Retail software · Live
Mags Tack
Commerce software for an equestrian shop whose working day follows riders onto the show circuit.
Project README · 2026-09-16Download README.md ↓
The brief.
Mags Tack needs a storefront and the less visible retail work behind it: inventory, purchasing, shows, loyalty, and finance.
What we built.
- A storefront connected to retail inventory workflows.
- Show workflows and purchasing tools built around the business.
- Loyalty, finance, and owner operations in the same system.
How it works.
- 01Browse the public shop by category.
- 02Follow a product through the retail workflow in an owner demonstration.
- 03Use show and purchasing workflows to manage the work beyond the storefront.
The result.
The July 13 handoff records acceptance testing of the commerce system against a historical catalog dataset. Those dataset sizes are not a statement of current sellable stock.
Where it stands.
The public shop is live. A newer storefront and supplier-direct flow are under active development; preview work is separate from completed delivery.
Earlier engineering notes · July–August 2026
Preserved from the original build record. Product status and offers may have changed; the current project summary appears above.
- Problem: acceptance journey 14 — a customer confirming email opt-in on the public path — deadlocked the database in a transaction our unit tests never hit. Fix: caught because acceptance runs on the real tenant data, patched, and re-proven the same session.
- Problem: the owner dashboard hung on 'Loading…' — module files imported from outside the served root, a dead module graph no build step flagged. Fix: relocated the shared modules, bumped the service-worker cache, and verified in-browser on the seeded real tenant.
- The storefront publishes through gates, not hope: WCAG AA contrast, landmarks, alt text, form labels, and a PII/cost/stock leakage denylist ran against all 4,323 generated pages — zero failures on the real catalog.
- Performance was measured, not assumed: barcode lookup and scan-acknowledge p95 under 6ms against the full 30,228-variation dataset, with every gate's threshold and evidence file kept in the repo.
- Rule we keep: the system runs on the owner's data from day one — the acceptance suite is twenty journeys through the shop's actual catalog, customers, and shows, not a demo dataset.