Solo Production Case Studies

Solo Production Case Studies

Status: Open for submissions · [Contributions Welcome ✍️]

Real stories of one operator shipping production software by collapsing the 12 layers. Honest accounts — including what went wrong — are worth far more than success theatre.

This page is the framework's proof. SAE makes its claims honestly, as a hypothesis — and this is where the hypothesis meets reality. Every submission that carries the four-metric scorecard (cycle time · regression escape rate · spec-drift time · blast-radius incidents) turns an assertion into evidence. We will publish the failures and the limits, not just the wins — that honesty is the point.

---

Submission template

Copy this block, fill it in, and add your case study below (or push it to the wiki):

### <Project name> — <one-line outcome>

**Operator:** <name / handle>
**Domain:** <e.g. B2B SaaS, internal tool, marketplace>
**Stack:** <tools used — link to Approved Zero-Ops Tooling List>
**Timeline:** <e.g. spec to production in 3 days>
**Context Entropy of the build:** <Low 1–3 / Medium 5–8 / High 13>

**The friction collapsed**
Which of the 12 layers did the SAE approach remove, and how?

**What single-shot vertical slicing looked like**
1–2 concrete features built end-to-end in one cycle.

**Where the Guardian earned its keep**
A bug or edge case adversarial testing caught before production.

**What the framework got WRONG for you**
Be honest. Where did "collapse the layer" bite back? What did you have
to re-introduce (a review step, a test gate, a human check)?

**Net result**
Metrics if you have them: time saved, scope shipped, cost.

---

Case studies

No community‑submitted case studies yet. Be the first — your honest write‑up (wins and scars) becomes a reference for every operator after you. The entry below is an illustrative example showing the format in action; replace this section with real submissions as they arrive.

---

📋 ILLUSTRATIVE EXAMPLE — not a real shipped project

This is a worked example to show the template filled in honestly, including the scars. It is fictional and labelled as such. Real, attributed case studies should follow this shape.

FieldNotes — internal field‑inspection tracker, spec‑to‑production in 4 days

Operator: (illustrative) Domain: Internal B2B tool — inspectors log site visits, attach photos, sync offline. Stack: Cursor + Claude Code · Next.js + Shadcn · Supabase (Postgres + storage + auth) · Vercel. See Approved Zero‑Ops Tooling List. Timeline: Spec to production in 4 days (≈ 3 focused half‑days of build). Context Entropy of the build: Medium (5–8) overall — three bounded contexts (auth, inspections, sync), decomposed into Low‑entropy slices.

The friction collapsed In a traditional setup this would have been a PM spec, a UX handoff, two engineers, and a QA pass across ~3 weeks. As an SAE build, layers 3–10 of the 12 Layers collapsed: the spec replaced the PRD, vertical slices removed the frontend/backend handoff, and the Guardian replaced the separate QA cycle. No tickets, no standups, no integration week.

What single‑shot vertical slicing looked like

1. inspections slice: schema.sql (inspection + photo tables) → routes.ts (CRUD + signed upload URLs) → view.tsx (list, detail, capture form) generated in one cycle from one spec. Types matched end‑to‑end on the first pass.
2. auth slice reused Supabase auth; spec just declared the authZ rules (an inspector sees only their own region's sites).

Where the Guardian earned its keep A fresh‑context adversarial pass on the sync slice found that two offline edits to the same inspection, synced in the wrong order, silently overwrote the newer one — a last‑write‑wins bug the happy‑path build never surfaced. It also caught a missing region check on the photo‑download endpoint (authZ was enforced on the record but not on the file). Both were high‑blast‑radius and would have shipped without the independent review.

What the framework got WRONG here

• The offline‑sync edge case wasn't in the first spec. "Spec the edge cases" is the rule, but conflict resolution for offline‑first sync simply wasn't imagined up front — the Guardian caught it, not the spec. Lesson: for genuinely novel problem domains, the first spec will have blind spots; budget a discovery pass.
• A late hot‑fix drifted from the spec for two days. Under deadline, the region‑check fix went straight into code and the spec lagged — exactly the stale‑spec failure mode the framework warns about. It was reconciled, but only because a drift check flagged it. The discipline is real because it's easy to break.
• "Vertical slice = aligned" held for structure, not concurrency. The layers lined up perfectly; the behaviour under concurrent writes did not. A reminder that single‑shot alignment is structural, not behavioural.

Net result Shipped in 4 days vs. an estimated ~3 weeks for an equivalent small team. Two serious bugs caught pre‑production by adversarial QA. One process scar (spec drift) logged and corrected. Net: the speed was real, and so were the two places it would have bitten without the non‑negotiables in place.

---

---

Reality check. We will publish case studies that report failures and limits, not just 10x wins. A framework that only collects success stories is marketing, not a playbook. If the SAE approach cost you a production incident, that lesson is exactly what belongs here.