Introduction evidence decisions

Every project was considered separately. Extra material is included only where it explains the application, failure, interface or a decision-driving constraint. These are editorial decisions, not automated checks or claims of production observation.

Project	Intro material and reason
Build a private bookmark API with ownership and version checks	Sizing: 10,000 users with 20 bookmarks each means 200,000 records
Build short links with unique aliases and safe redirects	Sizing: 30 million creations/day is roughly 300 requests/s (RPS) using 100,000 seconds/day, or 347 RPS using the exact day length
Enforce API quotas across concurrent gateways	Proposed response shows what rejection means to a real caller.
Build checkout that recovers from uncertain payments	Animated failure order and proposed status response make a lost reply distinct from a failed charge.
Record permission changes with durable audit evidence	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Enforce tenant access in APIs, caches and exports	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build restartable CSV export jobs	Ownership-takeover animation explains why late completion must be rejected.
Schedule reports without duplicate logical runs	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Erase account data across stores and in-flight work	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Migrate tenant data with a resumable backfill	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Design and rehearse regional write failover	Sizing: At 3,000 writes/s, a replica that is 20 seconds behind may be missing roughly 60,000 recent writes
Reserve rooms and handle recurring local times	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Reserve concert seats with expiring holds	Sizing: The sale receives 60,000 purchase attempts/s for 20 seconds: 1.2 million attempts for 30,000 seats
Assign drivers safely with expiring offers	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build chat with durable messages and reconnect recovery	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build a versioned shared document editor	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Deliver notifications with preferences and priority	Sizing: One million campaign recipients in ten minutes means about 1,667 submissions/s before retries
Deliver signed webhooks with retries and replay	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build restaurant discovery and authoritative checkout	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build a following feed with current access checks	Sizing: At 120,000 feed requests/s with 30 items per page, the service may need to consider 3.6 million candidate items/s
Collect news feeds with freshness and deduplication	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Process uploads and publish complete video renditions	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build resumable uploads and authorized video playback	Sizing: Five million uploads/day averages about 58 uploads/s
Run programming submissions inside isolated workers	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build versioned API routing and admission policies	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Ingest events with durable acceptance and replay	Sizing: A burst of 100,000 events/s at 1 KiB each carries about 97.7 MiB/s of raw input
Aggregate click events with late arrivals and reconciliation	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Compute trending topics from duplicate and late events	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Protect a database with versioned cache fills	Real code excerpt explains stale fills and explicitly names the missing distributed atomicity. Sizing: The assumed two million reads/s cannot all fall through to a database budgeted for 10,000 reads/s
Implement replicated writes and fenced leadership	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Search documents without leaking revoked content	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build typeahead with stale-response protection	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Synchronize files with resumable uploads and conflicts	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build a durable crawler with per-host limits	Sizing: One billion waiting URL entries at 200 bytes each is about 200 GB before indexes and history
Release invoice changes with stable cohorts and rollback	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Find database-pool waiting in slow API requests	Actual local diagnostic output and phase animation show why query duration alone misses the incident.
Ingest and query metrics with bounded cardinality	Illustrative telemetry input reveals why unique labels increase cardinality. Sizing: 300,000 hosts reporting every ten seconds produce 30,000 host reports/s
Reject excess API work before queues grow without bound	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build a document assistant with current permissions	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Draft support replies without granting tool authority	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Serve recommendations with safe fallback ranking	Sizing: The 250 ms response target has to include the whole user-visible operation
Trace requests through a bookmark API	Actual save request/response plus animated current local flow. No traffic calculation is needed to understand overlapping requests.
Enforce a single deadline across API dependencies	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build an SSRF-resistant link preview fetcher	Actual policy-model output explains the private/mixed-address cases without presenting a complete secure HTTP client.
Evolve tag responses without breaking old clients	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Move title lookup into restartable background jobs	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build duplicate-safe form submission and CSV export	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Store receipts with reviewable extraction and corrections	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build a shared reading-list UI with private reading state	Actual supplied conflict response anchors the UI behavior in a concrete interface.
Prevent overlapping shifts and enforce manager access	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Measure whether existing checks detect real defects	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Protect API response types, units and compatibility	Small API-spec excerpt exposes the precise type and unit issue.
Reproduce and remove order-dependent failures	Illustrative failure transcript makes the hidden input visible.
Measure API capacity with controlled arrival rates	Sizing: 120 arrivals/s minus 100 completions/s adds roughly 20 waiting requests each second when nothing is rejected
Design and run a synthetic reading-list journey	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Deploy from main and roll back compatible artifacts	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Diagnose a reading-list incident from existing telemetry	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Define and calculate a user-facing save SLO	Sizing: For one million eligible requests and a 99.9% success objective, the allowed failed-request count is 1,000
Implement burn-rate alert and incident state rules	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Bound retries across browser, API and SDK layers	Actual local output connects retry multiplication and remaining time to the user operation.
Prioritize API work within a fixed capacity budget	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Recover a service trapped in expired work and retries	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Keep bookmark saves usable when title lookup fails	Actual supplied function makes optional enrichment and controlled dependency behavior visible.
Rehearse detection, rollback and service recovery	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build ingestion with bounded backlog and explicit rejection	Sizing: During the ten-second burst, producers offer 1,000 events while workers finish at most 200
Build evidence-backed answers with permission rechecks	Proposed supported/abstaining response pair shows what the user receives.
Require exact human approval before agent actions	Illustrative action record gives exact meaning to parameter-bound approval.
Route extracted invoices through validation and review	Illustrative inconsistent invoice makes content errors different from transport failure.
Track AI evaluation evidence and serving versions	Small evaluation slice table shows why an average cannot decide permission safety.
Decide whether an AI feature improves a reading list	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Migrate tags across data, clients and workers	Before/after API data shapes make compatibility work concrete.
Build a service template with overridable defaults	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Write a data-platform policy from concrete decisions	Prose only: the concrete workflow or decision in the opening is sufficient. No scale, code or animation is needed.
Compare a small export script with a custom platform	Prose only: the concrete workflow or decision in the opening is sufficient. No scale, code or animation is needed.
Stage 1: Build the shared reading-list application	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Stage 2: Deploy, back up and recover the reading list	Explicit incident timeline explains recovery-point evidence without inflated traffic.
Stage 3: Add durable jobs and bounded caching	Sizing: 200 readers making one request/s offer 200 reads/s to a database budgeted for 100
Stage 4: Add optional AI tag suggestions	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Stage 5: Migrate the reading list to stable tag IDs	Illustrative stale-backfill timeline anchors version and deletion handling.
Record and revisit uncertain engineering decisions	Prose only: the concrete workflow or decision in the opening is sufficient. No scale, code or animation is needed.
Specify tag behavior for independent implementers	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Evaluate generated rate-limiter code with a simple reference	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build reusable AI instructions and evaluate transfer	Prose only: the concrete workflow or decision in the opening is sufficient. No scale, code or animation is needed.
Collect evidence about changed behavior and affected callers	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Rewrite private commit history to explain a change	Prose only: the concrete workflow or decision in the opening is sufficient. No scale, code or animation is needed.
Split a tagging feature into runnable changes	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Check the combined behavior of independently valid changes	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Automate deterministic review rules and retain human judgment	Prose only: the concrete workflow or decision in the opening is sufficient. No scale, code or animation is needed.
Demonstrate CI enforcement in a disposable repository	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.
Build a link monitor with durable history and change alerts	The user story and action/result walkthrough explain the key behavior. Keep the detailed workload and implementation evidence later in the page.