Skip to content
arula

Stage 08 / Lab synthesis / Learn

What we learned

The balance defect was the example. The lasting skill is knowing how to turn agent output into engineering evidence another person can inspect, challenge and use.

Seven stages complete

One evidence chain, from an ambiguous issue to a qualified release decision.

The method you built

The artifacts carried trust forward.

No stage started from a blank prompt. Each began with repository evidence and an accepted record from the stage before it.

  1. 01
    Context

    Trace before proposing

  2. 02
    Specification

    Decide before coding

  3. 03
    Tests

    Make claims disprovable

  4. 04
    Audit

    Challenge the handoffs

  5. 05
    Plan

    Bind edits to claims

  6. 06
    Execution

    Work in bounded increments

  7. 07
    Review

    Qualify the decision

Issue → context → specification → tests → audit → plan → change → decision

Three course goals

What you can now explain and do.

These are not completion labels. Each is a judgment you practiced against the pinned JHipster source.

01Course goal

Shared intuition

Trust follows evidence, not fluency.

Judgments you now make
  • Agent lead
  • Repository fact
  • Observable result
  • Human decision
  • Stop signal

Use agents for speed and breadth. Use source, observable behavior and accountable decisions for trust.

Repository proof

OperationResource returns 201 after saving an operation, while the account balance remains untouched. A plausible response is not the business outcome.

02Course goal

Common language

A change becomes reviewable when its reasoning has names.

Language you now own
  • Fact
  • Inference
  • Decision
  • Claim
  • Test
  • Evidence
  • Limit
  • Owner

Keep each category explicit so an assumption cannot quietly become shipped behavior.

Repository proof

The repository can reveal nullable relationships and CRUD paths; only an owner can decide signed-amount, retry, overdraft, ownership and immutability policy.

03Course goal

Everyday habits

Trustworthy delivery is a loop, not a final review step.

Habits you now use
  • Pin
  • Disclose
  • Trace
  • Specify
  • Test
  • Constrain
  • Inspect
  • Qualify

Repeat the loop on every material agent-assisted change, not only when final review begins.

Repository proof

A status-and-row-count test misses an unchanged balance. Reading both records before and after the request turns the intended invariant into evidence.

The operating shift

The agent is no longer the source of truth.

Before

“Update the balance and make the tests pass.”

The prompt hides policy, scope, evidence and stopping conditions.
After

“Implement accepted claims C-1–C-4 at the pinned revision. Start with the failing two-record observation. Stop on any new product decision.”

The task is bounded by artifacts the team owns.

Your new default

You can use this on Monday.

01

Ground the work

Pin the repository state and distinguish observed, reported and unavailable context.

02

Make intent testable

Convert vague outcomes into owned decisions and observable state transitions.

03

Commission bounded work

Give any coding agent an outcome, accepted context, boundaries, verification and stop conditions.

04

Defend the decision

Connect claim to test to code, record actual command results and state what remains unproven.

Carry it forward

Choose the next change that will use the loop.

Name a real repository, one delivery decision and the artifact that will keep the reasoning reviewable.

My next application

For [repository and change], we will use [artifact or habit] before [delivery decision].

Owner · next use · evidence to retain · review date