# The Biergarten Data Pipeline (Current — Synchronous Data Path)

The only concurrency in the current pipeline is a dedicated log dispatcher thread that drains `log_channel` and forwards entries to spdlog for the entire run (joined during shutdown after `log_channel` closes). `BiergartenPipelineOrchestrator::Run()` itself is synchronous — sampling, enrichment, generation, and export all happen on the main thread.

```mermaid
flowchart TD
    Start([start]) --> ParseArgs["main.cc: ParseArguments(argc, argv)"]
    ParseArgs --> ArgsValid{Arguments valid?}
    ArgsValid -- no --> LogErr[Log error / usage info] --> Stop1([stop])
    ArgsValid -- yes --> Init["main.cc: Init OpenSSL global state and LlamaBackendState"]
    Init --> DI["main.cc: di::make_injector(...)<br/>binds ILogger, IEnrichmentService, DataGenerator, IExportService, IPromptFormatter, WebClient based on --mocked vs. model-backed mode"]
    DI --> LogThread["main.cc: create BoundedChannel log_channel, spawn log dispatcher thread (LogDispatcher::Run)"]
    LogThread --> CreateOrch["main.cc: injector.create BiergartenPipelineOrchestrator"]
    CreateOrch --> Run["BiergartenPipelineOrchestrator::Run()"]

    Run --> SqliteInit["SqliteExportService.Initialize()<br/>builds biergarten_seed_&lt;UTC datetime&gt;.sqlite (numeric suffix if it exists), opens DB connection, executes schema DDL, begins transaction"]
    SqliteInit --> QueryLoc[Orchestrator: QueryLocations]
    QueryLoc --> LoadLoc["JsonLoader.LoadLocations(locations.json)"]
    LoadLoc --> SampleLoc["std::ranges::sample(all_locations, --location-count) — defaults to 10"]

    SampleLoc --> EnrichLoop{For each sampled Location}
    EnrichLoop --> GetContext["IEnrichmentService.GetLocationContext(loc)<br/>WikipediaEnrichmentService fetches a generic 'brewing' extract and a 'beer in {country}' extract (not city/region-specific); MockEnrichmentService (--mocked) returns an empty string"]
    GetContext --> LookupFailed{Lookup failed?}
    LookupFailed -- yes --> SkipCity1[Log warning, skip city] --> EnrichLoop
    LookupFailed -- no --> StoreEnriched["Store EnrichedCity{Location, region_context}"] --> EnrichLoop
    EnrichLoop -- done --> GenUsers[Orchestrator: GenerateUsers enriched_cities]

    GenUsers --> LoadUserData["JsonLoader: LoadPersonas / LoadForenamesByCountry / LoadSurnamesByCountry<br/>each memoized after first parse; MockCuratedDataService (--mocked) returns a fixed in-memory dataset instead, skipping these JSON files"]

    LoadUserData --> UserLoop{For each EnrichedCity}
    UserLoop --> SampleName["SampleName(forenames_by_country, surnames_by_country, location.iso3166_1)"]
    SampleName --> NamesAvail{Names available for country?}
    NamesAvail -- no --> SkipCity2[Log warning, skip city] --> UserLoop
    NamesAvail -- yes --> SamplePersona["Sample UserPersona (uniform random)"]
    SamplePersona --> UserGenMode{Generator mode}

    UserGenMode -- MockGenerator --> UserMockGen[DeterministicHash and Format]
    UserGenMode -- LlamaGenerator --> UserLlamaLoad["prompt_directory_.Load(USER_GENERATION) → USER_GENERATION.md in --prompt-dir"]
    UserLlamaLoad --> UserLlamaRetry["Infer(...) with kUserJsonGrammar → ValidateUserJson<br/>retry up to 3 attempts, appending validation error to the retry prompt on failure"]
    UserGenMode -- OpenAIGenerator --> UserOAILoad["prompt_directory_.Load(USER_GENERATION) — same prompt file as LlamaGenerator, content is backend-agnostic"]
    UserOAILoad --> UserOAIRetry["CallChatCompletionsApi(...) with kUserJsonSchema → ValidateUserJson<br/>retry up to 2 attempts — Structured Outputs (json_schema, strict) already guarantees schema shape, so retries only cover transient failures / placeholder text"]

    UserMockGen --> UserGenDone{Generation successful?}
    UserLlamaRetry --> UserGenDone
    UserOAIRetry --> UserGenDone

    UserGenDone -- yes --> UserPostProc["GetRandomCoordsWithinRange (5km jitter), BuildEmail, GenerateDateOfBirth<br/>email and date_of_birth are generated programmatically, never LLM-authored; generation + coordinate jitter run inside one try/catch per record"]
    UserGenDone -- no --> UserGenFail["Log warning: generation failed, skipping"] --> UserLoop

    UserPostProc --> UserExport["SqliteExportService.ProcessRecord(UserRecord)<br/>reuse cached location_id or insert Location and cache its id; insert User (FK location_id)"]
    UserExport --> UserExportFail{Exception during insert?}
    UserExportFail -- yes --> UserExportWarn["Log warning: SQLite export failed"] --> UserLoop
    UserExportFail -- no --> UserLoop
    UserLoop -- done --> GenBrew[Orchestrator: GenerateBreweries enriched_cities]

    GenBrew --> BrewLoop{For each EnrichedCity}
    BrewLoop --> BrewGenMode{Generator mode}

    BrewGenMode -- MockGenerator --> BrewMockGen[DeterministicHash and Format]
    BrewGenMode -- LlamaGenerator --> BrewLlamaLoad["PrepareRegionContext, prompt_directory_.Load(BREWERY_GENERATION) → BREWERY_GENERATION.md"]
    BrewLlamaLoad --> BrewLlamaRetry["Infer(...) with kBreweryJsonGrammar → ValidateBreweryJson<br/>retry up to 3 attempts; max_tokens is fixed across retries, not raised on truncation"]
    BrewGenMode -- OpenAIGenerator --> BrewOAILoad["PrepareRegionContext, prompt_directory_.Load(BREWERY_GENERATION) — same prompt file as LlamaGenerator"]
    BrewOAILoad --> BrewOAIRetry["CallChatCompletionsApi(...) with kBreweryJsonSchema → ValidateBreweryJson<br/>retry up to 2 attempts — Structured Outputs already guarantees schema shape"]

    BrewMockGen --> BrewGenDone{Generation successful?}
    BrewLlamaRetry --> BrewGenDone
    BrewOAIRetry --> BrewGenDone

    BrewGenDone -- yes --> BrewPostProc["GetRandomCoordsWithinRange (5km jitter)<br/>generation + jitter run inside one try/catch per record"]
    BrewGenDone -- no --> BrewGenFail["Log warning: generation failed, skipping"] --> BrewLoop

    BrewPostProc --> BrewExport["SqliteExportService.ProcessRecord(GeneratedBrewery)<br/>reuse cached location_id or insert Location and cache its id; insert Brewery (FK location_id)"]
    BrewExport --> BrewExportFail{Exception during insert?}
    BrewExportFail -- yes --> BrewExportWarn["Log warning: SQLite export failed (data loss prevented per-record; pipeline continues)"] --> BrewLoop
    BrewExportFail -- no --> BrewLoop
    BrewLoop -- done --> Finalize["SqliteExportService.Finalize()<br/>commits transaction, closes database connection"]

    Finalize --> CloseLog["main.cc: close log_channel, join log dispatcher thread"]
    CloseLog --> Stop2(["return 0 / stop"])
```