# Mabasa first public data snapshot / 6 November 2026

## The headline

316,273 retained source adverts found during the 56-day lookback ending 6 November 2026 at 10:43:14 CAT (08:43:14 UTC). The window begins 11 September at the same time. Earliest retained records are from 17 September: this is an eight-week lookback, not proof of eight complete weeks of continuous collection.

All job rows were included irrespective of active status. Dates use first_seen_at (local naive timestamps interpreted as UTC), falling back to found_at only if first_seen_at is absent. These describe Mabasa discovery, not original employer publication. Full raw local columns were inspected, including inactive records, split relationships, structured data and original page text. External projection kept job identifiers, titles, source URLs, dates, country/location fields and stored grouping fields. No account/CV/user-activity tables were read.

## Reconciliation

| Cohort | Stored rows | Source adverts |
|---|---:|---:|
| External sources | 315,875 | 312,589 |
| Zimbabwe-focused sites | 8,384 | 2,980 |
| South African sites | 704 | 704 |
| Total | 324,963 | 316,273 |

A source advert is reconciled within one provider using its native ID where recovered, otherwise its normalized source URL. Tracking parameters, URL fragments and trailing slashes are removed; identity-bearing query parameters remain. VacancyMail/iHarare trailing numeric IDs take precedence over changed slugs; LinkedIn numeric view IDs and Indeed jk IDs are used where present; Workday supported requisition-like URL suffixes are used where recoverable, otherwise canonical URL. Different provider adverts and reposts remain distinct. This is not worldwide cross-site vacancy deduplication, unique openings, verified still-open jobs, applicants or hiring demand.

The total includes 59 explicitly identified non-job opportunities (33 Zimbabwe-source and 26 external). They are excluded from role-family counts, retained in source inventory totals. Other unclassified adverts may include more opportunities. Thus the public headline says ‘adverts’, not ‘unique job vacancies’.

## Zimbabwe-focused sources

| Source | Stored rows | Source adverts |
|---|---:|---:|
| Vacancy Box | 1,209 | 1,066 |
| Apply Now | 1,820 | 932 |
| Vacancy Mail | 5,054 | 771 |
| iHarare Jobs | 301 | 211 |

A Zimbabwe-focused platform can publish roles abroad. These source cohorts do not verify every advert’s country. Child splits, invalidated parser products and repeat roots are collapsed into their source advert rather than counted as distinct jobs.

## Added coverage

Workday: 21,486 source adverts, first retained 24 October 2026. South African sources: Jobplacements 398; Careers24 171; Pnet 135. First South African record: 4 November 2026 at 01:38 CAT (3 November 23:38 UTC). Unequal collection periods make raw source-volume comparisons unsuitable as market-demand comparisons.

## Grouping related titles

The publication uses analysis-specific, rule-based title-evidence groups, not production taxonomy or an official occupational classification. Software/front-end/back-end/full-stack, cloud/DevOps, ML/AI, IT and data terms are grouped broadly as Software, IT & data. Generic engineering titles remain Engineering & technical roles; that bucket can still contain digital roles whose titles do not identify a specialism. Related specialisms are not being claimed to be interchangeable jobs.

Local evidence recovers original source metadata, parent titles, valid child titles, explicit ‘Job title’ labels and numbered job-role headings before detected related-post sections. Duties, employer names, generic ‘Reverify’ placeholders and oversized paragraph-like titles are not valid role evidence. This follows the prior Zimbabwe audit’s reconciliation approach, with tighter technical and finance rules.

One source advert counts once per matching family. Multi-role adverts and titles can overlap, so charts must not be summed as unique vacancies. Grouping remains imperfect: generic titles, sector qualifiers and untranslated titles can be unresolved or spill across broad groups. No employer requirement text is used to infer a person’s occupation. The unclassified count is shown beside each family chart.

Initial review caught machine-learning/SRE/API titles under generic engineering and finance-industry qualifiers misread as finance occupations. The final rules correct those cases and add AI/ML/MLOps/Kubernetes/test-automation terms. Plain ‘financial services’ does not alone imply a finance role.

Final classified external adverts: 216,504; unclassified: 96,085. Zimbabwe-source classified: 1,956; unclassified: 1,024. Display names and exact rules are in chart-data.json. These counts deliberately differ from the old P17 snapshot and should not be presented as a pure growth/time-series comparison.

## Country ranking decision

All 315,875 external rows have empty country_code. Inspected Workday payloads also include city-only and multi-location examples with missing country_iso. Reliable country classification across the full cohort was not established, so no worldwide ‘top five countries’ chart is published. Country inference from city names was not guessed. Future country reporting needs explicit normalized country provenance and a stated multi-location counting rule.

## Audit and reproducibility

Read-only PostgreSQL transactions through the existing Tailscale SSH connection. Cohort fixed at 2026-11-06T08:43:14.263920Z; extract started 08:44:24 UTC. Snapshot SQL, private compressed raw evidence, per-advert title evidence, source identities, refinement script and review samples remain in Documents/mabasa/analysis/market-snapshot-20261106, outside the public carousel server.

The full extract has 9,088 local and 315,875 external rows, checked against SQL expected counts. SHA256: ab48c0535af31d15b297d0414b59763684271440c5b3eb18a971a6970856093c. Original full local extract was streamed and retained privately. No source scraper, production database content or deployment configuration was modified.

Publication files contain aggregate counts and methods, not the raw full-row export. The cover and closing photographs reuse approved P17 nature artwork. No new synthetic proof or testimonials were generated.

## Writing references

The supplied creator documents were read as references. Precise virality formulas, fabricated stories, unsupported follower/engagement multipliers and claims that dwell time guarantees distribution were not adopted.

- UK Analysis Function, Writing about data: https://analysisfunction.civilservice.gov.uk/policy-store/writing-about-data/
- ONS chart-text guidance: https://service-manual.ons.gov.uk/data-visualisation/guidance/chart-text
- Mabasa CONTENT-WRITING-RULES.md and prior Zimbabwe analysis.


## Remote job titles / 6 November 2026

The carousel now has 12 slides. Slide 10 shows the five most repeated exact titles across Himalayas, RemoteFirstJobs, Jobicy, Remote OK and Remotive. Customer Success Manager:124; Sales Development Representative:120; Senior Software Engineer:118; Account Executive:115; Business Development Representative:101. Calculated from all 57,501 reconciled adverts in the frozen snapshot. Non-job opportunities excluded. Case and spacing are normalized; seniority and other title wording remain separate. Each advert counts once per distinct title. The same vacancy may appear on several sites. These are title frequencies, not application or viewing counts.
