add svgs

pipeline/diagrams/future-possible-activity.puml

@@ -1,5 +1,4 @@
-@startuml
-skinparam style strictuml
+@startuml future_possible_activity
 skinparam defaultFontName "DM Sans"
 skinparam defaultFontSize 13
 skinparam titleFontName "Volkhov"
@@ -8,270 +7,186 @@ skinparam backgroundColor #FAFCF9
 skinparam defaultFontColor #28342A
 skinparam titleFontColor #28342A
 skinparam ArrowColor #628A5B
-skinparam SequenceLifeLineBorderColor #547461
-skinparam SequenceParticipantBorderColor #547461
-skinparam SequenceParticipantBackgroundColor #EAF0E8
-skinparam SequenceBoxBorderColor #547461
+skinparam ActivityBackgroundColor #EAF0E8
+skinparam ActivityBorderColor #547461
+skinparam ActivityDiamondBackgroundColor #DCE8D8
+skinparam ActivityDiamondBorderColor #547461
 skinparam NoteBackgroundColor #EAF0E8
 skinparam NoteBorderColor #547461
-skinparam SequenceDividerBackgroundColor #EAF0E8
-skinparam SequenceDividerBorderColor #547461
 
-title The Biergarten Data Pipeline — Sequence Diagram v4 (Unified Orchestrator)
+title The Biergarten Data Pipeline — Activity Diagram
 
-participant "main.cc" as main #F2F6F0
-participant "Orchestrator" as orch #EAF0E8
-participant "Thread U1\nUserProducer" as u1 #DCE8D8
-participant "Thread U2\nUserExportConsumer" as u2 #E0EAE0
-participant "Thread B1\nEnrichmentProducer" as b1 #DCE8D8
-participant "Thread B2\nBreweryGenerationConsumer" as b2 #E5EDE1
-participant "Thread B3\nBreweryExportConsumer" as b3 #E0EAE0
-participant "Thread R1\nBeerGenerationProducer" as r1 #DCE8D8
-participant "Thread R2\nBeerExportConsumer" as r2 #E0EAE0
-participant "Thread C1\nCheckinGenerationProducer" as c1 #DCE8D8
-participant "Thread C2\nCheckinExportConsumer" as c2 #E0EAE0
-participant "Thread G1\nRatingGenerationProducer" as g1 #DCE8D8
-participant "Thread G2\nRatingExportConsumer" as g2 #E0EAE0
+|Main|
+start
+:ParseArguments(argc, argv);
+if (Invalid args?) then (yes)
+  :spdlog::error;
+  stop
+else (no)
+endif
+:Init CurlGlobalState & LlamaBackendState;
+:Build DI injector;
 
-' ─────────────────────────────────────────────
-' STARTUP
-' ─────────────────────────────────────────────
-main -> main : ParseArguments(argc, argv)
-alt Invalid args
-  main -> main : spdlog::error; stop
-end
+:JsonLoader::LoadLocations("locations.json");
+:JsonLoader::LoadBeerStyles("beer-styles.json");
 
-main -> main : Init CurlGlobalState & LlamaBackendState
-main -> main : Build DI injector
-note right of main
-  All dependencies bound with unique_ptr.
-  LlamaConfig or RestConfig injected
-  instead of ApplicationOptions.
+:EnrichmentService::PreWarmBeerStyleCache(beer_styles);
+note right
+  **NEW**: Beer styles do not need location context.
+  Wikipedia summaries for the entire palette are
+  fetched and cached globally at startup.
 end note
 
-main -> orch : exporter->Initialize()
-note right of orch
+:Initialize SqliteExportService;
+note right
   Opens SQLite connection.
-  Creates schema for all five fixture types
-  in one DDL pass. Begins IMMEDIATE TRANSACTION.
+  Begins a single transaction
+  covering all five fixture types.
+end note
+:BiergartenPipelineOrchestrator::Run();
+
+' ═══════════════════════════════════════════
+' PHASE 0 — USER GENERATION
+' ═══════════════════════════════════════════
+|Orchestrator|
+:RunUserPhase(sampled_locations);
+:Create BoundedChannels\n(user_llm_ch, user_exp_ch);
+
+fork
+  |Orchestrator|
+  :Loop: Send Locations → user_llm_ch;
+  :Close user_llm_ch;
+fork again
+  |LLM Worker|
+  while (user_llm_ch has items?) is (yes)
+    :Receive Location;
+    :GenerateUser(location)\nvia DataGenerator;
+    :Send GeneratedUser → user_exp_ch;
+  endwhile (no)
+  :Close user_exp_ch;
+fork again
+  |SQLite Worker|
+  while (user_exp_ch has items?) is (yes)
+    :Receive GeneratedUser;
+    :ProcessUser(user) → sqlite3_int64;
+    :Append → user_pool_;
+  endwhile (no)
+end fork
+
+|Orchestrator|
+:Join LLM Worker, SQLite Worker;
+
+' ═══════════════════════════════════════════
+' PHASE 1 — BREWERY & BEER GENERATION
+' Combined into a single dependent unit of work.
+' ═══════════════════════════════════════════
+:RunBreweryAndBeerPhase(sampled_locations);
+:Create BoundedChannels\n(loc_ch, llm_ch, exp_ch);
+
+fork
+  |Orchestrator|
+  :Loop: Send Locations → loc_ch;
+  :Close loc_ch;
+fork again
+  |Enrichment Workers (xN)|
+  while (loc_ch has items?) is (yes)
+    :Receive Location;
+    :GetLocationContext(location,\nBreweryContextStrategy);
+    :Send EnrichedCity → llm_ch;
+  endwhile (no)
+  |Orchestrator|
+  :Join Enrichment Workers;
+  :Close llm_ch;
+fork again
+  |LLM Worker|
+  while (llm_ch has items?) is (yes)
+    :Receive EnrichedCity;
+
+    :GenerateBrewery(location, context)\nvia DataGenerator;
+
+    :IBeerSelectionStrategy::SelectStyles(\n  brewery, beer_style_palette_);
+
+    while (For each selected BeerStyle?) is (remaining)
+      :GetStyleContextFromCache(style);
+      note right
+        Guaranteed cache hit from startup.
+      end note
+      :GenerateBeer(brewery, style_context)\nvia DataGenerator;
+      :Attach GeneratedBeer to Brewery bundle;
+    endwhile (done)
+
+    :Send BreweryWithBeers Bundle → exp_ch;
+    note right
+      The next generation of a brewery is
+      entirely dependent on the current
+      brewery and its beers completing.
+    end note
+  endwhile (no)
+  :Close exp_ch;
+fork again
+  |SQLite Worker|
+  while (exp_ch has items?) is (yes)
+    :Receive BreweryWithBeers Bundle;
+    :ProcessBrewery(brewery) → brewery_id;
+    :Append → brewery_pool_;
+
+    while (For each beer in bundle?) is (remaining)
+      :Set beer.brewery_id = brewery_id;
+      :ProcessBeer(beer) → sqlite3_int64;
+      :Append → beer_pool_;
+    endwhile (done)
+  endwhile (no)
+end fork
+
+|Orchestrator|
+:Join LLM Worker, SQLite Worker;
+note right
+  Both brewery_pool_ and beer_pool_
+  are now completely populated.
 end note
 
-main -> orch : JsonLoader::LoadLocations("locations.json")
-main -> orch : ISamplingStrategy::Sample(all_locations)
-main -> orch : BiergartenPipelineOrchestrator::Run()
+' ═══════════════════════════════════════════
+' PHASE 2 — CHECKIN GENERATION
+' Sequential now that Breweries/Beers are done.
+' ═══════════════════════════════════════════
+:RunCheckinPhase();
+:ICheckinDistributionStrategy::\nAssignActivityWeights(user_pool_);
 
-' ─────────────────────────────────────────────
-' PHASE 1 — USERS
-' ─────────────────────────────────────────────
-== Phase 1 — Users (no FK dependencies) ==
+while (For each GeneratedUser in user_pool_?) is (remaining)
+  :CheckinsForUser(user, brewery_pool_.size());
+  while (For each checkin index?) is (remaining)
+    :TimestampFor(user, index);
+    :Select brewery from brewery_pool_;
+    :GenerateCheckin(user, brewery, timestamp)\nvia DataGenerator;
+    :ProcessCheckin(checkin) → sqlite3_int64;
+    :Append → checkin_pool_;
+  endwhile (done)
+endwhile (done)
 
-orch -> u1 : spawn
-orch -> u2 : spawn
+' ═══════════════════════════════════════════
+' PHASE 3 — RATING GENERATION
+' ═══════════════════════════════════════════
+:RunRatingPhase();
 
-loop For each Location
-  u1 -> u1 : generator->GenerateUser(location)
-  u1 -> u2 : user_channel_.Send(GeneratedUser)
-end
-u1 -> u2 : user_channel_.Close()
+while (For each GeneratedCheckin in checkin_pool_?) is (remaining)
+  :Match brewery_id → select beer\nfrom beer_pool_ (same brewery_id);
+  if (Beer exists for brewery?) then (yes)
+    :GenerateRating(user, beer, checkin_id)\nvia DataGenerator;
+    :ProcessRating(rating);
+  else (no)
+    :Skip — brewery has no beers;
+  endif
+endwhile (done)
 
-loop user_channel_.Receive()
-  u2 -> u2 : exporter->ProcessUser(user) : sqlite3_int64
-  note right of u2
-    Returns committed row ID.
-    Stored on GeneratedUser.user_id.
-  end note
-  u2 -> orch : Append to user_pool_
-end
-
-orch -> orch : join(U1, U2)
-note right of orch
-  ── BARRIER 1 ──
-  user_pool_ is now fully committed.
-  Phase 2 begins.
-end note
-
-' ─────────────────────────────────────────────
-' PHASE 2 — BREWERIES
-' ─────────────────────────────────────────────
-== Phase 2 — Breweries (depends on locations only) ==
-
-orch -> b1 : spawn
-orch -> b2 : spawn
-orch -> b3 : spawn
-
-loop For each Location
-  b1 -> b1 : BreweryContextStrategy::QueriesFor(location)
-  b1 -> b1 : WikipediaService::GetLocationContext(location,\nbrewery_context_strategy_)
-  alt failure
-    b1 -> b1 : LocationContext{ Absent }
-  else truncated
-    b1 -> b1 : LocationContext{ Partial }
-  else success
-    b1 -> b1 : LocationContext{ Full }
-  end
-  b1 -> b2 : enrichment_channel_.Send(EnrichedCity)
-  note right of b1
-    Blocks if channel full.
-    Back-pressure against GPU consumer.
-  end note
-end
-b1 -> b2 : enrichment_channel_.Close()
-
-loop enrichment_channel_.Receive()
-  alt context.completeness == Absent
-    b2 -> b2 : spdlog::warn — proceeding with minimal prompt
-  end
-  b2 -> b2 : generator->GenerateBrewery(location, context)
-  b2 -> b3 : brewery_channel_.Send(GeneratedBrewery)
-end
-b2 -> b3 : brewery_channel_.Close()
-
-loop brewery_channel_.Receive()
-  b3 -> b3 : exporter->ProcessBrewery(brewery) : sqlite3_int64
-  note right of b3
-    Row ID stored on GeneratedBrewery.brewery_id.
-    No brewery_cache_ needed — orchestrator
-    threads the ID forward directly.
-  end note
-  b3 -> orch : Append to brewery_pool_
-end
-
-orch -> orch : join(B1, B2, B3)
-note right of orch
-  ── BARRIER 2 ──
-  brewery_pool_ is now fully committed
-  with live brewery_id values.
-  Phase 3 begins.
-end note
-
-' ─────────────────────────────────────────────
-' PHASE 3 — BEERS
-' ─────────────────────────────────────────────
-== Phase 3 — Beers (depends on brewery_pool_) ==
-
-orch -> r1 : spawn
-orch -> r2 : spawn
-
-loop For each GeneratedBrewery in brewery_pool_
-  r1 -> r1 : BeerContextStrategy::QueriesFor(location)
-  r1 -> r1 : WikipediaService::GetLocationContext(location,\nbeer_context_strategy_)
-  r1 -> r1 : generator->GenerateBeer(brewery.brewery_id,\nlocation, context)
-  r1 -> r2 : beer_channel_.Send(GeneratedBeer)
-end
-r1 -> r2 : beer_channel_.Close()
-
-loop beer_channel_.Receive()
-  r2 -> r2 : exporter->ProcessBeer(beer) : sqlite3_int64
-  note right of r2
-    Row ID stored on GeneratedBeer.beer_id.
-  end note
-  r2 -> orch : Append to beer_pool_
-end
-
-orch -> orch : join(R1, R2)
-note right of orch
-  ── BARRIER 3 ──
-  beer_pool_ is fully committed.
-  All three upstream pools ready.
-end note
-
-' ─────────────────────────────────────────────
-' CHECKIN WEIGHT ASSIGNMENT
-' ─────────────────────────────────────────────
-== Checkin Weight Assignment ==
-
-orch -> orch : ICheckinDistributionStrategy::\nAssignActivityWeights(user_pool_)
-note right of orch
-  J-curve weights written onto
-  GeneratedUser.user.activity_weight.
-  Small cohort gets high weight;
-  long tail gets low weight.
-  Requires the full pool — this is why
-  users were committed first.
-end note
-
-' ─────────────────────────────────────────────
-' PHASE 4 — CHECKINS
-' ─────────────────────────────────────────────
-== Phase 4 — Check-ins (depends on user_pool_ + brewery_pool_) ==
-
-orch -> c1 : spawn
-orch -> c2 : spawn
-
-loop For each GeneratedUser in user_pool_
-  c1 -> c1 : strategy->CheckinsForUser(user,\nbrewery_pool_.size())
-  loop For each checkin index
-    c1 -> c1 : strategy->TimestampFor(user, index)
-    note right of c1
-      Bursty weekend / evening
-      distribution applied here.
-    end note
-    c1 -> c1 : Select brewery from brewery_pool_\n(weighted random by activity_weight)
-    c1 -> c1 : generator->GenerateCheckin(user, brewery, timestamp)
-    c1 -> c2 : checkin_channel_.Send(GeneratedCheckin)
-  end
-end
-c1 -> c2 : checkin_channel_.Close()
-
-loop checkin_channel_.Receive()
-  c2 -> c2 : exporter->ProcessCheckin(checkin) : sqlite3_int64
-  note right of c2
-    Row ID stored on GeneratedCheckin.checkin_id.
-  end note
-  c2 -> orch : Append to checkin_pool_
-end
-
-orch -> orch : join(C1, C2)
-note right of orch
-  ── BARRIER 4 ──
-  checkin_pool_ is fully committed.
-  All FK dependencies for ratings satisfied.
-  Phase 5 begins.
-end note
-
-' ─────────────────────────────────────────────
-' PHASE 5 — RATINGS
-' ─────────────────────────────────────────────
-== Phase 5 — Ratings (depends on user_pool_ + beer_pool_ + checkin_pool_) ==
-
-orch -> g1 : spawn
-orch -> g2 : spawn
-
-loop For each GeneratedCheckin in checkin_pool_
-  g1 -> g1 : Resolve GeneratedUser from user_pool_\n(match user_id)
-  g1 -> g1 : Resolve GeneratedBeer from beer_pool_\n(match brewery_id, select one)
-  alt Beer found for this brewery
-    g1 -> g1 : generator->GenerateRating(user, beer,\ncheckin.checkin_id)
-    note right of g1
-      Strong positive skew applied
-      inside GenerateRating.
-    end note
-    g1 -> g2 : rating_channel_.Send(GeneratedRating)
-  else No beer found
-    g1 -> g1 : spdlog::warn — no beer for brewery,\nskipping rating
-  end
-end
-g1 -> g2 : rating_channel_.Close()
-
-loop rating_channel_.Receive()
-  g2 -> g2 : exporter->ProcessRating(rating)
-end
-
-orch -> orch : join(G1, G2)
-
-' ─────────────────────────────────────────────
+' ═══════════════════════════════════════════
 ' TEARDOWN
-' ─────────────────────────────────────────────
-== Teardown ==
-
-orch -> main : return
-main -> main : exporter->Finalize()
-note right of main
-  Single COMMIT covers all five fixture types:
-  users, breweries, beers, checkins, ratings.
-  All-or-nothing consistency per run.
+' ═══════════════════════════════════════════
+|Main|
+:Finalize SqliteExportService;
+note right
+  COMMIT covers all five fixture types.
 end note
-main -> main : spdlog::info "Pipeline complete in X ms"
-main -> main : return 0
+:spdlog::info "Pipeline complete in X ms";
+stop
 
 @enduml

pipeline/diagrams/future-possible-architecture.puml

@@ -1,4 +1,4 @@
-@startuml
+@startuml future_possible_architecture
 skinparam style strictuml
 skinparam defaultFontName "DM Sans"
 skinparam defaultFontSize 14
@@ -29,7 +29,7 @@ skinparam package {
   FontColor #28342A
 }
 
-title The Biergarten Data Pipeline — Architecture v4 (Unified Orchestrator)
+title The Biergarten Data Pipeline — Architecture
 
 ' ─────────────────────────────────────────────
 ' DOMAIN: VALUE OBJECTS
@@ -63,6 +63,24 @@ package "Domain: Value Objects & Contracts" {
     + context : LocationContext
   }
 
+  class BeerStyle {
+    + name : std::string
+    + description : std::string
+    + min_abv : float
+    + max_abv : float
+    + min_ibu : int
+    + max_ibu : int
+  }
+  note right of BeerStyle
+    Loaded once at startup from
+    beer-styles.json via JsonLoader.
+    Passed as std::span<const BeerStyle>
+    to IBeerSelectionStrategy.
+    Generator receives the selected
+    style as a parameter — it never
+    reads the palette directly.
+  end note
+
   class BreweryResult {
     + name_en : std::string
     + description_en : std::string
@@ -125,6 +143,11 @@ package "Domain: Value Objects & Contracts" {
     + user : UserResult
     + generated_at : std::string
   }
+  note right of GeneratedUser
+    user_id populated after SQLite
+    insert. Live FK carried in pool
+    for checkin and rating references.
+  end note
 
   class GeneratedCheckin {
     + checkin_id : sqlite3_int64
@@ -173,27 +196,47 @@ package "Domain Policy" {
     + Sample(locations : const std::vector<Location>&) : std::vector<Location>
   }
 
+  interface IBeerSelectionStrategy <<interface>> {
+    + SelectStyles(brewery : const GeneratedBrewery&,\n  palette : std::span<const BeerStyle>) : std::vector<BeerStyle>
+  }
+  note right of IBeerSelectionStrategy
+    Decides how many beers a brewery
+    gets and which styles are selected.
+    Count distribution and style
+    deduplication logic live here,
+    not in the orchestrator or generator.
+  end note
+
+  class RandomBeerSelectionStrategy {
+    - rng_ : std::mt19937
+    - min_beers_ : size_t
+    - max_beers_ : size_t
+    + SelectStyles(brewery : const GeneratedBrewery&,\n  palette : std::span<const BeerStyle>) : std::vector<BeerStyle>
+  }
+  note right of RandomBeerSelectionStrategy
+    Draws a random count in [min, max].
+    Samples without replacement from
+    palette to avoid duplicate styles
+    per brewery.
+  end note
+
   interface ICheckinDistributionStrategy <<interface>> {
     + AssignActivityWeights(users : std::vector<GeneratedUser>&) : void
-    + CheckinsForUser(user : const GeneratedUser&, brewery_count : size_t) : size_t
-    + TimestampFor(user : const GeneratedUser&, index : size_t) : std::string
+    + CheckinsForUser(user : const GeneratedUser&,\n  brewery_count : size_t) : size_t
+    + TimestampFor(user : const GeneratedUser&,\n  index : size_t) : std::string
   }
   note right of ICheckinDistributionStrategy
-    Injected into the orchestrator.
     Owns all statistical policy:
     J-curve weight assignment,
     bursty weekend timestamps,
     per-user checkin volume.
-    No mediator required to hold this —
-    the orchestrator calls it directly
-    before the checkin phase opens.
   end note
 
   class JCurveCheckinStrategy {
     - rng_ : std::mt19937
     + AssignActivityWeights(users : std::vector<GeneratedUser>&) : void
-    + CheckinsForUser(user : const GeneratedUser&, brewery_count : size_t) : size_t
-    + TimestampFor(user : const GeneratedUser&, index : size_t) : std::string
+    + CheckinsForUser(user : const GeneratedUser&,\n  brewery_count : size_t) : size_t
+    + TimestampFor(user : const GeneratedUser&,\n  index : size_t) : std::string
   }
 
 }
@@ -204,15 +247,21 @@ package "Domain Policy" {
 package "Infrastructure: Enrichment" {
 
   interface IEnrichmentService <<interface>> {
-    + GetLocationContext(loc : const Location&, strategy : const IContextStrategy&) : LocationContext
+    + GetLocationContext(loc : const Location&,\n  strategy : const IContextStrategy&) : LocationContext
   }
 
   class WikipediaService {
     - client_ : std::unique_ptr<WebClient>
     - extract_cache_ : std::unordered_map<std::string, std::string>
-    + GetLocationContext(loc : const Location&, strategy : const IContextStrategy&) : LocationContext
+    + GetLocationContext(loc : const Location&,\n  strategy : const IContextStrategy&) : LocationContext
     - FetchExtract(query : std::string_view) : std::string
   }
+  note right of WikipediaService
+    extract_cache_ keyed by query string.
+    Beer pass gets near-100% cache hits
+    since locations were already fetched
+    during the brewery pass.
+  end note
 
   interface WebClient <<interface>> {
     + Get(url : const std::string&) : std::string
@@ -232,12 +281,18 @@ package "Infrastructure: Enrichment" {
 package "Infrastructure: Generation" {
 
   interface DataGenerator <<interface>> {
-    + GenerateBrewery(location : const Location&, context : const LocationContext&) : BreweryResult
-    + GenerateBeer(brewery_id : sqlite3_int64, location : const Location&, context : const LocationContext&) : BeerResult
+    + GenerateBrewery(location : const Location&,\n  context : const LocationContext&) : BreweryResult
+    + GenerateBeer(brewery_id : sqlite3_int64,\n  location : const Location&,\n  context : const LocationContext&,\n  style : const BeerStyle&) : BeerResult
     + GenerateUser(location : const Location&) : UserResult
-    + GenerateCheckin(user : const GeneratedUser&, brewery : const GeneratedBrewery&, timestamp : const std::string&) : CheckinResult
-    + GenerateRating(user : const GeneratedUser&, beer : const GeneratedBeer&, checkin_id : sqlite3_int64) : RatingResult
+    + GenerateCheckin(user : const GeneratedUser&,\n  brewery : const GeneratedBrewery&,\n  timestamp : const std::string&) : CheckinResult
+    + GenerateRating(user : const GeneratedUser&,\n  beer : const GeneratedBeer&,\n  checkin_id : sqlite3_int64) : RatingResult
   }
+  note right of DataGenerator
+    GenerateBeer receives BeerStyle
+    as a parameter. Style selection
+    and count decisions live in
+    IBeerSelectionStrategy, not here.
+  end note
 
   class MockGenerator {
     + GenerateBrewery(...) : BreweryResult
@@ -260,26 +315,12 @@ package "Infrastructure: Generation" {
     + GenerateCheckin(...) : CheckinResult
     + GenerateRating(...) : RatingResult
     - Load(config : const LlamaConfig&) : void
-    - Infer(system_prompt, user_prompt, max_tokens, grammar) : std::string
+    - Infer(system_prompt, user_prompt,\n  max_tokens, grammar) : std::string
     - ValidateModelArchitecture() : void
   }
 
-  class RestGenerator {
-    - config_ : RestConfig
-    + GenerateBrewery(...) : BreweryResult
-    + GenerateBeer(...) : BeerResult
-    + GenerateUser(...) : UserResult
-    + GenerateCheckin(...) : CheckinResult
-    + GenerateRating(...) : RatingResult
-  }
-  note right of RestGenerator
-    Future REST-backed implementation.
-    Slots in at the DI root with zero
-    changes to orchestration logic.
-  end note
-
   interface IPromptFormatter <<interface>> {
-    + Format(system_prompt : std::string_view, user_prompt : std::string_view) : std::string
+    + Format(system_prompt : std::string_view,\n  user_prompt : std::string_view) : std::string
     + ExpectedArchitecture() : std::string_view
   }
 
@@ -297,12 +338,6 @@ package "Infrastructure: Generation" {
     + seed : int
   }
 
-  class RestConfig {
-    + endpoint : std::string
-    + api_key : std::string
-    + timeout : std::chrono::milliseconds
-  }
-
 }
 
 ' ─────────────────────────────────────────────
@@ -322,12 +357,12 @@ package "Infrastructure: Pipeline Channel" {
     + Close() : void
   }
   note right of BoundedChannel
-    Used within each phase to
-    decouple production from export.
-    Phase boundaries are explicit
-    sequential barriers in the
-    orchestrator's Run() method —
-    not channel-mediated.
+    Used for user, brewery, and
+    checkin/rating phases.
+    Beer phase uses a simple
+    sequential loop — enrichment
+    is all cache hits, no fan-out
+    needed.
   end note
 
 }
@@ -346,13 +381,6 @@ package "Infrastructure: Export" {
     + ProcessRating(rating : const GeneratedRating&) : void
     + Finalize() : void
   }
-  note right of IExportService
-    Process* methods return
-    sqlite3_int64 row IDs.
-    Orchestrator uses these to
-    populate FK fields on all
-    downstream fixture types.
-  end note
 
   class SqliteExportService {
     - date_time_provider_ : std::unique_ptr<IDateTimeProvider>
@@ -365,6 +393,7 @@ package "Infrastructure: Export" {
     - insert_rating_stmt_ : SqliteStatementHandle
     - transaction_open_ : bool
     - location_cache_ : std::unordered_map<std::string, sqlite3_int64>
+    - brewery_cache_ : std::unordered_map<std::string, sqlite3_int64>
     + Initialize() : void
     + ProcessBrewery(brewery : const GeneratedBrewery&) : sqlite3_int64
     + ProcessBeer(beer : const GeneratedBeer&) : sqlite3_int64
@@ -378,11 +407,11 @@ package "Infrastructure: Export" {
     - FinalizeStatements() : void
   }
   note right of SqliteExportService
-    brewery_cache_ removed — row IDs
-    are now carried on GeneratedBrewery
-    and GeneratedBeer value objects
-    and threaded through by the
-    orchestrator directly.
+    brewery_cache_ restored.
+    Keyed by location string for
+    location deduplication, and
+    by brewery identity for beer
+    FK resolution without re-querying.
   end note
 
   interface IDateTimeProvider <<interface>> {
@@ -407,7 +436,9 @@ package "Orchestration" {
     - brewery_context_strategy_ : std::unique_ptr<IContextStrategy>
     - beer_context_strategy_ : std::unique_ptr<IContextStrategy>
     - sampling_strategy_ : std::unique_ptr<ISamplingStrategy>
+    - beer_selection_strategy_ : std::unique_ptr<IBeerSelectionStrategy>
     - checkin_strategy_ : std::unique_ptr<ICheckinDistributionStrategy>
+    - beer_style_palette_ : std::vector<BeerStyle>
     --
     - user_pool_ : std::vector<GeneratedUser>
     - brewery_pool_ : std::vector<GeneratedBrewery>
@@ -422,29 +453,25 @@ package "Orchestration" {
     - RunRatingPhase() : void
   }
   note right of BiergartenPipelineOrchestrator
-    Single component owns all
-    sequencing. Run() reads as a
-    linear narrative:
-      1. RunUserPhase
-      2. RunBreweryPhase
-      3. RunBeerPhase
-      4. checkin_strategy_->AssignActivityWeights
-      5. RunCheckinPhase
-      6. RunRatingPhase
-    The checkin gate is an explicit
-    sequential barrier between steps
-    3 and 5 — not a hidden internal
-    trigger in a separate object.
-    Pools are members: each phase
-    appends to them and the next
-    phase reads from them directly.
-    No mediator. No shared_ptr.
-    Ownership is unambiguous.
+    beer_style_palette_ loaded once
+    at startup from beer-styles.json.
+    Passed as std::span<const BeerStyle>
+    to IBeerSelectionStrategy per brewery.
+    RunBeerPhase() is a sequential loop —
+    no channels, no fan-out. Enrichment
+    is cache hits; LLM is the only cost.
   end note
 
   class JsonLoader {
     + {static} LoadLocations(filepath : const std::filesystem::path&) : std::vector<Location>
+    + {static} LoadBeerStyles(filepath : const std::filesystem::path&) : std::vector<BeerStyle>
   }
+  note right of JsonLoader
+    LoadBeerStyles() added.
+    Reads beer-styles.json once
+    at startup into the palette
+    held by the orchestrator.
+  end note
 
 }
 
@@ -458,12 +485,14 @@ BiergartenPipelineOrchestrator *-- DataGenerator : owns
 BiergartenPipelineOrchestrator *-- IExportService : owns
 BiergartenPipelineOrchestrator *-- ICheckinDistributionStrategy : owns
 BiergartenPipelineOrchestrator *-- ISamplingStrategy : owns
+BiergartenPipelineOrchestrator *-- IBeerSelectionStrategy : owns
 BiergartenPipelineOrchestrator ..> JsonLoader : uses
 
 ' Policy implementations
 IContextStrategy <|.. BreweryContextStrategy : implements
 IContextStrategy <|.. BeerContextStrategy : implements
 ISamplingStrategy <|.. UniformSamplingStrategy : implements
+IBeerSelectionStrategy <|.. RandomBeerSelectionStrategy : implements
 ICheckinDistributionStrategy <|.. JCurveCheckinStrategy : implements
 
 ' Enrichment
@@ -475,10 +504,8 @@ WebClient <|.. CURLWebClient : implements
 ' Generation
 DataGenerator <|.. MockGenerator : implements
 DataGenerator <|.. LlamaGenerator : implements
-DataGenerator <|.. RestGenerator : implements
 LlamaGenerator *-- IPromptFormatter : owns
 LlamaGenerator ..> LlamaConfig : constructed with
-RestGenerator ..> RestConfig : constructed with
 IPromptFormatter <|.. Gemma4JinjaPromptFormatter : implements
 
 ' Export
@@ -486,7 +513,7 @@ IExportService <|.. SqliteExportService : implements
 SqliteExportService *-- IDateTimeProvider : owns
 IDateTimeProvider <|.. SystemDateTimeProvider : implements
 
-' Data flow
+' Domain containment
 EnrichedCity *-- Location : contains
 EnrichedCity *-- LocationContext : contains
 GeneratedBrewery *-- Location : contains