomy — is held after three equivalent foreground Opus attempts timed out before writing it. The completed slices and the 27-receipt corpus remain usable; the next merge must first use a deterministic compact extract, then a smaller local-only classification pass. Independent trust/safety research moves forward meanwhile.
planned target: 01_RESEARCH/evidence/B2-mature-market-comparators.md · corpus: reports/evidence/B2-S-raw/ -
B3 — trust and verification mechanics Not started
Every verification claim collected so far has been recorded as a claim and deferred here. What any operator actually checks, who pays for it, and when it runs is unestablished.
-
B4 — legal, regulatory, contractual Not started
Contract forms, employment-agency registration, and the lawfulness of criminal-record checks in Poland are all quoted from operator marketing so far and must be re-derived from primary sources before any load-bearing use.
-
B5 — rates and economics Not started
Blocks at least one downstream figure: one agency prices its entire fee as a percentage of the statutory gross minimum wage, and the corpus holds two conflicting values for that wage. See Market signals.
-
B6 — visual and interaction scan Initial pass complete
Three platforms inspected in a browser — composition, primary interaction, and where trust signals sit relative to search. Two further targets returned a verification page and a navigation timeout and yielded nothing.
Screenshots were inspected at runtime but are not stored as image files; the written record is the durable evidence. Design questions raised there are open questions, not conclusions.
01_RESEARCH/evidence/B6-initial-visual-scan.md -
Publication to baby.nurt.cloud Live · read-only
This evidence cockpit was published on 30 July 2026 after the target route and public HTTPS response were verified. It remains deliberately read-only: no booking, payment, identity verification, waiting list, analytics, or user-data collection exists here.
Deployment proof: public HTTPS 200 · containerbaby-nurtcloud
Tooling and provenance change
The Flash / OpenCode-Go collection lane is unavailable. The failure is a real insufficient-balance receipt, not a transient error, so it is not being retried automatically. Mechanical source collection that was planned for that lane has been carried by the verified Claude Code Opus lane instead.
This matters for reading everything above: the provenance of the corpus changed mid-programme, and the change is recorded rather than smoothed over. Collection volume in B2 was shaped by that constraint.
What the current corpus cannot support
- No market sizing, share, or ranking. Every scale figure held anywhere in the corpus is an operator's self-report on its own page. There is no third-party traffic, revenue, registration, or transaction data.
- No demand-side evidence. No parent, no carer, no survey, no review-mining, no forum reading. What families want, fear, or will pay is absent — partly by charter (no outreach), partly because review-mining has not been done.
- Nothing behind registration walls. One major operator requires an account before any carer profile is visible. Profile quality, response rates, and real liquidity are invisible.
- One moment, one vantage point. The Polish corpus is a single unauthenticated snapshot taken on 2026-07-30 from one host. No trend, no seasonality, no comparison point.
- The project's own subject site is uncharacterised. sprawdzonaopiekunka.pl failed DNS resolution from the collection host. That is a receipt about one fetch, not evidence that the site is defunct — and it sits directly under the sponsor's intent.
Grounded readings · Poland corpus
Four patterns the captures support, and what each one stops short of
These are readings of captured text, not market conclusions. Where an operator states something about itself, it is marked as a claim and stays a claim. Nothing below ranks, scores, or recommends anything.
-
Multi-care is already being attempted in Poland, from two opposite directions
Direct capture The group that operates the largest Polish childcare listings site also runs pomoc domowa, opieka seniora and opieka zwierząt as separate sibling sites, linked from its own footer. That is the separate-site precedent: one owner, four surfaces, four audiences.
Direct capture The opposite precedent also exists. One Polish app covers childcare, senior support, home help, perinatal support and disability support inside a single product; and one Warsaw agency sells nanny, housekeeper and pet care as one household hire, with the housekeeper role explicitly including the family's animals.
What this does not show. Whether either structure works. There is no traffic, revenue, retention, or cross-category usage data anywhere in the corpus. The sibling sites were identified inside a captured footer and never fetched; whether they share supply, identity, or trust machinery is unknown and directly tests the expansion question.
01_RESEARCH/evidence/B1-poland-direct-platforms.md §5 · 01_RESEARCH/evidence/B6-initial-visual-scan.md §1 -
Marketplace and managed agency are different products, not different polish
Direct capture Self-service platforms in the Polish corpus monetise the contact layer: one subscription marketplace gates outbound messaging, charging parents 99,99 zł for one month against 19,99 zł on the carer side — a five-to-one split that subsidises supply. Agencies monetise placement: one charges a single commission set at 60% of the current gross minimum wage, refundable if no engagement follows.
The axis underneath is who performs the match. On the platforms the family screens, chooses, and carries the risk. In the agencies an intermediary interviews, checks references, and presents candidates — and three of the five agencies have independently converged on a three-month guarantee with no standards body in sight.
Comparator note The same split appears in the mature-market slice, where the matching party divides three ways across the four operators captured — family, platform, or local office. That comparison is collected but not yet merged, so it is a pointer, not a finding.
What this does not show. Which model earns more, retains better, or transfers to Poland. Agency commission size is undisclosed for two of the three agencies that confirm one exists, and the one quantified fee cannot be converted into złoty yet — the corpus holds two conflicting values for the statutory minimum wage it is indexed to (4 300 zł in one operator's editorial copy, 4 806 zł recurring as a salary floor in dated listings on the same page). That conflict is recorded and unresolved.
01_RESEARCH/evidence/B1-poland-direct-platforms.md §5, §6 C-2 · 01_RESEARCH/evidence/B2-C-managed-and-specialists.md §7 F1 -
“Sprawdzona” is doing unearned work across the whole category
Operator claim · unverified Three consumer platforms use the word sprawdzona — vetted, verified — on their landing surfaces. Only the agencies describe an actual check, and the checks they describe are reference and document checks.
Direct capture The clearest inconsistency is internal to one operator, which advertises thousands of sprawdzonych profili on its homepage while its own FAQ states:
“Nie przeprowadzamy weryfikacji przeszłości ani nie sprawdzamy nowych członków, ponieważ nie jesteśmy agencją pracy ani agencją opieki nad dziećmi.” We do not run background checks or vet new members, because we are not an employment agency or a childcare agency. Verification, it goes on to say, rests with members.
A second platform lists mandatory identity verification and a criminal-record certificate among its safety items, without stating who obtains the certificate, when, at whose cost, or whether it is required at all. Two incompatible public postures serve the same market from the same category position.
What this does not show. That anyone is lying, or that any check is absent. No operator in the corpus claims a completed criminal-record check as a platform action, and whether more is lawful in Poland is an open legal question, not a finding. Verification mechanics are deferred to an unstarted bundle; until it runs, none of these claims may be treated as established.
01_RESEARCH/evidence/B1-poland-direct-platforms.md §5, §6 C-1, §7 -
Money mostly moves off-platform, and nobody visible is insuring the transaction
Direct capture Of eleven Polish operators and channels captured, exactly one describes payment and receipt inside its own product. The rest monetise at the contact layer or charge a placement commission, and no captured Polish operator describes escrow, insurance, or a dispute process.
Where money is used as a trust instrument, it is abroad: one global comparator documents escrow with release held until 24 hours after a booking ends, plus a no-show refund; another explicitly disclaims all liability for payments made through its own facilitation. Both are family-paid subscription marketplaces and they take opposite positions.
What this does not show. That escrow is desirable, lawful, or affordable here. Fee levels, take rates, and unit economics are absent from every slice collected so far.
01_RESEARCH/evidence/B1-poland-direct-platforms.md §5 · 01_RESEARCH/evidence/B2-A-global-marketplaces.md §3.5
No niche has been selected
The wedge — childcare, elder care, pet care, or a combination — is deliberately open, and stays open until the strategy phase. Nothing on this page should be read as a preference between them. Demand concentration on one voivodeship, the carer-side subsidy pattern, and the multi-care precedents are inputs to that decision, not the decision.
Candidate hypothesis · not a decision
A first flow to argue about, not a specification to build
What follows is a single candidate direction, written down so it can be attacked. It has not been selected, and the process that would select it has not run.
Pending the three-round council
The programme requires a three-round council with distinct Fable, Opus and Sonnet seats, each round preserving its inputs, seat views, disagreements, synthesis, and the changes it hands to the next round — with false consensus explicitly forbidden. Zero rounds have run. Follow-up research driven by council gaps has therefore also not run, and no beachhead has been chosen or ruled out.
Until then this direction has the status of a hypothesis with known evidence gaps beneath it: trust mechanics, Polish legal ground, and unit economics are all unstarted bundles.
Proposed first flow
Four steps, in this order. The ordering is the argument: the family's need is stated before any supply is shown, and a human check stands between a request and a contact.
-
Family need
A family describes the shape of the care it needs — who it is for, when, where, and what kind — before seeing any carer. The captured incumbents invert this: the first decision on the largest Polish surface is a role selector, after which profiles sit behind a free-registration wall.
Not built · no input is collected on this page
-
Care profile / evidence
A carer record that shows what was checked, by whom, and when, rather than the word sprawdzona with nothing behind it. Each line of evidence would carry its own provenance, and an unchecked field would read as unchecked rather than as absent.
Directly dependent on the unstarted trust and legal bundles: what may lawfully be verified in Poland, by whom, and at whose cost is not established.
Not built · no verification is performed
-
Matching request
A routed request rather than a browsable directory — the family asks, and a match is proposed. Whether the family or the operator should perform the match is genuinely open: the comparators split three ways on exactly this axis, and it is the choice that determines the trust apparatus, the unit economics, and the failure modes.
Not built · nothing is submitted or routed
-
Human trust review
A named human step before any contact is released, with an explicit account of what that person looked at. This is where the wedge would earn the word the whole Polish category currently uses for free.
It is also the most expensive step in the flow and the least evidenced: no cost, capacity, or willingness-to-pay figure exists anywhere in the corpus.
Not built · no review capacity exists
What this direction deliberately excludes for now
- No niche commitment. The flow is written to be legible for childcare first, but nothing here selects childcare over elder or pet care.
- No booking, payment, or escrow. Money as a trust instrument appears once in the Polish corpus and is unpriced everywhere; it is not in a first flow.
- No ratings-first design. Ratings and testimonials are the cheapest trust surface to build and the least supported by anything collected.
- No scale claims. The category's headline numbers are self-reported and, in at least one case, internally inconsistent. A first surface that cannot show honest local liquidity should show none.
What would send this back
- Evidence that Polish families overwhelmingly find care through informal channels that a request-and-review flow cannot reach — currently the largest hole in the corpus.
- A legal finding that the checks implied by step two cannot lawfully be performed, or cannot be performed by a platform.
- Economics showing the human review step cannot be paid for at any price families will accept — the agencies charge for exactly this work, and two of three will not say how much.
- A council round that rejects the ordering, particularly the decision to route requests rather than open a directory.
Provenance
Where every statement on this page comes from
Each artifact below is a file in the project repository. Paths are shown as written references rather than links: this page fetches nothing, and nothing on it is loaded from an external source.
Source-label rule
No platform feature, scale figure, price, or safety claim on this site has been accepted as fact without its source label. Operator marketing is recorded as an operator claim and stays one. A failed fetch is recorded as a receipt about that fetch, never as evidence about the operator behind it. Where the corpus is silent, this site says so instead of filling the gap.
Artifact register
-
Charter and current state
Sponsor intent, north-star question, research objectives, constraints, and the definition of done; plus the single source of where the programme stands.
PROJECT_CHARTER.md · 00_ADMIN/NOW.md -
Poland operator map
Eleven operators and channels with content, four unreachable targets kept as receipts, a conflict register, an explicit unverified-claims list, and a section stating what the corpus cannot support.
01_RESEARCH/evidence/B1-poland-direct-platforms.md · reports/evidence/P1-B1-B2-raw/ · 01_RESEARCH/P1-B1-QA.md -
Mature-market comparator slices
Three separate slices, each with its own capture inventory and limits. The merge that would turn them into one taxonomy is pending.
01_RESEARCH/evidence/B2-A-global-marketplaces.md · B2-B-global-marketplaces.md · B2-C-managed-and-specialists.md reports/evidence/B2-A-raw/ · B2-B-raw/ · B2-C-raw/ · B2-S-raw/ -
Visual and interaction scan
Public-page browser navigation with DOM and landmark inspection. No accounts, no forms, no search submissions.
01_RESEARCH/evidence/B6-initial-visual-scan.md -
Source register and failure record
One row per capture across all slices, plus preserved records of the runs that failed — a write refusal, several timeouts — kept rather than deleted.
01_RESEARCH/evidence/SOURCE_REGISTER.md · reports/evidence/ -
Deployment gates
The conditions under which anything may be published, including the TLS gate currently blocking publication.
05_PROTOTYPE/DEPLOYMENT_GATES.md
Platforms in the initial visual scan
Captured and described, 2026-07-30:
- Niania.pl — Poland, childcare. Photo hero, a two-option role selector as the first decision, sparse top navigation, and dense city and job-city link lists beneath. Sibling services for home help, senior care and pet care are linked from the footer as separate sites.
- Babysits Poland — Poland, childcare marketplace. Search-first hero with a city field and category tabs; a carer carousel appears before the safety section; a three-step section narrates search → contact → book, pay, receipt.
- Care.com — multi-category comparator. Search-first hero, then six equal category cards for child, senior, adult, housekeeping, pet care and tutoring, and a caregiver carousel drawing from more than one category.
Attempted and not captured — recorded as non-results, with no product or visual claim drawn from either:
- Rover — a verification page was served instead of the site.
- Papa — browser navigation timed out.
Everything quoted from the three captured platforms is public landing-page copy. Their scale numbers, ratings, and safety lists appear in the artifacts as platform claims awaiting verification — including the two largest, an eight-million-family figure with unstated geography and a forty-five-million figure with no date or activity definition.
Method and its limits
How sources were collected
Unauthenticated public-page fetches returning tag-stripped text, plus one browser pass for the visual scan. Content behind JavaScript, accordions, or registration walls is absent by construction — in at least one case a captured FAQ's questions are present and its answers never rendered.
What the capture format does not record
The stored receipts hold status, final URL, title, and text — and nothing else. The requested URL, the fetch client, and the timestamp are not in the files, so redirect behaviour cannot be audited from the corpus alone and access dates rest on the capture-run convention rather than on captured data. Several 404s in the comparator slices were later shown to be wrong-path misses, not missing content: the same operators carried that content elsewhere on their own sites.
Bias in what survives
Almost everything collected is operator-controlled first-party marketing copy. No press, filings, app-store listings, or third-party data are in the corpus. It is therefore systematically biased toward what operators wish to say about themselves — which is exactly why the labels on this page exist.
Read-only compliance
No sign-ups, no logins, no messages, no authenticated scraping, no submitted forms. Nothing was sent to the project's subject site. This page continues that posture: it has no forms, no analytics, no external requests, and no dependencies.