# AWS interview field guide — product requirements

## Outcome and provenance

Create `aws-interview-guide.html`, a comprehensive, interactive, printable study document based on every supplied image in `imgs/`. Preserve the photographed question wording and answer format, explain defensible answers in depth, and connect every question to technical background, interview discussion, Leadership Principles, and an abbreviation register. English is the teaching language; German originals remain available alongside translations.

This PRD adapts `[local workspace]/amazon_aws_interview/prd.json`, `AGENTS.md`, and `ralph.py`: critique first → research → materially improve → verify → preserve evidence → repeat. The older project's unrelated service-catalog size floors do not apply to this focused document. The user's explicit request to reconstruct supplied images governs this project. Do not retrieve additional alleged assessment questions. Do not claim the photographs constitute a complete official test or that educational recommendations are Amazon's answer key. Selected radio buttons in photographs are observations, not correctness evidence.

## Required experience

1. A source-complete reconstruction: inventory all 93 images, inspect each visually, preserve every visible question and choice, consolidate overlapping photos, document cropped/unreadable/missing content without invention, and classify the Leadership Principles poster separately. Include image provenance links and confidence/uncertainty notes. Do not transcribe browser/session identifiers.
2. Quiz: preserve five-point effectiveness ratings, single/multiple choice when present, experience self-reports, and paired work-style statements. The five-point scale measures effectiveness, not agreement. Preserve visible endpoints; middle labels are explanatory study labels. Every objective option receives a specific rationale; ratings explain the chosen score, why adjacent scores fit less well, assumptions, and circumstances that would change the rating. Ambiguous cases allow alternatives and explain constraints. Self-reports/work-style choices receive nuanced interpretation and reflection, without fabricated official correct personality answers.
3. Large technical background: at least 32 substantial chapters and at least 30,000 words of technical background excluding quiz, glossary, leadership, and repeated text. Cover every concept actually present in the images and important prerequisites. Each chapter contains mechanics, request/data flow, design decisions, AWS mappings, trade-offs, failure modes, diagnosis, security, reliability, cost, a worked example, practical verification, interview follow-ups, and citations. No filler or repetitive boilerplate to satisfy floors.
4. Leadership: all 16 current Amazon Leadership Principles, each with at least 500 words of explanatory deep dive, observable behaviors, common misinterpretations, principle tensions, a truthful STAR evidence workshop, follow-up probes, and links to related quiz themes. Add at least 8 realistic prompts per principle beyond the explanatory-word floor. Explain the difference between the older 14-principle poster and the current 16 principles.
5. Register: at least 200 useful abbreviation/term entries, with expansion, plain-English definition, practical relevance, topic area, and working chapter cross-links. All technical abbreviations in chapter titles and key explanations must be covered. Include a service/technology decision register and topic-to-question coverage map.
6. Study controls: global search; chapter and quiz topic filters; study/test/review modes; reveal explanations; persistent selections, attempts, bookmarks, confidence and notes; progress export/import and reset; clear self-reflection handling; printable complete content; mobile and keyboard accessibility. Works as a standalone HTML file with inline CSS/JS/data; source photos may be relative links.
7. Evidence: source citations near factual statements, dated research ledger, iteration reports, persistent critiques, coverage and validation report, and a concise local README.

## Ralph execution

- Model exactly `gpt-6-astra`, reasoning effort `high` (the user's `gpt-6-astra-high`). Do not silently substitute models.
- At least 20 successful, material, verified fresh-process iterations. Retries do not count as additional successful iterations. No completion claim before 20.
- Each pass reads PRD, iteration task, content contract, progress, critiques, and relevant current output; records a critique before edits; opens at least two relevant primary URLs (at least one Amazon/AWS); makes substantive content improvements; runs validation; writes an honest JSON report and research/progress updates.
- Preserve per-attempt prompt, model invocation, output log, report, timestamps, material file hashes, and verifier result. Do not count failed process, invalid reports, missing artifacts, empty changes, or failed required verification.
- Initial sequence: 8 image reconstruction passes; 8 technical chapter passes; 2 Leadership passes; 1 glossary/register pass; 1 integration/final audit pass. Continue corrective passes beyond 20 if required to meet acceptance gates.

## Acceptance gates

- All 93 source images accounted for; all readable visible questions represented; overlaps and omissions explicitly recorded.
- Detailed explanations for every educational answer and distractor; no invented hidden options, false official keys, or future-interview certainty.
- 32+ technical chapters, 30,000+ substantive technical words; all observed technical quiz topics covered and cross-linked.
- 16 Leadership Principles with 128+ prompts; 200+ register entries.
- Standalone HTML passes content/schema/reference/link checks and real browser interaction checks, including persistence, import/export, filtering, print, mobile, and accessible controls.
- At least 20 actual successful GPT-6 Astra/high iterations evidenced in the runner state and reports.
- No unresolved critical/high content or engineering issues. Remaining source ambiguity is visibly disclosed and not misrepresented as resolved.