# Project export: Scout - New Age Interview Pattern This document was generated by HackStack to give an AI agent context about a hackathon project. Sections are labeled with their provenance; content marked as truncated was cut to keep this document small. ## Project metadata - Hackathon: OpenAI Build Week - Tagline: Scout is made to replace traditional style Leetcode Interviews, and replace them with real software engineering challenges, which reflects their day to day job scenarios - Devpost: https://devpost.com/software/scout-interview-pattern - GitHub: https://github.com/Navigam1108/BuildWeekProject - Video: https://www.youtube.com/embed/Tb3L3NIe17w?enablejsapi=1&hl=en_US&rel=0&start=&version=3&wmode=transparent - Team: 1 GitHub contributor(s) — Navigam (12 commits) ## Devpost submission (written by the team) ### Overview Project name Scout - Repository Interviews Elevator pitch Scout replaces LeetCode-style coding rounds with browser-based repository interviews, benchmarks, and evidence-based scoring. Built with Next.js, React, TypeScript, Node.js, SQLite, Docker, code-server, Python, C++, OpenAI Codex, GPT-5.6, Vercel AI SDK Try it out Source code: https://github.com/Navigam1108/BuildWeekProject Demo video: https://youtu.be/Tb3L3NIe17w About the project ### Inspiration Most engineering interviews still test isolated puzzles in a blank editor. That makes it easy to optimize for memorized patterns or a single AI-generated answer, while missing the work engineers actually do: understand an unfamiliar system, find the important bottleneck, preserve behavior, test the change, and explain the trade-off. Scout turns that work into the interview. ### What it does Scout gives candidates a realistic, browser-based VS Code workspace backed by an isolated Docker container. Each session starts from a production-shaped repository, engineering ticket, tests, benchmark, and intentional TODO seams. The prototype includes nine Python, TypeScript, and C++ interview repositories with 45 measurable optimization opportunities across routing, caching, scheduling, matching, payments, recommendations, search, log processing, and risk allocation. Candidates are not expected to finish every mission. The scorecard rewards meaningful progress: correctness, benchmark improvement, progress toward a golden reference, and hidden-test confidence. Two or three strong improvements can be a strong interview result. Interviewers can view elapsed time, the live candidate VS Code workspace, submissions, test evidence, benchmark deltas, mission scores, and the data-structure concepts demonstrated. ### How we built it Scout uses Next.js, React, TypeScript, SQLite, Docker, and code-server. Starting a session copies a candidate repository, applies a reproducible workload variant, launches the matching language container, and opens VS Code in the browser. The terminal and hidden grader use the same workspace, so candidate tests, benchmarks, and the submitted code run in one toolchain. The grader runs hidden invariants and benchmarks, parses a structured report, and generates the interviewer scorecard. Each pack has a candidate version, golden reference, visible tests, hidden tests, and calibrated mission rubric. The AI interview guide is honestly shown as Coming soon. This demo focuses on the working repository workflow and deterministic evidence rather than presenting an unfinished assistant as a completed feature. How we used OpenAI Codex and GPT-5.6 I used OpenAI Codex with GPT-5.6 as a development collaborator. It helped decompose the work, iterate on the Next.js and Docker workflow, structure challenge packs, improve the UI, troubleshoot grading and build issues, and refine documentation and tests. I set the interview philosophy, designed the repository missions and scoring rubric, reviewed the implementation, ran the validation checks, and made the final product decisions. GPT-5.6 is not Scout's runtime interviewer or grader: current scores are deterministic and come from tests and benchmarks. Challenges and learnings The main challenge was balancing realism with reproducible evaluation. Scout uses persisted replay variants to change workload shape and ticket framing while keeping the public API and rubric comparable. We also avoided an all-or-nothing grade because real candidates will not optimize every path in one interview. We learned that the best interview evidence is not a pass/fail label. It is a clear record of what the candidate changed, what they tested, how performance moved, and what opportunities remained. ### What's next Next steps are authentication, protected grader storage, read-only live observation, more calibrated repositories, and a constrained AI guide for repository navigation rather than solution generation. ## README (from the GitHub repository) # Scout - repository interviews for the AI era > An OpenAI Builder Week project that replaces puzzle-style coding rounds with realistic, browser-based codebase investigations. Scout gives a candidate a production-shaped repository instead of an isolated algorithm prompt. They identify and improve the bottlenecks they can reach; the interviewer receives reproducible tests, benchmark evidence, and mission-level progress. The AI interview guide is intentionally presented as a coming-soon preview while the repository experience takes center stage. The goal is not to make interviews "AI-proof." It is to make engineering judgment visible: how someone maps an unfamiliar system, chooses a bottleneck, checks invariants, improves performance, and explains the trade-offs. ## What judges can try | Capability | What it provides | | --- | --- | | **9 repository environments** | Python, TypeScript, and C++ services spanning logs, payments, caching, matching, scheduling, recommendations, search, routing, and risk allocation. | | **45 measurable missions** | Every pack has five independent performance opportunities, distributed across a multi-file codebase rather than exposed as named DSA exercises. | | **Partial-success grading** | Completing two or three missions well is a strong result. The scorecard shows correctness, benchmark progress, and proximity to a golden reference. | | **Browser-native workspace** | Each session creates a fresh repository inside a Docker-backed code-server IDE, with an integrated terminal for `make test` and `make bench`. | | **Replay variants** | A persisted per-session seed changes workload shape and ticket framing while keeping the public API and mission rubric comparable. | | **AI interview guide (coming soon)** | The demo previews a planned, constrained codebase guide instead of exposing an empty chatbox. The repository, scoring, and browser workflow are available today. | | **Interviewer evidence** | The live view shows elapsed time, submissions, visible and hidden test totals, benchmark deltas, mission cards, concepts, and the session log. | | **Open-ended sessions** | Tasks are designed for roughly an hour, but Scout never hard-stops a session; the interviewer sees elapsed time and ends it manually. | ## Fastest judge path The shortest end-to-end demo starts the Python LogScope interview. It requires Node.js **22.6+**, Git, ripgrep (`rg`), and a Docker daemon running Linux containers. Use Docker Desktop on Windows or macOS, or Docker Engine on Linux. On Linux, make sure your current user can run `docker` without `sudo`. Python is included in the candidate image, so it is not needed on the host for this flow. ### Windows PowerShell ```powershell git clone Set-Location openai_interview_platform Copy-Item .env.example .env npm ci # Builds the image needed for the LogScope Python task. docker build -t challenge-py images/challenge-py npm run dev ``` ### Linux / macOS Bash ```bash git clone cd openai_interview_platform cp .env.example .env npm ci # Builds the image needed for the LogScope Python task. docker build -t challenge-py images/challenge-py npm run dev ``` Open [http://localhost:3000/admin](http://localhost:3000/admin), click **Start session** on **LogScope**, then open the candidate workspace link shown by the app. Enter the generated IDE password when code-server asks for it. In the browser terminal, run: ```bash make test make bench ``` Click **Submit**, then open the interviewer view to inspect the grade report. Use **End session** when finished to remove the container. The first image build downloads the pinned code-server bundle and can take a little longer. Docker Hub and GitHub access are needed during that build. ### Run the entire portfolio Build all three language images before starting TypeScript or C++ tasks: ```powershell npm run images:build ``` This is equivalent to building `challenge-py`, `challenge-ts`, and `challenge-cpp` individually. The images run code-server as a non-root candidate user and include only the language runtime, compiler/tooling, Git, Make, and support needed for the browser workspace and grader. `PUBLIC_HOST` in `.env` is a **hostname or IP only**. Keep the default `localhost` for a local demo; do not include the Next.js port in that value. ## A 90-second judge demo For the clearest walkthrough, prepare one baseline session and one improved submission rather than asking a judge to implement a change live. 1. Start a LogScope session and show the task ticket, multi-file tree, and browser terminal. 2. Run `make test` and `make bench` to establish the candidate baseline. 3. Open the **AI guide** tab to show the intentional coming-soon preview rather than an unconfigured chatbox. 4. Open the interviewer view to show the matching coming-soon status alongside the session controls and replay profile. 5. Reveal a prepared candidate improvement, rerun the benchmark, and submit. 6. In the interviewer view, show hidden-test totals, mission-level timings, the replay seed/profile, elapsed time, session pulse, and session log. ## Why these interviews feel like engineering work Each pack contains a candidate repository with correct-but-inefficient paths, visible tests, a benchmark workload, and intentional TODO extension seams. The candidate can improve a coherent subset without rewriting the service. Golden implementations and hidden grader assets are included in this hackathon checkout so the evaluation is reviewable; in a real deployment they belong in private grader storage. The following labels are for interview calibration and follow-up; candidates see product tickets and service vocabulary rather than a list of DSA topics. | Pack | Scenario | Language | Interviewer-calibration concepts | | --- | --- | --- | --- | | `logscope-py` | Time-range log analysis | Python | Binary search, hashing, inverted index, heap, sliding window | | `payfix-py` | Nightly payment reconciliation | Python | Hash maps, composite keys, grouping, memoization, heaps | | `gateway-cache-ts` | Bounded API response cache | TypeScript | Hash maps, linked list, reverse index, fan-in, batching | | `orderbook-py` | Exchange matching engine | Python | Price levels, hash maps, heaps, ordered lookup, deques | | `dispatch-scheduler-ts` | Dependency-aware job scheduling | TypeScript | Graphs, in-degree, priority queues, min-heaps, interval search | | `recofeed-py` | Recommendation feed assembly | Python | Hash sets/maps, top-K heaps, cache invalidation, batching | | `route-mesh-ts` | API route resolution | TypeScript | Tries, prefix maps, prefix sums, binary search, cache invalidation | | `inventory-search-py` | Commerce inventory search | Python | Tries, inverted indexes, hash joins, binary search, pagination | | `risk-router-cpp` | Low-latency risk allocation | C++20 | Secondary indexes, prefix sums, binary search, versioned caches | ## Grading philosophy Scout is deliberately not a "finish every TODO or fail" system. Each mission has an equal default weight, and the automated score combines: - correctness and invariant evidence (35%), - improvement from the candidate baseline (35%), - progress toward the golden benchmark (20%), and - hidden-test confidence (10%). The interviewer still evaluates code quality, prioritization, communication, and trade-offs. As a guide, one meaningful mission is progress; two or three well-executed missions is a strong interview; broad, correct improvement is exceptional. See [INTERVIEW-FRAMEWORK.md](INTERVIEW-FRAMEWORK.md) for the full challenge and scorecard contract. ## How the browser workflow works ```mermaid flowchart LR C[Candidate browser] --> W[Scout web app
Next.js] W --> S[Fresh session repository
Docker container] S --> I[code-server IDE
integrated terminal] C --> I W --> D[(SQLite
sessions, reports, transcript)] C -. future AI guide .-> P[Deterministic policy gate] P -. guarded [README truncated for size] ## Detected evidence (automated analysis) Indexed codebase: 167 recognized source files, 221 KB. - Anthropic (technology) — detected in the code - C++ (language) — detected in the code - CSS (language) — detected in the code - Next.js (technology) — detected in the code - Python (language) — detected in the code - React (technology) — detected in the code - TypeScript (language) — detected in the code - Vercel AI SDK (technology) — detected in the code - Docker (technology) — claimed on Devpost, not found in the code - JavaScript (language) — claimed on Devpost, not found in the code - Node.js (technology) — claimed on Devpost, not found in the code - Vercel (technology) — claimed on Devpost, not found in the code ## Codebase structure (from repository index) ### Files (120 of 258) ``` .env.example .gitignore app/admin/page.tsx app/admin/sessions/[id]/page.tsx app/api/admin/login/route.ts app/api/candidate/[token]/chat/route.ts app/api/candidate/[token]/route.ts app/api/candidate/[token]/submit/route.ts app/api/candidate/[token]/transcript/route.ts app/api/challenges/route.ts app/api/sessions/[id]/end/route.ts app/api/sessions/[id]/report/route.ts app/api/sessions/[id]/route.ts app/api/sessions/[id]/toggle-agent/route.ts app/api/sessions/route.ts app/globals.css app/layout.tsx app/page.tsx app/s/[token]/page.tsx challenges/dispatch-scheduler-ts/agent/onboarding.md challenges/dispatch-scheduler-ts/agent/symbols.json challenges/dispatch-scheduler-ts/challenge.yaml challenges/dispatch-scheduler-ts/golden/benchmarks/bench.mjs challenges/dispatch-scheduler-ts/golden/README.md challenges/dispatch-scheduler-ts/golden/src/priority-queue.mjs challenges/dispatch-scheduler-ts/golden/src/scheduler.mjs challenges/dispatch-scheduler-ts/grader/grade.sh challenges/dispatch-scheduler-ts/grader/hidden/scheduler.invariant.mjs challenges/dispatch-scheduler-ts/grader/mission-baselines.json challenges/dispatch-scheduler-ts/repo/benchmarks/bench.mjs challenges/dispatch-scheduler-ts/repo/Makefile challenges/dispatch-scheduler-ts/repo/src/calendar-index.mjs challenges/dispatch-scheduler-ts/repo/src/dispatch-state.mjs challenges/dispatch-scheduler-ts/repo/src/ready-state.mjs challenges/dispatch-scheduler-ts/repo/src/scheduler.mjs challenges/dispatch-scheduler-ts/repo/src/worker-pool.mjs challenges/dispatch-scheduler-ts/repo/TASK.md challenges/dispatch-scheduler-ts/repo/tests/scheduler.test.mjs challenges/dispatch-scheduler-ts/repo/TODO.md challenges/dispatch-scheduler-ts/solution/solution.patch challenges/dispatch-scheduler-ts/variants/calibration.json challenges/dispatch-scheduler-ts/variants/manifest.json challenges/gateway-cache-ts/agent/onboarding.md challenges/gateway-cache-ts/agent/symbols.json challenges/gateway-cache-ts/challenge.yaml challenges/gateway-cache-ts/golden/README.md challenges/gateway-cache-ts/golden/src/response-cache.mjs challenges/gateway-cache-ts/golden/src/response-cache.ts challenges/gateway-cache-ts/grader/grade.sh challenges/gateway-cache-ts/grader/hidden/cache.invariant.mjs challenges/gateway-cache-ts/repo/benchmarks/bench.mjs challenges/gateway-cache-ts/repo/benchmarks/bench.ts challenges/gateway-cache-ts/repo/Makefile challenges/gateway-cache-ts/repo/src/cache-seams.mjs challenges/gateway-cache-ts/repo/src/cache-seams.ts challenges/gateway-cache-ts/repo/src/entry-seam.mjs challenges/gateway-cache-ts/repo/src/request-seam.mjs challenges/gateway-cache-ts/repo/src/response-cache.mjs challenges/gateway-cache-ts/repo/src/response-cache.ts challenges/gateway-cache-ts/repo/TASK.md challenges/gateway-cache-ts/repo/tests/cache.test.mjs challenges/gateway-cache-ts/repo/tests/cache.test.ts challenges/gateway-cache-ts/repo/TODO.md challenges/gateway-cache-ts/solution/solution.patch challenges/gateway-cache-ts/variants/calibration.json challenges/gateway-cache-ts/variants/manifest.json challenges/inventory-search-py/agent/onboarding.md challenges/inventory-search-py/agent/symbols.json challenges/inventory-search-py/challenge.yaml challenges/inventory-search-py/golden/catalog/__init__.py challenges/inventory-search-py/golden/catalog/service.py challenges/inventory-search-py/golden/README.md challenges/inventory-search-py/grader/grade.sh challenges/inventory-search-py/grader/hidden/test_invariants.py challenges/inventory-search-py/grader/measure.py challenges/inventory-search-py/repo/benchmarks/bench.py challenges/inventory-search-py/repo/catalog/__init__.py challenges/inventory-search-py/repo/catalog/browse_index.py challenges/inventory-search-py/repo/catalog/identity_index.py challenges/inventory-search-py/repo/catalog/indexes.py challenges/inventory-search-py/repo/catalog/models.py challenges/inventory-search-py/repo/catalog/service.py challenges/inventory-search-py/repo/Makefile challenges/inventory-search-py/repo/TASK.md challenges/inventory-search-py/repo/tests/test_service.py challenges/inventory-search-py/repo/TODO.md challenges/inventory-search-py/solution/solution.patch challenges/inventory-search-py/variants/calibration.json challenges/inventory-search-py/variants/manifest.json challenges/logscope-py/agent/onboarding.md challenges/logscope-py/agent/symbols.json challenges/logscope-py/challenge.yaml challenges/logscope-py/golden/logscope/__init__.py challenges/logscope-py/golden/logscope/ingestion.py challenges/logscope-py/golden/logscope/query_engine.py challenges/logscope-py/golden/logscope/retention.py challenges/logscope-py/golden/README.md challenges/logscope-py/grader/grade.sh challenges/logscope-py/grader/hidden/test_invariants.py challenges/logscope-py/repo/benchmarks/bench.py challenges/logscope-py/repo/logscope/__init__.py challenges/logscope-py/repo/logscope/compression.py challenges/logscope-py/repo/logscope/ingestion.py challenges/logscope-py/repo/logscope/models.py challenges/logscope-py/repo/logscope/query_engine.py challenges/logscope-py/repo/logscope/query_hints.py challenges/logscope-py/repo/logscope/retention.py challenges/logscope-py/repo/logscope/source_index.py challenges/logscope-py/repo/logscope/time_index.py challenges/logscope-py/repo/logscope/time_parser.py challenges/logscope-py/repo/Makefile challenges/logscope-py/repo/TASK.md challenges/logscope-py/repo/tests/test_query_engine.py challenges/logscope-py/repo/TODO.md challenges/logscope-py/solution/solution.patch challenges/logscope-py/variants/calibration.json challenges/logscope-py/variants/manifest.json challenges/orderbook-py/agent/onboarding.md challenges/orderbook-py/agent/symbols.json challenges/orderbook-py/challenge.yaml [138 more files omitted for size] ``` ### Dependencies - package.json: @ai-sdk/anthropic@^1.2.12, @ai-sdk/openai-compatible@^0.2.14, @types/better-sqlite3@^7.6.12, @types/dockerode@^3.3.38, @types/js-yaml@^4.0.9, @types/node@^22.15.21, @types/react@^19.0.0, @types/react-dom@^19.0.0, ai@^4.3.16, better-sqlite3@^11.10.0, dockerode@^4.0.7, eslint@^9.39.5, eslint-config-next@^15.5.20, js-yaml@^4.1.0, nanoid@^5.1.5, next@^15.3.3, react@^19.0.0, react-dom@^19.0.0, typescript@^5.8.3, zod@^3.25.20 ### Recent commits (newest first) - Remove AGENTS.md from tracking - Remove demo_recording_shotlist from cache and history - Remove images from tracking, add to .gitignore; update README and homepage - Update homepage and styles with platform screenshots and assets - Remove documentation markdown files from tracking, add to .gitignore - Update admin session page, session API, homepage, and global styles - H10: Per-session replay variants — variant seed, manifests, calibration, benchmark integration, and tests - Add supporting modules, TODO.md files, and latest refinements across all packs - Update challenges with expanded repo/golden/grader files - Complete all 7 handoff tasks - Remove build artifacts and pycache files - Initial commit ## Key source files (fetched from GitHub, selected and truncated for size) ### challenges/gateway-cache-ts/agent/onboarding.md ```markdown # Gateway project brief Gateway provides a bounded response cache for a hot API path. `ResponseCache` also owns expiry cleanup, header canonicalization, shared in-flight requests, and tag-based removal. The repository includes replay-style benchmarks and tests; run `make test` and `make bench` from the repository root. ``` ### challenges/payfix-py/agent/onboarding.md ```markdown # PayFix project brief PayFix closes a nightly transaction export. Its reconciliation module contains the matching, duplicate-reporting, currency, account-validation, and exception queue stages used by the workflow. Preserve output ordering and public behavior. Run `make test` and `make bench` from the repository root. ``` ### package.json ``` { "name": "scout-interview-platform", "version": "0.1.0", "private": true, "scripts": { "dev": "next dev", "build": "next build", "start": "next start", "lint": "eslint .", "typecheck": "tsc --noEmit", "images:build": "node scripts/build-images.mjs", "verify": "npm run typecheck && npm run lint && npm run build && node tests/test_grading.mjs && npm run variant-test && npm run test:public-host", "index-challenge": "node scripts/index-challenge.mjs", "guardrails": "node scripts/test-guardrails.mjs", "test:public-host": "node --no-warnings --experimental-strip-types tests/test_public_host.mjs", "variant-test": "node scripts/test_variants.mjs" }, "dependencies": { "@ai-sdk/anthropic": "^1.2.12", "@ai-sdk/openai-compatible": "^0.2.14", "ai": "^4.3.16", "better-sqlite3": "^11.10.0", "dockerode": "^4.0.7", "js-yaml": "^4.1.0", "nanoid": "^5.1.5", "next": "^15.3.3", "react": "^19.0.0", "react-dom": "^19.0.0", "zod": "^3.25.20" }, "devDependencies": { "@types/better-sqlite3": "^7.6.12", "@types/dockerode": "^3.3.38", "@types/js-yaml": "^4.0.9", "@types/node": "^22.15.21", "@types/react": "^19.0.0", "@types/react-dom": "^19.0.0", "eslint": "^9.39.5", "eslint-config-next": "^15.5.20", "typescript": "^5.8.3" } } ``` ### images/challenge-cpp/Dockerfile ``` FROM debian:bookworm-slim AS code-server ARG CODE_SERVER_VERSION=4.121.0 ARG TARGETARCH RUN apt-get update \ && apt-get install -y --no-install-recommends ca-certificates curl tar \ && rm -rf /var/lib/apt/lists/* \ && architecture="${TARGETARCH:-$(dpkg --print-architecture)}" \ && case "${architecture}" in \ amd64) code_server_sha256="3860893f15376e5f984492c5c92e87c51975d67e3902410f422823d9f60e06af" ;; \ arm64) code_server_sha256="de5101a3c1f86b3853431d6920dcf4daaca879f613dec9139496099e31baa569" ;; \ *) echo "Unsupported code-server architecture: ${architecture}" >&2; exit 1 ;; \ esac \ && curl --fail --location --silent --show-error \ "https://github.com/coder/code-server/releases/download/v${CODE_SERVER_VERSION}/code-server-${CODE_SERVER_VERSION}-linux-${architecture}.tar.gz" \ --output /tmp/code-server.tar.gz \ && echo "${code_server_sha256} /tmp/code-server.tar.gz" | sha256sum --check --status \ && mkdir -p /opt/code-server \ && tar -xzf /tmp/code-server.tar.gz --strip-components=1 -C /opt/code-server \ && rm -rf /opt/code-server/lib/vscode/extensions/copilot \ /opt/code-server/lib/vscode/extensions/github \ /opt/code-server/lib/vscode/extensions/github-authentication \ /opt/code-server/lib/vscode/extensions/mermaid-markdown-features \ /opt/code-server/lib/vscode/extensions/microsoft-authentication \ && rm /tmp/code-server.tar.gz FROM debian:bookworm-slim RUN apt-get update \ && apt-get install -y --no-install-recommends bash ca-certificates g++ git make python3 \ && ln -s /usr/bin/python3 /usr/local/bin/python \ && useradd --create-home --shell /bin/bash candidate \ && mkdir /workspace \ && chown candidate:candidate /workspace \ && rm -rf /var/lib/apt/lists/* COPY --from=code-server /opt/code-server /opt/code-server RUN ln -s /opt/code-server/bin/code-server /usr/local/bin/code-server ENV HOME=/home/candidate ENV PATH="/opt/code-server/bin:${PATH}" USER candidate WORKDIR /workspace EXPOSE 8080 CMD ["code-server", "--bind-addr", "0.0.0.0:8080", "--auth", "password", "--disable-telemetry", "/workspace"] ``` ### images/challenge-py/Dockerfile ``` ARG CODE_SERVER_VERSION=4.121.0 FROM debian:bookworm-slim AS code-server-download ARG CODE_SERVER_VERSION ARG TARGETARCH RUN apt-get update \ && apt-get install -y --no-install-recommends ca-certificates curl tar \ && architecture="${TARGETARCH:-$(dpkg --print-architecture)}" \ && case "$architecture" in \ amd64) code_server_sha256="3860893f15376e5f984492c5c92e87c51975d67e3902410f422823d9f60e06af" ;; \ arm64) code_server_sha256="de5101a3c1f86b3853431d6920dcf4daaca879f613dec9139496099e31baa569" ;; \ *) echo "Unsupported architecture: $architecture" >&2; exit 1 ;; \ esac \ && curl --fail --location --silent --show-error --retry 3 \ --output /tmp/code-server.tar.gz \ "https://github.com/coder/code-server/releases/download/v${CODE_SERVER_VERSION}/code-server-${CODE_SERVER_VERSION}-linux-${architecture}.tar.gz" \ && echo "${code_server_sha256} /tmp/code-server.tar.gz" | sha256sum --check --status \ && mkdir -p /opt/code-server \ && tar --extract --gzip --file /tmp/code-server.tar.gz --strip-components=1 --directory /opt/code-server \ && rm -rf /opt/code-server/lib/vscode/extensions/copilot \ /opt/code-server/lib/vscode/extensions/github \ /opt/code-server/lib/vscode/extensions/github-authentication \ /opt/code-server/lib/vscode/extensions/mermaid-markdown-features \ /opt/code-server/lib/vscode/extensions/microsoft-authentication \ && rm -rf /var/lib/apt/lists/* /tmp/code-server.tar.gz FROM python:3.12-slim-bookworm RUN apt-get update \ && apt-get install -y --no-install-recommends bash ca-certificates git libstdc++6 make \ && rm -rf /var/lib/apt/lists/* \ && useradd --create-home --shell /bin/bash candidate \ && mkdir -p /workspace \ && chown candidate:candidate /workspace COPY --from=code-server-download /opt/code-server /opt/code-server RUN ln -s /opt/code-server/bin/code-server /usr/local/bin/code-server ENV HOME=/home/candidate ENV PATH="/opt/code-server/bin:${PATH}" USER candidate WORKDIR /workspace EXPOSE 8080 CMD ["code-server", "--bind-addr", "0.0.0.0:8080", "--auth", "password", "--disable-telemetry", "/workspace"] ``` ### images/challenge-ts/Dockerfile ``` FROM debian:bookworm-slim AS code-server ARG CODE_SERVER_VERSION=4.121.0 ARG TARGETARCH RUN set -eux; \ apt-get update; \ apt-get install -y --no-install-recommends ca-certificates curl tar; \ rm -rf /var/lib/apt/lists/*; \ architecture="${TARGETARCH:-$(dpkg --print-architecture)}"; \ case "$architecture" in \ amd64) checksum="3860893f15376e5f984492c5c92e87c51975d67e3902410f422823d9f60e06af" ;; \ arm64) checksum="de5101a3c1f86b3853431d6920dcf4daaca879f613dec9139496099e31baa569" ;; \ *) echo "Unsupported code-server architecture: $architecture" >&2; exit 1 ;; \ esac; \ archive="code-server-${CODE_SERVER_VERSION}-linux-${architecture}.tar.gz"; \ curl --fail --location --silent --show-error \ "https://github.com/coder/code-server/releases/download/v${CODE_SERVER_VERSION}/${archive}" \ --output "/tmp/${archive}"; \ echo "${checksum} /tmp/${archive}" | sha256sum --check --status; \ tar --extract --gzip --file "/tmp/${archive}" --directory /opt; \ mv "/opt/code-server-${CODE_SERVER_VERSION}-linux-${architecture}" /opt/code-server; \ rm -rf /opt/code-server/lib/vscode/extensions/copilot \ /opt/code-server/lib/vscode/extensions/github \ /opt/code-server/lib/vscode/extensions/github-authentication \ /opt/code-server/lib/vscode/extensions/mermaid-markdown-features \ /opt/code-server/lib/vscode/extensions/microsoft-authentication; \ rm "/tmp/${archive}" FROM node:22-bookworm-slim COPY --from=code-server /opt/code-server /opt/code-server RUN ln -s /opt/code-server/bin/code-server /usr/local/bin/code-server RUN apt-get update \ && apt-get install -y --no-install-recommends bash ca-certificates git libstdc++6 make python3 \ && ln -s /usr/bin/python3 /usr/local/bin/python \ && useradd --create-home --shell /bin/bash candidate \ && mkdir -p /workspace \ && chown candidate:candidate /workspace \ && rm -rf /var/lib/apt/lists/* ENV HOME=/home/candidate ENV PATH="/opt/code-server/bin:${PATH}" USER candidate WORKDIR /workspace EXPOSE 8080 CMD ["code-server", "--bind-addr", "0.0.0.0:8080", "--auth", "password", "--disable-telemetry", "/workspace"] ``` ### app/layout.tsx ```typescript import "./globals.css"; export const metadata = { title: "Scout Interviews", description: "Fair, evidence-based live coding interviews" }; export default function RootLayout({ children }: Readonly<{ children: React.ReactNode }>) { return ( {children} ); } ``` ### app/page.tsx ```typescript import Image from "next/image" import Link from "next/link" const interviewSignals = [ ["Explore", "Read a real service, trace the important paths, and understand the constraints before changing code."], ["Prioritize", "Choose the improvement with the strongest likely impact instead of chasing every available task."], ["Improve", "Make a focused change while preserving behavior and existing contracts."], ["Prove", "Use tests and benchmarks to show that the repository is both correct and meaningfully better."], ] const evidence = [ ["Live workspace", "A browser-based VS Code session where candidates investigate and implement in a real repository."], ["Measured progress", "Tests, benchmarks, and mission-level progress make partial improvements visible."], ["Interview evidence", "Review the work, elapsed time, scorecard, and the trade-offs behind the change."], ] export default function HomePage() { return (
Engineer facing an empty editor and countdown clock
The hiring-signal problem

A blank editor cannot show how an engineer works.

Traditional coding rounds reward puzzle recall under artificial pressure. They rarely reveal how a candidate understands an unfamiliar system, chooses a bottleneck, or proves an improvement is safe.

See the interview problemRun the live demo
Real repositoryMultiple improvement pathsEvidence-based review
Why traditional rounds break down

One isolated answer is a weak proxy for engineering judgment.

The problem is not the candidate. It is an assessment that removes context, encourages pattern recall, and asks one answer to carry too much signal.

01

No system context

A single prompt cannot show how someone reads a service, maps constraints, or finds the real hot path.

02

Recall over judgment

Recognizing a trick is different from deciding what improvement is worth making in an existing system.

03

One-screen AI answers

When a hidden tool can answer one prompt, the interview needs stronger evidence than the final code alone.

Conceptual illustration of the AI assessment dilemma
AI exposed the flaw

Surveillance is not a better interview.

One-screen answer tools make a puzzle-only assessment harder to trust. Adding more browser locks or camera checks does not reveal how someone thinks inside a real system.

Scout changes the format instead: candidates work in context, choose a path, and demonstrate impact with evidence.

The Scout format

Start with a real system. Let engineering judgment become visible.

Scout is a growing library of production-shaped interview repositories. Every repository includes a realistic ticket, multiple files, tests, benchmarks, and several independently measurable opportunities to improve the system.

There is no prescribed solution and no expectation to finish everything. A strong candidate identifies a high-value opportunity, makes a focused change, and proves it worked.

Browse interview environments
Conceptual illustration of navigating and improving a repository
A better interview signal

One repository. Several credible ways to demonstrate impact.

{interviewSignals.map(([title, description], index) =>
{String(index + 1).padStart(2, "0")}

{title}

{description}

)}
The live interview

Review evidence, not just the final answer.

([]) const [sessions, setSessions] = useState([]) const [busySlug, setBusySlug] = useState("") const [message, setMessage] = useState("") const [created, setCreated] = useState(null) const [copied, setCopied] = useState("") const [now, setNow] = useState(Date.now()) const load = useCallback(async () => { const [challengeResponse, sessionResponse] = await Promise.all([fetch("/api/challenges"), fetch("/api/sessions")]) if (challengeResponse.ok) setChallenges(await challengeResponse.json()) if (sessionResponse.ok) setSessions(await sessionResponse.json()) }, []) useEffect(() => { void load() const timer = setInterval(() => setNow(Date.now()), 1000) return () => clearInterval(timer) }, [load]) async function copyText(value: string, label: string) { try { await navigator.clipboard.writeText(value) setCopied(`${label} copied`) window.setTimeout(() => setCopied(""), 1800) } catch { setCopied(`Could not copy ${label.toLowerCase()}`) } } async function startSession(challenge: Challenge) { setBusySlug(challenge.slug) setMessage(`Starting ${challenge.title}...`) setCreated(null) const response = await fetch("/api/sessions", { method: "POST", headers: { "content-type": "application/json" }, body: JSON.stringify({ challenge_slug: challenge.slug, candidate_name: "Demo candidate" }), }) const data = await response.json() if (!response.ok) { setMessage(data.error || "Could not create session") setBusySlug("") return } setCreated({ ...data, title: challenge.title }) setMessage("") setBusySlug("") await load() } return

Scout.
Repository interview demo
Home
Pick a production-shaped task

Browse a codebase. Start a session.

Each task is an open-ended repository investigation with five independently measurable opportunities. Sessions stay active until you end them.

{message &&
{message}
}
{created &&
Workspace ready

{created.title}

Share the workspace link and IDE password with the candidate. They can open the browser IDE immediately.

Open candidate workspace Open interviewer view
Candidate workspace link
{created.candidateUrl}
IDE password
{created.idePassword}
{copied &&
{copied}
}
}
Task catalog

{challenges.length} interview environments

{challenges.map((challenge) =>
{challenge.domain} · {challenge.repo_shape}

{challenge.title}

{challenge.summary}

{challenge.mission_count} missions · {challenge.language}

DSA coverage: {challenge.dsa_topics.join(", ")} · {challenge.intentional_todos} implementation seams

[truncated — 1120 more characters] ``` ### app/api/sessions/route.ts ```typescript import { NextResponse } from "next/server"; import { getDb } from "@/lib/db"; import { createSession, sweepExpiredSessions } from "@/lib/sessions"; export async function GET() { await sweepExpiredSessions(); const rows = getDb().prepare(`SELECT s.*, c.title FROM sessions s JOIN challenges c ON c.slug=s.challenge_slug ORDER BY s.created_at DESC`).all(); return NextResponse.json(rows); } export async function POST(request: Request) { try { const body = await request.json() as { challenge_slug?: string; candidate_name?: string; variant_seed?: number }; const result = await createSession({ challengeSlug: body.challenge_slug || "", candidateName: body.candidate_name || "Candidate", variantSeed: body.variant_seed, candidateOrigin: new URL(request.url).origin, }); return NextResponse.json(result, { status: 201 }); } catch (error) { return NextResponse.json({ error: error instanceof Error ? error.message : "Could not create session" }, { status: 500 }); } } ``` [151 more indexed source files omitted to keep this export small. The full file list is in the Codebase structure section above.]