Start here. Clone this challenge repository locally, then create a new private repository under your own GitHub account and push your completed work there. Add GitHub user paddie as a collaborator (select the Read role where available), and confirm the repository shows paddie as active or invited. Then email prm@dreamdata.io with the repository URL and submitted commit hash. The invitation may still be pending when you submit. Do not submit only in the supplied challenge repository.
Add your Go connector, tests, documentation, decision notes, and AI_USE.md, then complete the checklist in ASSIGNMENT.md. Extend this README with your connector's exact build, full-run, and incremental-run commands so it remains the repository landing page for review.
The supplied mock-crm executables are challenge infrastructure. Their source is intentionally absent; do not modify or replace them.
ASSIGNMENT.md— task, baseline success criteria, deliverables, checklist, and technical review.docs/crm-api.md— authoritative HTTP resources, schemas, relationships, pagination, tombstones, and retry contract.docs/local-mock.md— authoritative startup and process-local scenario behavior.AGENTS.md— concise working agreement for you and coding agents.
If a summary conflicts with the detailed API, follow docs/crm-api.md; for executable behavior and flags, follow docs/local-mock.md. Ask rather than inventing unspecified behavior.
Identify your host:
uname -s
uname -mSelect the matching executable. x86_64 means amd64; arm64 and aarch64 mean arm64.
uname -s | uname -m | Executable |
|---|---|---|
Darwin | arm64 | bin/darwin-arm64/mock-crm |
Darwin | x86_64 | bin/darwin-amd64/mock-crm |
Linux | aarch64 or arm64 | bin/linux-arm64/mock-crm |
Linux | x86_64 | bin/linux-amd64/mock-crm |
Windows executables are not supplied.
From the repository root, copy/paste the pair for your host:
# macOS on Apple silicon
chmod +x bin/darwin-arm64/mock-crm
bin/darwin-arm64/mock-crm --listen=:8080
# macOS on Intel
chmod +x bin/darwin-amd64/mock-crm
bin/darwin-amd64/mock-crm --listen=:8080
# Linux on arm64
chmod +x bin/linux-arm64/mock-crm
bin/linux-arm64/mock-crm --listen=:8080
# Linux on amd64
chmod +x bin/linux-amd64/mock-crm
bin/linux-amd64/mock-crm --listen=:8080Run only the matching command and leave that terminal running. For an exactly reproducible scenario, append --simulation-start=2031-02-03; otherwise startup visibly selects the current UTC date. In another terminal, check readiness and make the first page of a full pull:
curl --fail http://localhost:8080/healthz
curl --fail 'http://localhost:8080/v1/companies?page=0&size=2'Follow each next_page until it is absent. HTTP requests never advance the simulation, so repeated full and inclusive incremental pulls remain unchanged. After the full pull completes, save its greatest updated_at and make an inclusive boundary pull:
CHECKPOINT='<greatest updated_at from the completed full pull>'
curl --fail --get 'http://localhost:8080/v1/companies' \
--data-urlencode page=0 --data-urlencode size=500 \
--data-urlencode order=asc --data-urlencode "since=$CHECKPOINT"Follow next_page until it is absent. Then return to the mock terminal, press Enter on an empty line, and wait for the completed-day confirmation. Repeat the same inclusive command and follow its pages again; it now returns boundary ties plus records touched during the simulated day. Finally, rerun the full-pull command to observe the refreshed current state.
Press Enter only between completed pulls; pagination across a publication is intentionally unsupported. In connector code, derive the checkpoint from the greatest updated_at safely emitted, not merely observed. See docs/local-mock.md for the complete terminal and process lifecycle.
Keep this mock process alive for the required validation: for every implemented entity, perform an initial full pull and retrieve at least 10 Enter-driven incremental ticks using the prior emitted state each time. You can retrieve several entities after each Enter press. Your connector process may restart freely; restarting the mock starts a fresh scenario and invalidates old connector checkpoints.
Resilience is required: retry 429 and 5xx responses, honor Retry-After-Ms, and fail safely after a documented bounded retry policy. Test this by restarting the matching mock binary with --enable-fault-profile=true; the profile is disabled by default and affects /v1 only. See the assignment and docs/local-mock.md for the exact command.
With mise and Go available, build all supplied mock CRM targets from source into this repository's generated bin/ directory:
# Required once when mise first sees this repository's configuration.
mise trust
mise run buildThis produces bin/{darwin,linux}-{arm64,amd64}/mock-crm. The generated binaries are ignored; use mock-crm/scripts/build-release.sh when creating the sanitized candidate package under mock-crm/dist/.