Phase 11 Secondary Validation Synthesis — 2026-07-19

Status: SECONDARY/ONLINE EVIDENCE SYNTHESIS + SYNTHETIC READINESS CONTEXT — DOES NOT REPLACE PRIMARY PILOT EVIDENCE
Governing issue: #112

Purpose

The Phase 11 plan requires a real controlled Zambia pilot before formal exit. Secondary research is used only to narrow assumptions, lock conservative scope and avoid asking participants to rediscover facts that authoritative sources can establish.

Since this synthesis was first prepared, DIREKT has also completed its production-shaped synthetic controlled-pilot activation. That synthetic/system evidence proves current state-machine and integration readiness, but it does not convert any real-world question into a participant result.

This synthesis is valid for:

  • locking conservative Wave 1 scope;
  • choosing claims/features that remain disabled;
  • defining provider evidence pathways;
  • defining what real participants/operations must still measure;
  • reducing open assumptions before the real pilot;
  • interpreting synthetic readiness without overstating it.

It is not valid as a substitute for:

  • actual recruitment/onboarding completion or abandonment;
  • trust-language comprehension;
  • real enquiry/response behavior;
  • field/reviewer/support timing;
  • real-device/network failure rates;
  • willingness to pay;
  • regulator/legal approval;
  • participant consent or outcomes.

Source families used

Current authoritative/credible source families considered in Phase 11 research include:

  • Zambia Data Protection Commission — registration, rights, data-protection framework and overseas storage/transfer requirements;
  • Zambia Information and Communications Technology Authority — ICT/mobile/connectivity market evidence;
  • Zambia Statistics Agency — population, labour, device/internet context where applicable;
  • Electoral Commission of Zambia / Lusaka local-government sources — ward/constituency geography;
  • TEVETA — vocational/trade training and qualification pathways;
  • Energy Regulation Board — electrical/energy-sector regulatory boundaries;
  • National Council for Construction — contractor/construction registration boundaries;
  • PACRA — business registration/identity source;
  • Bank of Zambia — payment-system/provider boundaries;
  • Competition and Consumer Protection Commission — consumer-protection/redress context;
  • Firebase/Google official documentation — Android phone authentication and backend ID-token verification.

All real product claims must still use the exact authoritative issuer/source appropriate to the individual check.

11C — Provider cohort and evidence validation

Conclusions locked before primary pilot

Four-category boundary

Keep the approved four-category bounded scope:

  1. plumbing/water repair;
  2. electrical repair/services;
  3. motor-vehicle mechanics;
  4. appliance/electronics repair.

Reasons:

  • practical local-service discovery/use cases;
  • formal trade/qualification evidence pathways exist for relevant subsets;
  • business-registration evidence can be checked separately where applicable;
  • category-specific statutory boundaries can be tested without assuming one credential fits all providers;
  • the categories naturally include fixed-premises, mobile and hybrid operating models.

Do not add categories merely to make recruitment easier without a documented scope decision.

Three provider pathways

Retain:

  • registered businesses;
  • qualified/trade-trained individuals;
  • experienced informal providers where participation and public wording remain lawful and accurate.

This tests whether DIREKT can present partial, check-specific trust without turning formal business registration into the only route to visibility.

Claims remain decomposed

Never collapse trust into a blanket verified provider state.

Independent claim concepts may include:

  • identity checked;
  • phone/contact checked;
  • business registration checked;
  • qualification/trade certificate checked;
  • category-specific licence/registration checked where applicable;
  • premises/location checked where actually performed and permitted;
  • not checked;
  • pending;
  • expired.

Electrical category remains high-caution

A training certificate does not automatically authorize every electrical activity.

Before any public qualification/licence claim:

  • identify exact issuer;
  • identify credential type/scope;
  • identify expiry/revocation where applicable;
  • identify the authoritative regulator/issuer for the specific activity.

NCC is not a blanket artisan requirement

NCC registration may be relevant to certain contractors/projects/entities. Do not automatically require it from every plumber, electrician, mechanic or small repair provider.

PACRA is registration evidence, not competence proof

A PACRA record may support a business-identity/registration claim. It must not become a competence, safety or work-quality guarantee.

Synthetic readiness now proven

The managed synthetic checkpoint contains:

  • 24 provider fixtures;
  • exactly 6 per category;
  • exactly 2 registered-business, 2 qualified-individual and 2 experienced-informal fixtures per category;
  • 48 complete evidence/review/recommendation/decision/claim lifecycles;
  • 24 provider/category publications only after required claims existed.

This proves the current architecture can represent the intended mix and verification lifecycle.

Still requires PRIMARY-PILOT evidence

  • recruitment acceptance rate;
  • onboarding completion/abandonment by pathway/category;
  • evidence real providers actually possess/can submit;
  • rejection/resubmission reasons;
  • trust-claim comprehension;
  • real verification turnaround;
  • perceived fairness across provider pathways;
  • real support burden.

11C conclusion: scope/evidence architecture and synthetic functionality can be locked; real usability/performance outcomes remain primary required.

11D — Discovery, location and trust comprehension

Manual/list-first remains the entry mode

Do not make Google Maps a Wave 1 prerequisite.

Required discovery inputs remain:

  • approved area/ward;
  • landmark/free-text area description where supported;
  • provider service area;
  • fixed/mobile/hybrid operating model.

Location-permission denial must not block core discovery.

Geography remains bounded

Kabwata + Chilenje remains the controlled operating boundary. This does not imply representativeness for Lusaka or Zambia as a whole.

Matero remains a later comparison/expansion candidate.

Exact private provider coordinates remain private

For mobile/hybrid providers, discovery should rely on service area rather than inventing an exact public pin.

For fixed premises, public precise location should be enabled only when genuinely customer-facing, useful, separately authorized and safe.

Trust must be understood claim by claim

Ask real participants what each claim means rather than whether they simply “trust the badge.”

Synthetic readiness now proven

Synthetic providers exercise:

  • Kabwata/Chilenje locality labels;
  • mobile/fixed/hybrid operating models;
  • bounded service areas;
  • scoped claims/limitations;
  • publication gating;
  • manual/list-first boundary with Maps disabled.

Synthetic coordinates are fixture values only and are not evidence of real service-area accuracy.

Still requires PRIMARY-PILOT evidence

  • how customers naturally describe/search locations;
  • manual search success/no-result behavior;
  • permission-denial recovery on real devices;
  • service-area accuracy perception;
  • interpretation of identity/business/qualification/location claims;
  • understanding of not checked, pending and expired;
  • whether a map provides enough incremental value to justify later provider/privacy cost.

11D conclusion: manual/list-first and reduced-precision privacy architecture can be locked; comprehension and discovery effectiveness remain primary required.

11E — Enquiries, contact handoff, reviews and complaints

Keep tracked enquiry before off-platform contact

Baseline:

provider discovery
→ tracked enquiry
→ provider response
→ separate time-bounded contact handoff consent where enabled
→ interaction history
→ review/complaint eligibility

This preserves accountability without prematurely building full chat.

Do not add full chat without evidence

A request for chat is a feature request, not proof that tracked enquiry + consented handoff fails.

Automated WhatsApp/call providers remain off

Wave 1 can test whether participants understand/want contact handoff without activating automated communications delivery.

Any later provider requires legal/privacy/provider review and updated processing records.

Reviews remain interaction-linked

Review eligibility stays tied to an accountable DIREKT interaction rather than unrestricted anonymous creation.

Synthetic readiness now proven

Managed synthetic states include:

  • 60 enquiries;
  • 36 closed;
  • 12 accepted/active;
  • 6 declined;
  • 6 received/unanswered;
  • 48 tracked interactions;
  • 24 interaction-eligible reviews;
  • 6 complaint scenarios.

These distributions are deterministic test scenarios, not observed response or complaint rates.

Still requires PRIMARY-PILOT evidence

  • real enquiry completion/abandonment;
  • response time and unanswered rate;
  • customer understanding of DIREKT versus off-platform contact;
  • contact-consent/revocation comprehension;
  • review completion/moderation burden;
  • complaint rate/type and support workload;
  • whether full chat is actually necessary.

11E conclusion: interaction architecture and state-machine readiness can be locked; real response/accountability behavior remains primary required.

11F — Operations and field capacity

Field-visit claims remain disabled

No public field checked, field visited or premises visited claim until:

  • a Zambia-based operator/lead is named;
  • safety/identity procedures are approved;
  • travel/assignment workflow is tested;
  • cost/capacity can be measured honestly.

Wave 1 can still validate evidence review, corrections, complaints and queue handling without a field claim.

Four-eyes/high-risk controls remain necessary

High-impact trust changes, exceptions and overrides must preserve authorization/audit separation. Pilot convenience never justifies direct DB status fixes for real cases.

Operations capacity is a release variable

Queue age, support backlog and review time determine whether the pilot must pause/narrow; they are not merely analytics.

Synthetic readiness now proven

The synthetic run proves:

  • reviewer assignments execute;
  • evidence state transitions execute;
  • recommendations and immutable decisions execute;
  • scoped claims and publication gating execute;
  • enquiry/review/complaint state synchronization executes;
  • zero field visits were created.

Still requires PRIMARY-PILOT evidence

  • reviewer minutes per case/evidence item;
  • support contacts per participant;
  • queue ageing;
  • rework/rejection burden;
  • complaint/escalation load;
  • field-route/time/cost if later activated;
  • fraud/collusion indicators.

11F conclusion: field claims remain disabled and operations controls stay fixed; real capacity/timing/cost evidence remains primary required.

11G — Device, connectivity and reliability

Android-first remains appropriate

Continue optimizing the native Android participant path rather than introducing another participant client during Phase 11.

Design for variable connectivity

Retain:

  • timeouts and safe retry;
  • idempotent mutations;
  • interrupted upload recovery;
  • app/process restart recovery;
  • low-bandwidth behavior;
  • manual location fallback;
  • no background-location dependency.

Repository support is not real device evidence

A supported Android/API version does not prove usable performance on low-memory/low-storage devices.

The real pilot must capture recruited-device/network diversity privately/minimally.

Synthetic/system readiness now proven

Before synthetic activation, a broken Android dependency/wrapper bundle was detected through regression-first auditing, repaired and revalidated through:

  • Android CI;
  • Android performance budget;
  • supply-chain/security checks.

The backend/container/recovery/openAPI/synthetic gates also passed on the promoted Phase 11 source.

This is valuable SYSTEM-METRIC evidence, not a real Zambia device/network result.

Still requires PRIMARY-PILOT evidence

  • actual Android distribution in recruited cohort;
  • low-memory/storage failures;
  • startup/navigation latency on real devices;
  • intermittent-network failures;
  • upload retry/recovery behavior;
  • OTP delivery/auth behavior across Zambia networks;
  • font/screen/accessibility issues;
  • real crash/ANR/error rates.

11G conclusion: Android/offline-resilient architecture remains appropriate; real reliability rates remain primary required.

11H — Pricing and unit economics

Keep real payments disabled

Phase 11 measures value and willingness to pay without moving real money.

This avoids premature payment-provider/legal/tax/reconciliation risk.

Payment stays independent from trust

No subscription/payment state may automatically:

  • approve evidence;
  • create verification claims;
  • increase trust level;
  • bypass expiry/review;
  • change competence/safety wording;
  • unfairly alter ranking.

Use cost accounting rather than guessed profitability

Measure later:

provider acquisition/recruitment effort
+ evidence review time/cost
+ support/moderation cost
+ infrastructure/auth/SMS cost
+ field cost if activated
= activation/servicing cost by pathway/category

Then compare separately against stated willingness to pay.

Pricing questions should test trade-offs

Distinguish:

  • one-time verification/application fee;
  • recurring provider subscription;
  • category/pathway differences;
  • value of visibility versus trust checks;
  • willingness to pay for optional features;
  • refusal reasons.

Synthetic assumptions retained

Scenario-only monthly bands:

  • ZMW 50;
  • ZMW 100;
  • ZMW 150;
  • ZMW 250.

No price is validated. Managed synthetic activation created zero payment intents.

Still requires PRIMARY-PILOT evidence

  • perceived provider value;
  • acceptable price bands;
  • billing-cycle preference;
  • willingness to pay by category/pathway;
  • real support/review/infrastructure cost;
  • field cost if activated;
  • acquisition/activation friction.

11H conclusion: payment-disabled research model and unit-economics framework can be locked; price/revenue viability remains primary required.

Cross-stage decisions locked for Wave 1

Decision Current status
Geography LOCKED: Kabwata + Chilenje only for Wave 1
Comparison area Matero retained; not automatically activated
Categories LOCKED: four bounded categories
Cohort LOCKED: invite-only, maximum 8 providers + 20 customers per wave
Provider pathways LOCKED: registered + qualified + experienced-informal where lawful/accurate
Android LOCKED as participant client
Maps DISABLED / NOT REQUIRED for Wave 1
Precise location OPTIONAL/SEPARATE; never required for core discovery
Sentry real-participant telemetry DISABLED unless separately approved/privacy-tested
Automated WhatsApp/call provider DISABLED unless separately approved
Payments DISABLED; research only
Field-visit trust claim DISABLED until field gate passes
Full chat DEFERRED pending evidence
Blanket verified badge PROHIBITED
Firebase phone auth IMPLEMENTED TECHNICALLY; REAL CONFIG/USE EXTERNAL-GATED
Public launch PROHIBITED
Synthetic functional readiness COMPLETE
PRIMARY-PILOT validation PENDING

Effect on Phase 11

Secondary research plus synthetic/system testing has substantially reduced uncertainty about the shape of the pilot and the technical ability of the current architecture to support it.

It has not answered the core real-world questions that justify production release.

Therefore:

Scope architecture: LOCKED
Synthetic functional readiness: COMPLETE
Managed synthetic activation: COMPLETE
Regulator/legal approvals: PENDING EXTERNAL EVIDENCE
11C–11H PRIMARY-PILOT evidence: PENDING
11J decision: PENDING
Formal Phase 12 authorization: NOT GRANTED

The next real evidence must come from the approved controlled Zambia pilot, not from further relabelling of desk research or synthetic scenarios.