Practice · turn understanding into visible evidence

You have worked through a lesson and want to find out what you can now do independently. Choose one session below. Each has a concrete product task, files you may inspect, and a separate answer pack with changed requirements. For example, the coding session asks you to consume inventory events without counting a repeated delivery twice. The full-stack session asks you to keep a newer draft visible while an earlier save completes.

Your next action: open one candidate brief, read its contract and allowed materials, and make your own attempt. If another person is assessing you, give them the matching assessor pack. If you are working alone, keep that pack closed until your first attempt, then use its schedules to find a gap. Record the observation and a specific next task in the attempt record.

All assessment prompts here are constructed, without company attribution. Training timeboxes are not claims about employer round lengths. Keep reference checks separate from assessed learner work. A green repository suite tests supplied code; it never establishes a candidate's independent skill.

Start with an observable diagnostic

Choose coding plus practical/debug, then design and the role-specific full-stack or infra session. Give the candidate only the linked candidate page and explicitly allowed files. The assessor keeps the separate pack closed until its release time. For genuine unfamiliarity, another person chooses the variant; a public repository cannot technically conceal its answers. If a candidate has read a variant, retire it for that candidate and use a changed representation on a later date.

The pack includes fully scored senior and lead examples and an attempt record. The examples are fictional calibration performances, not claims about a learner or hiring outcome.

Curriculum gate: on at least two different occasions, complete unfamiliar variants with an independent assessor; include both implementation and design/role evidence. For senior, seek at least 2 in correctness, implementation, clarification, and operations/security on each occasion, with no unresolved core-invariant failure. For lead, retain that technical floor and demonstrate 3 in ownership/influence and at least one cross-team tradeoff/architecture dimension. A missed gate names the next practice task; totals do not cancel a data-loss or authorization failure. These thresholds organize learning. They are not validated hiring cutoffs or a pass probability, and a solo mock cannot demonstrate real multi-team execution history.

Worked diagnostic: a learner names idempotency but overwrites an equal ID with a different payload. Score correctness 0 or 1 according to whether they recover with prompting, not 3 for vocabulary. After debrief, use an inventory reservation event instead of a transaction import; require the new invariant and a runnable regression.

Design mocks

Use the worked designs as prompts before reading their solutions. Record the session or keep a timestamped outline. At minute 25, introduce a changed constraint: ten times the burst, offline clients, stricter consistency, a failed region, or half the budget. Ask the candidate to revise the design, not merely name another service.

Times are training choices, not promised company round lengths.

Behaviorally anchored rubric

Score each dimension 0–3. Use dimensions individually; a total is not a hiring prediction.

For junior practice, establish correctness and a complete small flow first. For senior, expect independent failure/tradeoff reasoning. For staff, add cross-team scope and executable sequencing while retaining technical correctness. Do not demand staff scope from a junior candidate.

Behavioral and project deep dives

Prepare real stories, not invented metrics. Use this sequence: context → constraint → your action → alternatives → evidence/outcome → reflection. Distinguish “I” from “we.” If you do not know a historical number, state the uncertainty and use a verifiable qualitative outcome.

A worked fictional example: a team wants a cache for a slow endpoint. You measure the query, discover a missing owner/time index, and propose an index rollout. Compare read improvement, write overhead, and the avoided invalidation work. A strong story includes what you measured and who accepted the tradeoff. Replace this with a true project; do not present the fictional story as your experience.

A useful practice week

Choose three sessions: one concept + implementation, one coding/practical mock, one design + project-story mock. Spend the next session on the weakest observed dimension. Repeat the failed exercise with a changed input after a day, then a week. Adjust session lengths to your schedule; do not equate pages read with readiness.

Feedback log

Date / target role / prompt:
Round mode: independent | AI allowed
Assumptions stated:
Invariant and implementation:
Counterexample or failure found:
Weakest rubric dimension, with evidence:
Next exercise and success criterion:
Revisit date: