How trust scores actually work
No black boxes. No “trust us”. Here is exactly what we check, how we score it, and why a cross-agency platform can be honest in a way single-agency sites structurally cannot.
Read this first. The pillars below describe the full verification model. In the current public build some checks run on demo data only — the claims register below states exactly which are live, which are simulated, and which are planned. We would rather show you the blueprint with its status than blur the line.
1. Title & Ardhisasa cross-check
The title deed is checked against official land registry records — ownership, encumbrances, caveats, charges and whether the parcel touches planned public infrastructure. Listings show Verified / Pending / Flagged status, with the date of the last official search.
2. Photo authenticity scan
Every listing photo is fingerprinted and reverse-matched across all agencies on the platform. Recycled photos from other buildings, stock images passed off as real units, and old listings re-used under new names are all caught automatically.
3. Duplicate-listing detection
The same unit listed five times at five prices is one of the oldest scams in Kenyan real estate. Our cross-agency view detects duplicate and near-duplicate listings instantly — something no single-agency platform can ever do.
4. Pricing anomaly detection
Statistical models compare each asking price against the market band for its area, type and size. A price 30%+ below market isn’t a bargain — it’s bait. Urgency-scam patterns are flagged before they reach you.
5. Agent reputation scoring
Phone-number history, complaint patterns, response behaviour and prior listing outcomes build a reputation profile for every agent and agency. Repeat bad actors are removed from the verified network.
Score bands — what the badge means
90–100
Highly Verified
All checks passed. Clean title, unique photos, in-band pricing, reputable agency.
75–89
Verified
Passed all critical checks; one minor note (e.g. off-plan renders or similar unit elsewhere).
60–74
Under Review
Pending searches or minor anomalies. Proceed with standard due diligence.
Below 60
Flagged
Failed checks detected. We show it — clearly labelled — so you know what to avoid.
The claims register
What is real, and what is a demo
Most property platforms ask you to take their word for it. We publish the register below instead: every capability this product claims, its honest status, what actually backs it today, and what has to happen before it can be called live. If a claim is not here, we don’t make it.
| Claim | Status | What backs it today |
|---|---|---|
Title deeds are cross-checked against official land-registry records. Listing pages · Trust Center · reviewed 2026-08-31 | Simulated | Checks run on seeded demo data with recorded outcomes and dates; no live connection to Ardhisasa or Ministry of Lands systems exists in this build. Path to live: Integrate an Ardhisasa-access partner or advocate network, record search references, and date-stamp every result. |
Listing photos are fingerprinted and reverse-matched across agencies. Listing pages · Trust Center · reviewed 2026-08-31 | Simulated | Duplicate and reuse outcomes are pre-computed on the demo inventory; no external reverse-image provider is called at runtime. Path to live: Perceptual-hash pipeline over real multi-agency photo sets with a dispute workflow. |
Duplicate and near-duplicate listings are detected automatically. Auto-Pilot pipeline (scripts/auto-listings/dedupe.mjs) · reviewed 2026-08-31 | Live | Fuzzy-match deduplication runs as part of the ingest pipeline and in-app anomaly checks; covered by unit tests. |
Asking prices are screened against market bands for anomalies. Auto-Pilot pipeline · anomaly detection (adminStore) · reviewed 2026-08-31 | Live | Deterministic price-band screening runs on every ingested listing and inside the admin anomaly sweep; covered by unit tests. |
Agents and agencies carry a reputation score. Listing pages · Trust Center · reviewed 2026-08-31 | Simulated | Scores are demo values on seeded agencies; no complaint database, response-time tracking or operational history feeds them yet. Path to live: Response-time SLAs, complaint adjudication and renewal/suspension policy with real telemetry. |
Every listing carries a 0–100 trust score with five weighted pillars. Listing pages · Trust Center · reviewed 2026-08-31 | Live | Scores and their five-pillar weighting are computed and displayed deterministically; factor definitions are published on the Trust Center. |
Verification evidence carries a check date, scope and validity window. Listing pages (evidence panel) · reviewed 2026-08-31 | Live | Evidence panels derive scope, check date and a 90-day validity window per check, and expose a report / human-review path. |
Natural-language property search across inventory. Ask Keja AI · Properties · reviewed 2026-08-31 | Live | Deterministic parser + matcher over the demo inventory; covered by unit tests. |
Rental ROI, mortgage and affordability calculators with labelled assumptions. Investment Calculator · Ask Keja AI · reviewed 2026-08-31 | Live | Pure functions with FACT / ESTIMATE / ASSUMPTION labels; covered by finance unit tests. |
Auto-Pilot ingests partner feeds with quality gates and review queues. scripts/auto-listings/* · Admin console · reviewed 2026-08-31 | Live | Feed adapters, enrichment, dedupe, quality scoring and a capped trust score run end-to-end in CI on sample feeds. |
Keja sits above multiple agencies as a cross-agency trust layer. Trust Center · Home · reviewed 2026-08-31 | Simulated | Multiple demo agencies exist in seeded data; no live agency integrations or partner feeds are connected in production. Path to live: Signed partner feed contracts with attribution, freshness and revocation terms. |
Viewing fees and deposits can be held in M-Pesa escrow. Ask Keja AI (viewing flow) · reviewed 2026-08-31 | Planned | Copy describes a future capability. No payment provider integration, settlement model or funds safeguarding exists in this build, and none should until the legal role of the platform is defined. Path to live: Licensed PSP partnership, documented settlement and reconciliation model, and a written legal opinion on the platform's payment role. |
The platform processes any payments at all. Terms & Privacy · reviewed 2026-08-31 | Planned | No payment collection of any kind exists in this build. Nothing asks for, or should receive, real money. Path to live: Licensed PSP partnership, documented settlement and reconciliation model, and a written legal opinion on the platform’s payment role. |
Property tokenization with fractional ownership from $100. Keja Tokenize · reviewed 2026-08-31 | Simulated | Fully simulated environment: assets, tokens, ledger, KYC and distributions are demo data. No securities are offered, no money is accepted, and no blockchain is involved. Path to live: Separate legal entity, CMA classification opinion, offering documents, licensed KYC/AML and custody arrangements — education-only until then. |
KYC gates investor onboarding. Keja Tokenize (KYC modal) · reviewed 2026-08-31 | Simulated | The KYC flow demonstrates the UX only; no identity documents are collected, verified or stored — by design, since there is no backend. Path to live: A vetted identity-verification provider with purpose-limited access, explicit retention rules and audit logging. |
Accounts, roles and sessions. Auth modal · Dashboard · Admin · reviewed 2026-08-31 | Simulated | Demo-grade, client-side only: roles, sessions and password hashes live in the visitor's own browser storage. Suitable for demonstrating flows; not real security. Path to live: Server-side identity, modern password hashing, MFA and revocable sessions before any real account. |
Keja AI gives property guidance. Ask Keja AI · reviewed 2026-08-31 | Live | Deterministic intent engine with FACT / ESTIMATE / ASSUMPTION labels, explicit escalation on legal, tax, valuation and suitability questions, and a human handoff path; covered by unit tests. |
The AI is a large language model. Ask Keja AI · reviewed 2026-08-31 | Planned | It is not — it is a transparent rules engine. That is a feature (predictability, no hallucination) and is disclosed so nobody assumes otherwise. Path to live: A governed LLM integration with evaluation sets, escalation policies and hallucination monitoring. |
This register is generated from src/data/claims.ts and rendered verbatim — the site cannot claim a capability that is not declared here, and each entry carries the date it was last reviewed against the implementation.
The structural moat
Why a single-agency AI can never do this
An AI advisor owned by one developer or agency can never tell you a listing looks suspicious — even when it is. Its employer’s revenue depends on you transacting. Keja sits above multiple agencies: our incentives are aligned with the buyer, because trust is the product.
Duplicate detection, agent reputation and pricing anomalies only become possible when more than one seller’s data flows through the same verification layer. That’s why agencies onboard to Keja first, and why the trust layer compounds as the network grows.
Browse verified listingsTrust questions, answered
Straight answers
What is Ardhisasa and why does it matter?
Ardhisasa is Kenya’s official digital land information platform run by the Ministry of Lands. It records ownership, tenure, acreage and encumbrances for registered land. We cross-check every listing’s title against it where access permits — and we tell you when we couldn’t, rather than pretending.
Does a high trust score guarantee the deal is safe?
No — and be suspicious of anyone who promises that. A trust score means our checks passed as of the date shown. You should still run your own advocate-led due diligence: official land search, rates clearance, encumbrances, and a physical beacon walkthrough for land. We make that easier, not optional.
Why do you show flagged listings instead of hiding them?
Three reasons: (1) buyers deserve to see what fraud looks like, with labels; (2) hiding listings would push bad actors to less transparent channels; (3) a marketplace that only shows perfect inventory isn’t telling the truth about the Kenyan market. We show reality, clearly marked.
How do agents get onto the verified network?
Agencies apply or are invited, submit company registration and operator IDs, and their first listings go through enhanced verification. Their ongoing behaviour — complaint patterns, listing accuracy, closing honesty — feeds their reputation score. Bad actors are removed; their history stays on record.
Can I get scammed anyway?
Fraud evolves, so stay alert even on verified platforms. Our golden rules: never pay "holding fees" to personal wallets, insist on escrow, never skip the official land search, walk the beacons, and treat urgency ("someone else is paying today") as a red flag in itself.