What we found, what we built, and what is still only a plan

1. Purpose and status of this document

This is a review and rebuild brief for Pr-ed Institute. It accompanies a working review build of a public website and a candidate professionalisation platform. The build exists so that decisions can be made against something real rather than a slide deck.

Nothing in this document or the accompanying build is a council endorsement, an accreditation claim, a registration outcome or a compliance certificate. All standards named are targets. All candidate records shown are synthetic.

Observations about the existing public site were taken from a desktop inspection on 11 September 2026. The comparison date for this brief is 14 September 2026. Where a source could not be retrieved, that is stated rather than filled in.

2. Evidenced observations from the 11 September inspection

The following were observed directly on the public pages inspected: the homepage, services, pricing and contact pages. They are described as findings about those pages only.

• Authenticated admin portal access was blocked by the permission system. No claim is made here about the portal's interior, its current features, its security or its usability. That area remains uninspected.
• A logo resource named PREd---logo---White.png returned 404. Two image requests failed on the pages inspected, so the brand mark was missing in at least one position.
• No portal link was visible in the navigation or the footer, so a returning candidate has no obvious way back into an authenticated area from the public site.
• Homepage content repeated across sections, and the differentiating journey — what actually happens to a candidate over time — was weak.
• Outcome wording was absolute in places without supporting evidence on the page. Absolute claims about results should either carry evidence or be reworded.
• A shared generic page title was used across pages, and the served shell extracted as JavaScript-dependent. This warrants a rendering and SEO investigation. It is not by itself proof of de-indexing, and no indexing status was measured.
• The assessment feedback duration was stated as one hour in the service description and thirty minutes in the pricing area. The two do not agree.
• The Masterclass PayFast total appeared as R6,500 while the list price is published excluding VAT. This warrants reconciliation of price display and tax treatment. It is not a finding of a tax breach.
• Form fields used placeholder text without associated labels, which fails a basic accessibility expectation and also loses the label once typing starts.
• An optional CV upload was offered without a contextual explanation of what happens to the document.
• Terms and refund text was visible as prose, but the referenced documents were not linked.

3. Positioning

Pr-ed Institute prepares people for professional registration across South Africa's construction, engineering, built-environment and environmental professions. It is an independent preparation provider. The council is always the decision-maker.

• Training completion and preparation progress are never professional registration status.
• Expanded services, mentor coverage and any fees for newly added professions are proposals and need Pr-ed confirmation before publication.
• No council endorsement or CPD accreditation is asserted anywhere in the build.
• The typographic Pr-ed wordmark used in the build is provisional. No council logo or accreditation seal is recreated.

4. Expanded council and category matrix

The registry below is configurable data, not hard-coded page content. Each record carries a council, category, profession, discipline, stage set, rule version, source URL, review date, evidence template, preparation themes, mentor requirements and publication status. Membership of the CBE core group is held separately from adjacent regulators.

SACPCMP detailed content and SACQSP detailed requirements could not be retrieved during inspection. Those categories are published in the build with an explicit pending status and no invented criteria.

5. Voice service blueprint

Voice capture is the centre of the product, not a feature bolted onto a text form. A candidate must be able to complete the journey without authoring narrative text: speaking, confirming, choosing dates and categories is enough. Keyboard editing remains available throughout as an accessibility and correction route.

1. Capture. A user-triggered consent step, then getUserMedia and MediaRecorder in the browser. Obvious recording state, timer, pause, resume, stop, playback, per-answer re-record, delete, and optional audio file import. All microphone tracks are stopped on stop and on unmount. Nothing listens on page load.
2. Guided interview. One question at a time, readable on screen, with optional text-to-speech. Nine core questions, plus optional follow-ups tailored to the pathway family.
3. Transcription. Browser speech recognition is offered only where the browser supports it and only after explicit consent, with disclosure that audio may be sent to the browser's speech service and that a connection is needed. On failure, refusal or absence, the audio is retained for playback and download and the answer is marked as needing transcription. Canned text is never substituted.
4. Structuring. Only candidate-confirmed utterances are compiled, under Role, Context, Actions, Judgement, Outcome, Reflection and Supporting evidence. Missing specifics are requested, never invented. Anything the system proposes is labelled a suggestion.
5. Correction. Re-recording a single answer creates a new version; earlier versions remain visible and the candidate must confirm again.
6. Review. A confirmed account can be sent to a mentor queue. Mentor actions in this build are demonstrations.
7. Export. A reviewed preparation draft can be exported as Markdown or printed.

Preparation states are: recorded, transcribed, candidate confirmed, mentor review requested, revisions requested or reviewed for preparation, and exportable preparation draft. None of these is a council status.

• Language: the build requests en-ZA by default. Which languages actually work depends on the browser and its speech provider. No promise is made that all South African languages are supported. Multilingual validation is a planned work item, not a delivered capability.
• Provenance: each transcript records its source (browser speech recognition, typed, sample data or server provider) and links to the question and segment. Word-level timestamps are not claimed, because the browser API does not reliably supply them.
• Storage: in this review version audio and data stay on the device in IndexedDB. Audio is never placed in localStorage. The storage position is explained in the capture consent step rather than as a persistent banner.
• Server transcription: a provider adapter is structured but not connected. No paid third-party connector is silently enabled and no output is fabricated when credentials or permission are absent.

6. Intelligence dictionary and sample internal rules

The build computes indicators from the data actually present and recomputes them when data changes. There is no single probability of registration, and no randomised or hard-coded success percentage.

Sample internal rules, all labelled provisional and all editable in the admin rule registry:

• Fourteen days without a planned activity prompts an internal check-in. Paused and on-leave enrolments are excluded.
• A review waiting more than seven days prompts a mentor or administrator.
• A repeated knowledge check gap on the same theme suggests relevant learning plus a practical assignment.
• An unmet experience requirement suggests supervisor-approved project exposure rather than another generic course.

Every recommendation card shows the observed facts, the date range, the rule and reason, the source data, what is missing, the recommended action, the owner and a due date, with acknowledge, snooze and resolve actions. Missing data is labelled insufficient information rather than treated as zero. There is no personality, emotion, accent or demographic scoring anywhere in the model. No external notification is sent; actions create internal demonstration intervention records only.

7. Role and permission map

In the review build the role switch is a demonstration view only. It is not a security boundary. Production requires authenticated, tenant-scoped access with row-level rules.

8. Information architecture

• Public: home, pathways (explorer and per-pathway detail), how it works, training, mentorship, for employers, fees, institute, support, review and rebuild brief.
• Candidate portal: home (your next step), my pathway and plan, voice studio, evidence library and item detail, training, mentorship, insights, CPD, settings.
• Programme areas: cohort insights with filters and interventions, admin rule registry.
• Every card in the portal links to a working view. There are no decorative dashboards.

9. Proposed production data model

• organisations — tenant boundary for employers and Pr-ed itself.
• users — identity, role assignments held in a separate role table, never on the profile row.
• enrolments — one per pathway per person; the isolation boundary for all evidence.
• pathway_rule_versions — immutable versions of a council category's requirements, with source URL, retrieval date and reviewer.
• projects — candidate work context, optionally linked to an employer.
• experience_sessions — a captured interview, with mode and preparation state.
• audio_assets — private object storage references, duration, mime, checksum, retention class.
• transcript_versions — text, provider, language, created_at, superseded_by.
• evidence_links — mapping from confirmed segments to rule-version requirements.
• learning_attempts — lesson and knowledge check results.
• mentor_assignments and reviews — assignment, state, structured feedback, next actions.
• interventions — rule-generated internal actions with owner, due date and outcome.
• events — append-only activity log that drives the indicators.
• cpd_records — activity, certificate, reflection, evidence and renewal planning entries.
• outcomes — recorded council results where a candidate chooses to share them, never inferred.

10. Production design requirements

• Authenticated, tenant-scoped backend with row-level access control on every table.
• Private audio and object storage with short-lived signed links; no public buckets.
• Consent, retention, deletion and audit records for every piece of candidate material.
• Server-side transcription with an async job queue, retries and idempotency keys.
• Any AI assistance grounded in a versioned, approved source set, with provenance shown and a human gate before anything reaches a mentor or an assessor-facing document.
• Redaction and controlled access where employer project material is involved.
• Verified supervisor and referee processes, with an auditable verification trail.
• A change-review process for source rules: when a council updates a requirement, affected enrolments are identified and reviewed.
• Candidate material is not used to train models by default. No real candidate data is used in demonstrations. Actual retention periods and any cross-border processing are approved before launch.

Targets: WCAG 2.2 AA; Core Web Vitals at p75 of LCP 2.5 seconds or less, INP 200 milliseconds or less, CLS 0.1 or less; ISO 9241-210 human-centred design process; POPIA-appropriate privacy practice and OWASP-appropriate production controls. These are targets to be verified, not achievements to be claimed.

11. Benchmarks

Useful patterns were noted from three public examples, described in general terms and not reproduced: PMI's certification pathway pages, which make eligibility routes and the steps between them explicit before anyone applies; RICS candidate tooling, which treats structured experience recording and mentor review as the core of the candidate relationship; and PeopleCert's learning and certification lifecycle, which keeps learning, assessment and certification visibly separate. No audited scores or comparative rankings are claimed.

12. Services and fees as inspected

• Published excluding VAT as at 11 September 2026. Validity period and tax treatment to be confirmed.
• Assessment feedback duration is displayed as "Confirm duration" because the inspected sources disagreed.
• No fees are invented for ECSA-specific services or newly added professions.
• No checkout exists in this build.

13. Indicative roadmap

Indicative and conditional. Sequence matters more than dates, and every stage depends on professional content sign-off and on decisions Pr-ed has not yet made.

1. Discovery and content audit: confirm the current offer, confirm the fee schedule, resolve the duration and VAT discrepancies, and agree the wording rules for outcome claims.
2. Core foundations: authenticated tenant backend, pathway registry with versioned rules, consent and retention model.
3. One voice pathway pilot: a single council category end to end with server transcription and real mentor review.
4. Extend to all six CBE core councils, each gated on professional content sign-off for that council.
5. Adjacent pathways: SACPLAN, SAGC, EAPASA and SACNASP, with validated categories.
6. Cohort intelligence: employer views, interventions and consent-aware reporting.
7. Migration and release: content migration, redirects for existing URLs, accessibility and performance verification, then launch.

No claim is made that a full production rebuild has been completed. What exists today is the review build described in the deliverable ledger below.

14. Deliverable ledger

15. Testing actually performed

Only tests actually run are listed. No Lighthouse scores, audit results or performance measurements are reported, because none were run in a representative environment.

• TypeScript compilation and production build of the review application.
• Automated browser rendering checks of the public and portal routes at desktop and mobile viewport sizes.
• State transition checks: confirmation blocked on empty transcripts, confirmation invalidated after an edit, review restricted to its own enrolment, deleted audio removed from storage, paused enrolments excluded from inactivity flags, and indicators recomputing when data changes.
• Microphone behaviour was exercised only as far as an automated browser allows. Real-device capture across browsers, and speech recognition across languages, remain untested.

16. Sources consulted

• https://www.pr-edinstitute.com/ and its /services, /pricing and /contact pages
• https://cbe.org.za/Pages/Professionalisation.asp
• https://www.ecsa.co.za/resources/registration-documents/ and the ECSA category pages for professional engineer, engineering technologist, engineering technician and certificated engineer
• https://sacpcmp.org.za/ (current detailed retrieval restricted; validate)
• https://www.sacapsa.com/registration/categories
• https://www.saclap.org.za/ landscape architecture and landscape management category pages
• https://www.sacqsp.org.za/ (detailed retrieval blocked; validate)
• https://www.sacpvp.co.za/
• https://sacplan.org.za/, https://sagc.org.za/, https://eapasa.org/, https://www.sacnasp.org.za/
• https://www.pmi.org/certifications/project-management-pmp, https://www.rics.org/candidates, https://www.peoplecert.org/
• https://www.w3.org/TR/WCAG22/, https://web.dev/articles/vitals, https://www.iso.org/standard/77520.html, https://owasp.github.io/www-project-top-ten/, https://inforegulator.org.za/popia/
• https://developer.mozilla.org/en-US/docs/Web/API/SpeechRecognition and https://developer.mozilla.org/en-US/docs/Web/API/MediaRecorder

Third-party pages are paraphrased, not reproduced.