From 18bb1564342ca66f08a0bbf939aae4406dfa2357 Mon Sep 17 00:00:00 2001
From: devRaGonSa <rgonsal93@gmail.com>
Date: Tue, 2 Jun 2026 11:21:15 +0200
Subject: [PATCH] feat: add database maintenance and update home copy

---
 ...SK-database-maintenance-cleanup-command.md |   0
 .../TASK-database-maintenance-docs.md         |   0
 .../TASK-db-retention-audit.md                |   0
 ...K-final-maintenance-and-copy-validation.md |   0
 ...historical-runner-maintenance-scheduler.md |   0
 ...home-remove-vietnam-copy-except-trailer.md |   0
 .../review/TASK-db-retention-audit-report.md  | 518 ++++++++++++++
 ...-maintenance-and-copy-validation-report.md | 122 ++++
 backend/app/config.py                         |  70 ++
 backend/app/database_maintenance.py           | 638 ++++++++++++++++++
 backend/app/historical_runner.py              |  92 +++
 backend/tests/test_database_maintenance.py    | 448 ++++++++++++
 .../test_historical_runner_maintenance.py     | 153 +++++
 docs/database-maintenance.md                  | 307 +++++++++
 frontend/index.html                           |   8 +-
 15 files changed, 2352 insertions(+), 4 deletions(-)
 rename ai/tasks/{pending => done}/TASK-database-maintenance-cleanup-command.md (100%)
 rename ai/tasks/{pending => done}/TASK-database-maintenance-docs.md (100%)
 rename ai/tasks/{pending => done}/TASK-db-retention-audit.md (100%)
 rename ai/tasks/{pending => done}/TASK-final-maintenance-and-copy-validation.md (100%)
 rename ai/tasks/{pending => done}/TASK-historical-runner-maintenance-scheduler.md (100%)
 rename ai/tasks/{pending => done}/TASK-home-remove-vietnam-copy-except-trailer.md (100%)
 create mode 100644 ai/tasks/review/TASK-db-retention-audit-report.md
 create mode 100644 ai/tasks/review/TASK-final-maintenance-and-copy-validation-report.md
 create mode 100644 backend/app/database_maintenance.py
 create mode 100644 backend/tests/test_database_maintenance.py
 create mode 100644 backend/tests/test_historical_runner_maintenance.py
 create mode 100644 docs/database-maintenance.md

diff --git a/ai/tasks/pending/TASK-database-maintenance-cleanup-command.md b/ai/tasks/done/TASK-database-maintenance-cleanup-command.md
similarity index 100%
rename from ai/tasks/pending/TASK-database-maintenance-cleanup-command.md
rename to ai/tasks/done/TASK-database-maintenance-cleanup-command.md
diff --git a/ai/tasks/pending/TASK-database-maintenance-docs.md b/ai/tasks/done/TASK-database-maintenance-docs.md
similarity index 100%
rename from ai/tasks/pending/TASK-database-maintenance-docs.md
rename to ai/tasks/done/TASK-database-maintenance-docs.md
diff --git a/ai/tasks/pending/TASK-db-retention-audit.md b/ai/tasks/done/TASK-db-retention-audit.md
similarity index 100%
rename from ai/tasks/pending/TASK-db-retention-audit.md
rename to ai/tasks/done/TASK-db-retention-audit.md
diff --git a/ai/tasks/pending/TASK-final-maintenance-and-copy-validation.md b/ai/tasks/done/TASK-final-maintenance-and-copy-validation.md
similarity index 100%
rename from ai/tasks/pending/TASK-final-maintenance-and-copy-validation.md
rename to ai/tasks/done/TASK-final-maintenance-and-copy-validation.md
diff --git a/ai/tasks/pending/TASK-historical-runner-maintenance-scheduler.md b/ai/tasks/done/TASK-historical-runner-maintenance-scheduler.md
similarity index 100%
rename from ai/tasks/pending/TASK-historical-runner-maintenance-scheduler.md
rename to ai/tasks/done/TASK-historical-runner-maintenance-scheduler.md
diff --git a/ai/tasks/pending/TASK-home-remove-vietnam-copy-except-trailer.md b/ai/tasks/done/TASK-home-remove-vietnam-copy-except-trailer.md
similarity index 100%
rename from ai/tasks/pending/TASK-home-remove-vietnam-copy-except-trailer.md
rename to ai/tasks/done/TASK-home-remove-vietnam-copy-except-trailer.md
diff --git a/ai/tasks/review/TASK-db-retention-audit-report.md b/ai/tasks/review/TASK-db-retention-audit-report.md
new file mode 100644
index 0000000..c6f8a67
--- /dev/null
+++ b/ai/tasks/review/TASK-db-retention-audit-report.md
@@ -0,0 +1,518 @@
+---
+id: TASK-db-retention-audit-report
+source_task: TASK-db-retention-audit
+status: review
+type: report
+---
+
+# Database Retention Audit
+
+## Scope
+
+This report audits the current storage model used by HLL Vietnam for:
+
+- live server snapshots;
+- public-scoreboard historical matches and player stats;
+- RCON AdminLog events and materialized matches;
+- displayed historical snapshots;
+- player event raw ledger;
+- paused Elo/MMR storage surfaces.
+
+Confirmed facts below are based on the current repository code, not on assumptions.
+
+## Files Inspected
+
+- `AGENTS.md`
+- `ai/repo-context.md`
+- `ai/architecture-index.md`
+- `backend/app/config.py`
+- `backend/app/historical_runner.py`
+- `backend/app/historical_storage.py`
+- `backend/app/historical_snapshot_storage.py`
+- `backend/app/player_event_aggregates.py`
+- `backend/app/player_event_storage.py`
+- `backend/app/postgres_display_storage.py`
+- `backend/app/postgres_rcon_storage.py`
+- `backend/app/rcon_admin_log_materialization.py`
+- `backend/app/rcon_admin_log_storage.py`
+- `backend/app/rcon_historical_backfill.py`
+- `backend/app/rcon_historical_leaderboards.py`
+- `backend/app/rcon_historical_read_model.py`
+- `backend/app/rcon_historical_worker.py`
+- `backend/app/writer_lock.py`
+
+## Confirmed Storage Areas
+
+### Live and displayed data
+
+- `server_snapshots`
+  - Present in SQLite live storage and PostgreSQL displayed storage.
+  - Used for live/current server status history.
+- `displayed_historical_snapshots`
+  - PostgreSQL primary key is `(server_key, snapshot_type, metric, snapshot_window)`.
+  - Persistence uses upsert semantics, so normal operation is already bounded by logical identity.
+
+### Public-scoreboard historical fallback
+
+- `historical_servers`
+- `historical_maps`
+- `historical_matches`
+- `historical_players`
+- `historical_player_match_stats`
+- `historical_ingestion_runs`
+- `historical_backfill_progress`
+
+These remain part of the displayed historical fallback/read path.
+
+### RCON historical primary path
+
+- `rcon_historical_targets`
+- `rcon_historical_capture_runs`
+- `rcon_historical_samples`
+- `rcon_historical_checkpoints`
+- `rcon_historical_competitive_windows`
+- `rcon_admin_log_events`
+- `rcon_player_profile_snapshots`
+- `rcon_materialized_matches`
+- `rcon_match_player_stats`
+- `rcon_scoreboard_match_candidates`
+
+### Player-event ledger
+
+- `player_event_raw_ledger`
+- `player_event_ingestion_runs`
+- `player_event_backfill_progress`
+
+### Elo/MMR paused surfaces
+
+Confirmed in backend documentation and runtime payloads as preserved but paused.
+This audit did not find cleanup code or explicit retention logic for Elo/MMR tables in the task’s read-first files, so they should stay out of cleanup scope.
+
+## Snapshot Boundedness Findings
+
+### `displayed_historical_snapshots`
+
+Confirmed bounded in PostgreSQL by primary key plus upsert:
+
+- same logical snapshot identity overwrites previous payload;
+- normal scheduled refresh does not append unlimited rows;
+- routine cleanup is not required for healthy records.
+
+Safe policy:
+
+- do not bulk-delete this table in the first maintenance command;
+- only report suspicious legacy/corrupt rows if found later.
+
+### File-based snapshots
+
+Confirmed bounded by deterministic file paths under `backend/data/snapshots/<server>/...`.
+Normal writes replace files in place.
+
+Safe policy:
+
+- do not add file cleanup in the first database-maintenance command.
+
+## Retention Windows Confirmed In Runtime Code
+
+Two different historical paths currently exist, and they do not use identical monthly rules.
+
+### Public-scoreboard fallback (`historical_*`)
+
+Confirmed in `backend/app/historical_storage.py`:
+
+- weekly fallback can use previous week when current-week sample is insufficient;
+- monthly fallback uses previous month only when current month has zero closed matches;
+- `is_early_month` is computed with `current_time.day <= 3`, but selection still depends on zero current-month matches.
+
+### RCON materialized leaderboards (`rcon_materialized_matches`)
+
+Confirmed in `backend/app/rcon_historical_leaderboards.py`:
+
+- monthly leaderboard selection uses previous month through day 7 inclusive;
+- weekly selection falls back to previous week when current week lacks the configured minimum closed matches.
+
+## Cleanup Policy Recommendation
+
+Because the new maintenance task explicitly requires:
+
+- latest 100 materialized matches;
+- current month;
+- previous month during first 7 days of a new month;
+- current week;
+- previous week when weekly fallback may need it;
+
+the cleanup command should implement the stricter policy directly instead of copying the older `historical_storage.py` month fallback behavior.
+
+## Safe-To-Delete Areas
+
+### 1. `server_snapshots`
+
+Safe to delete old rows by `captured_at` age once outside retention.
+
+Why:
+
+- live UI only needs recent history;
+- no dependent child tables were found;
+- latest snapshot queries use recent rows only.
+
+Recommended initial retention:
+
+- keep last 14 days by default.
+
+### 2. Non-critical `rcon_admin_log_events`
+
+Safe to delete by event age after retention.
+
+Confirmed critical event types:
+
+- `kill`
+- `match_start`
+- `match_end`
+
+Everything else should be treated as non-critical for first-pass cleanup:
+
+- `chat`
+- `message`
+- `connected`
+- `disconnected`
+- `team_switch`
+- `unknown`
+- `kick`
+- `ban`
+- any other non-materialization event type.
+
+Recommended initial retention:
+
+- non-critical: 30 days.
+
+### 3. Critical `rcon_admin_log_events`
+
+Conditionally safe to delete only when both are true:
+
+- older than the critical retention threshold;
+- not needed by protected materialized matches/windows.
+
+Recommended initial retention:
+
+- critical: 90 days.
+
+### 4. Old `rcon_materialized_matches`
+
+Conditionally safe to delete when all of the following are true:
+
+- not in latest N closed matches;
+- not in current month;
+- not in previous month during the first 7 days of a new month;
+- not in current week;
+- not in previous week when weekly fallback may still be needed.
+
+### 5. Child rows in `rcon_match_player_stats`
+
+Safe to delete only as a dependent delete for non-protected materialized matches.
+
+Delete these before deleting parent matches because no cascade constraint was found.
+
+## Areas That Must Be Protected
+
+### RCON materialized read path
+
+Protect:
+
+- latest 100 closed `rcon_materialized_matches`;
+- all `rcon_match_player_stats` for protected matches;
+- `kill`, `match_start`, `match_end` events that still support protected matches or currently relevant windows.
+
+### Live/current match surfaces
+
+Protect recent/current RCON data:
+
+- recent `rcon_admin_log_events` still used by current-match kill feed and current-match player stats;
+- recent `server_snapshots`.
+
+Reason:
+
+- current-match pages read recent AdminLog windows directly;
+- premature deletion would break kill feed or partial player stats.
+
+### Displayed snapshots
+
+Protect:
+
+- `displayed_historical_snapshots` as a bounded read cache;
+- file-based snapshots under `backend/data/snapshots/`.
+
+### Public-scoreboard `historical_*`
+
+Do not delete in the first cleanup command.
+
+Reason:
+
+- these tables still back visible historical endpoints and fallback behavior;
+- the task’s cleanup requirements are expressed around materialized matches and AdminLog, not around legacy scoreboard imports;
+- deleting this domain now would widen scope materially.
+
+### Player-event ledger
+
+Do not delete in the first cleanup command.
+
+Reason:
+
+- it feeds visible player-event snapshots and monthly MVP V2 support;
+- no task requirement currently defines safe retention windows for this ledger;
+- deleting it without product retention rules is risky.
+
+### Elo/MMR paused data
+
+Do not delete.
+
+Reason:
+
+- task explicitly warns against unrelated Elo/MMR deletions;
+- runtime is paused, but persisted surfaces are intentionally preserved.
+
+## Foreign Keys, Cascades, and Orphan Risk
+
+### Confirmed foreign keys
+
+Public-scoreboard historical tables have foreign keys:
+
+- `historical_matches.historical_server_id -> historical_servers.id`
+- `historical_matches.historical_map_id -> historical_maps.id`
+- `historical_player_match_stats.historical_match_id -> historical_matches.id`
+- `historical_player_match_stats.historical_player_id -> historical_players.id`
+
+RCON/postgres tables also define references for some capture-oriented tables.
+
+### Missing cascade behavior relevant to cleanup
+
+No `ON DELETE CASCADE` behavior was found for:
+
+- `rcon_materialized_matches`
+- `rcon_match_player_stats`
+
+The materialization code explicitly deletes/rebuilds child player rows itself, which is another signal that cleanup must manage delete order manually.
+
+### Required delete order
+
+For materialized RCON cleanup:
+
+1. identify protected match keys;
+2. identify candidate old matches outside protected windows;
+3. delete `rcon_match_player_stats` for candidate match keys;
+4. delete candidate rows from `rcon_materialized_matches`;
+5. delete eligible `rcon_admin_log_events`;
+6. delete eligible `server_snapshots`.
+
+This order avoids orphaned player-stat rows.
+
+## Risks and Edge Cases
+
+### Timestamp parsing risk
+
+Some match rows may have:
+
+- missing `started_at`;
+- missing `ended_at`;
+- only server-time fields;
+- malformed or non-ISO timestamps.
+
+Safe rule:
+
+- if a materialized match timestamp cannot be interpreted safely, protect it and skip deletion.
+
+### Cross-path monthly rule mismatch
+
+There is a confirmed mismatch today:
+
+- public-scoreboard fallback code does not strictly use previous month through day 7;
+- RCON leaderboard code does.
+
+Cleanup should follow the task requirement, not the older fallback implementation.
+
+### Current-match dependency on recent AdminLog
+
+Current match pages query recent AdminLog rows directly, not only materialized matches.
+Deleting recent critical events too aggressively could break live/current-match UI even if historical leaderboards still work.
+
+### Optional table presence
+
+Some environments may not have every migrated or optional table yet.
+Maintenance code should treat absent tables as controlled skips, not hard failures.
+
+### SQLite versus PostgreSQL
+
+The repo still supports SQLite fallback when `HLL_BACKEND_DATABASE_URL` is unset.
+Maintenance code should either:
+
+- support SQLite safely for the targeted tables; or
+- emit a clear unsupported message.
+
+Given the codebase still keeps local SQLite testability, SQLite support is preferable for the cleanup command and its tests.
+
+## Recommended Cleanup Scope For First Implementation
+
+Implement only:
+
+- `server_snapshots`;
+- `rcon_admin_log_events`;
+- `rcon_materialized_matches`;
+- `rcon_match_player_stats`.
+
+Explicitly skip in first implementation:
+
+- `displayed_historical_snapshots`;
+- `historical_*`;
+- `player_event_raw_ledger` and related worker tables;
+- Elo/MMR tables;
+- `rcon_historical_samples`, `rcon_historical_competitive_windows`, `rcon_player_profile_snapshots`, `rcon_scoreboard_match_candidates`.
+
+Reason:
+
+- keeps scope aligned to the task;
+- avoids deleting sources still needed for visible history fallback or paused domains;
+- stays within the repository’s change-budget expectations.
+
+## Implementation Plan For `app.database_maintenance`
+
+1. Add maintenance config accessors for all retention values and batch size.
+2. Add a CLI with `cleanup --dry-run` and `cleanup --apply`.
+3. Build one plan phase that calculates:
+   - protected materialized match keys;
+   - candidate match keys;
+   - candidate admin-log rows by criticality and age;
+   - candidate server snapshots by age.
+4. Log the plan in structured JSON before any deletes.
+5. On apply:
+   - acquire the shared backend writer lock;
+   - delete `rcon_match_player_stats` in batches for candidate match keys;
+   - delete `rcon_materialized_matches` in batches;
+   - delete `rcon_admin_log_events` in batches;
+   - delete `server_snapshots` in batches;
+   - optionally run `VACUUM ANALYZE` only when explicitly requested.
+6. Treat missing optional tables as logged skips.
+7. Keep dry-run as default behavior.
+
+## Recommended SQL Audit Commands
+
+### Table-size audit
+
+```sql
+select
+  schemaname,
+  relname as table_name,
+  pg_size_pretty(pg_total_relation_size(relid)) as total_size,
+  pg_size_pretty(pg_relation_size(relid)) as table_size,
+  pg_size_pretty(pg_total_relation_size(relid) - pg_relation_size(relid)) as indexes_size,
+  n_live_tup as estimated_rows,
+  n_dead_tup as estimated_dead_rows
+from pg_stat_user_tables
+order by pg_total_relation_size(relid) desc;
+```
+
+### AdminLog by type and age
+
+```sql
+select
+  event_type,
+  count(*) as row_count,
+  min(event_timestamp) as first_event_timestamp,
+  max(event_timestamp) as last_event_timestamp,
+  min(server_time) as first_server_time,
+  max(server_time) as last_server_time
+from rcon_admin_log_events
+group by event_type
+order by row_count desc, event_type asc;
+```
+
+### Materialized matches by server and closure time
+
+```sql
+select
+  target_key,
+  source_basis,
+  count(*) as matches,
+  min(coalesce(ended_at, started_at)) as first_closed_at,
+  max(coalesce(ended_at, started_at)) as last_closed_at
+from rcon_materialized_matches
+group by target_key, source_basis
+order by target_key asc, source_basis asc;
+```
+
+### Materialized player stats coverage
+
+```sql
+select
+  target_key,
+  count(*) as player_stat_rows,
+  count(distinct match_key) as match_count
+from rcon_match_player_stats
+group by target_key
+order by target_key asc;
+```
+
+### Live server snapshots by age
+
+```sql
+select
+  server_id,
+  min(captured_at) as first_captured_at,
+  max(captured_at) as last_captured_at,
+  count(*) as snapshot_rows
+from server_snapshots
+group by server_id
+order by last_captured_at desc;
+```
+
+### Displayed snapshot coverage
+
+```sql
+select
+  snapshot_type,
+  metric,
+  snapshot_window,
+  count(*) as snapshot_rows,
+  min(generated_at) as first_generated_at,
+  max(generated_at) as last_generated_at
+from displayed_historical_snapshots
+group by snapshot_type, metric, snapshot_window
+order by snapshot_type asc, metric asc, snapshot_window asc;
+```
+
+### Public-scoreboard fallback historical matches
+
+```sql
+select
+  hs.slug as server_slug,
+  count(*) as matches,
+  min(coalesce(hm.ended_at, hm.started_at, hm.created_at_source)) as first_match_at,
+  max(coalesce(hm.ended_at, hm.started_at, hm.created_at_source)) as last_match_at
+from historical_matches hm
+join historical_servers hs on hs.id = hm.historical_server_id
+group by hs.slug
+order by hs.slug asc;
+```
+
+### Player-event ledger age
+
+```sql
+select
+  event_type,
+  count(*) as row_count,
+  min(occurred_at) as first_occurred_at,
+  max(occurred_at) as last_occurred_at
+from player_event_raw_ledger
+group by event_type
+order by row_count desc, event_type asc;
+```
+
+## Summary
+
+The first cleanup command can be implemented safely if it stays narrow:
+
+- clean `server_snapshots`;
+- clean aged `rcon_admin_log_events`;
+- clean non-protected `rcon_materialized_matches`;
+- delete `rcon_match_player_stats` first for those matches;
+- skip bounded snapshot caches and broader fallback domains for now.
+
+The main correctness risk is not delete syntax. It is retention-window selection and preserving current/live dependencies. The cleanup command should therefore be plan-first, dry-run-first, and lock-protected.
diff --git a/ai/tasks/review/TASK-final-maintenance-and-copy-validation-report.md b/ai/tasks/review/TASK-final-maintenance-and-copy-validation-report.md
new file mode 100644
index 0000000..ea8de3c
--- /dev/null
+++ b/ai/tasks/review/TASK-final-maintenance-and-copy-validation-report.md
@@ -0,0 +1,122 @@
+---
+id: TASK-final-maintenance-and-copy-validation-report
+title: Final maintenance and copy validation report
+status: review
+type: report
+team: PM
+supporting_teams: [Backend Senior, Frontend Senior, Arquitecto de Base de Datos]
+roadmap_item: validation
+priority: high
+---
+
+# TASK-final-maintenance-and-copy-validation-report
+
+## Scope
+
+Validation-only report for:
+
+- database maintenance cleanup command
+- historical-runner maintenance scheduler
+- database maintenance operations documentation
+- home copy update removing `Vietnam` outside the trailer block
+
+No runtime code was changed during this validation task.
+
+## Validation Commands Run
+
+```powershell
+python -m compileall backend/app
+PYTHONPATH=backend python -m unittest backend.tests.test_database_maintenance
+PYTHONPATH=backend python -m unittest backend.tests.test_historical_runner_maintenance
+PYTHONPATH=backend python -m unittest discover -s backend/tests -p "*historical*"
+node --check frontend/assets/js/main.js
+node --check frontend/assets/js/historico.js
+node --check frontend/assets/js/partida-actual.js
+git diff --check
+Invoke-RestMethod "http://127.0.0.1:8000/api/historical/snapshots/recent-matches?server=all-servers&limit=10" | ConvertTo-Json -Depth 10
+Invoke-RestMethod "http://127.0.0.1:8000/api/historical/snapshots/leaderboard?server=all-servers&timeframe=weekly&metric=kills&limit=10" | ConvertTo-Json -Depth 10
+Invoke-RestMethod "http://127.0.0.1:8000/api/historical/snapshots/leaderboard?server=all-servers&timeframe=monthly&metric=kills&limit=10" | ConvertTo-Json -Depth 10
+$env:PYTHONPATH='backend'; python -m app.database_maintenance cleanup --dry-run
+Invoke-WebRequest "http://127.0.0.1:8080/?v=maintenance-copy-validation"
+Invoke-WebRequest "http://127.0.0.1:8080/historico.html?v=maintenance-copy-validation"
+Invoke-WebRequest "http://127.0.0.1:8080/partida-actual.html?server=comunidad-hispana-01&v=maintenance-copy-validation"
+```
+
+## Result
+
+- `compileall`: passed
+- `test_database_maintenance`: passed
+- `test_historical_runner_maintenance`: passed
+- `*historical*` discovery suite: passed
+- frontend JavaScript syntax checks: passed
+- `git diff --check`: passed
+
+## Docker Status
+
+Docker smoke validation could not run because the local Docker daemon was unavailable.
+
+Observed error:
+
+`failed to connect to the docker API at npipe:////./pipe/dockerDesktopLinuxEngine`
+
+This prevented:
+
+- `docker compose up -d --build backend frontend postgres`
+- `docker compose --profile advanced up -d --build historical-runner rcon-historical-worker`
+- `docker compose --profile advanced ps`
+- `docker compose logs --tail=200 historical-runner`
+- `docker compose exec backend python -m app.database_maintenance cleanup --dry-run`
+
+## Local Smoke Checks
+
+Fallback local processes were started instead:
+
+- backend: `python -m app.main`
+- frontend: `python -m http.server 8080`
+
+Endpoint summary:
+
+- recent matches endpoint returned `200` with snapshot data for `all-servers`
+- weekly leaderboard endpoint returned `200` with a valid empty weekly result for the current dataset
+- monthly leaderboard endpoint returned `200` with valid leaderboard rows
+
+Maintenance dry-run summary:
+
+- backend mode: `sqlite`
+- protected materialized matches: `22`
+- candidate materialized matches: `0`
+- candidate player-stat rows: `0`
+- candidate admin-log rows: `0`
+- candidate server snapshots: `298`
+- skipped tables: none
+
+Frontend HTTP checks:
+
+- `/`: `200`
+- `/historico.html`: `200`
+- `/partida-actual.html?server=comunidad-hispana-01`: `200`
+
+## Copy Validation
+
+Confirmed in the diff for `frontend/index.html`:
+
+- meta description changed from `HLL Vietnam` to `HLL`
+- page title changed from `Comunidad Hispana - HLL Vietnam` to `Comunidad Hispana - HLL`
+- logo alt text changed from `HLL Vietnam` to `HLL`
+- hero accent changed from `HLL Vietnam` to `HLL`
+- trailer text `Primer vistazo a HLL Vietnam` was preserved
+- iframe title `Trailer HLL Vietnam` was preserved
+
+No browser-based visual QA was performed in this task.
+
+## Unrelated Notes
+
+Python test runs emitted existing `ResourceWarning` messages about unclosed SQLite connections from current initialization helpers. The test suites still passed, and this task did not change that behavior.
+
+An unintended local runtime artifact at `backend/backend/data/hll_vietnam_dev.sqlite3` was created during smoke setup and removed before final scope review.
+
+## Branch/Push Status
+
+- branch push: not performed
+- commit: not performed
+
diff --git a/backend/app/config.py b/backend/app/config.py
index fa8f447..49aefcb 100644
--- a/backend/app/config.py
+++ b/backend/app/config.py
@@ -38,6 +38,13 @@ DEFAULT_RCON_HISTORICAL_CAPTURE_RETRY_DELAY_SECONDS = 15
 DEFAULT_RCON_BACKFILL_CHUNK_HOURS = 6
 DEFAULT_RCON_BACKFILL_SLEEP_SECONDS = 1.0
 DEFAULT_RCON_BACKFILL_MAX_DAYS_BACK = 45
+DEFAULT_RECENT_MATCHES_KEEP = 100
+DEFAULT_ADMIN_LOG_NONCRITICAL_RETENTION_DAYS = 30
+DEFAULT_ADMIN_LOG_CRITICAL_RETENTION_DAYS = 90
+DEFAULT_SERVER_SNAPSHOT_RETENTION_DAYS = 14
+DEFAULT_DB_MAINTENANCE_BATCH_SIZE = 5000
+DEFAULT_DB_MAINTENANCE_ENABLED = False
+DEFAULT_DB_MAINTENANCE_INTERVAL_SECONDS = 43200
 DEFAULT_SQLITE_WRITER_TIMEOUT_SECONDS = 30.0
 DEFAULT_SQLITE_BUSY_TIMEOUT_MS = 30000
 DEFAULT_WRITER_LOCK_TIMEOUT_SECONDS = 120.0
@@ -514,6 +521,69 @@ def get_rcon_backfill_max_days_back() -> int:
     )
 
 
+def get_recent_matches_keep() -> int:
+    """Return how many recent closed materialized matches maintenance must protect."""
+    return _read_int_env(
+        "HLL_RECENT_MATCHES_KEEP",
+        str(DEFAULT_RECENT_MATCHES_KEEP),
+        minimum=1,
+    )
+
+
+def get_admin_log_noncritical_retention_days() -> int:
+    """Return retention days for non-critical AdminLog events."""
+    return _read_int_env(
+        "HLL_ADMIN_LOG_NONCRITICAL_RETENTION_DAYS",
+        str(DEFAULT_ADMIN_LOG_NONCRITICAL_RETENTION_DAYS),
+        minimum=1,
+    )
+
+
+def get_admin_log_critical_retention_days() -> int:
+    """Return retention days for critical AdminLog events."""
+    return _read_int_env(
+        "HLL_ADMIN_LOG_CRITICAL_RETENTION_DAYS",
+        str(DEFAULT_ADMIN_LOG_CRITICAL_RETENTION_DAYS),
+        minimum=1,
+    )
+
+
+def get_server_snapshot_retention_days() -> int:
+    """Return retention days for live server snapshots."""
+    return _read_int_env(
+        "HLL_SERVER_SNAPSHOT_RETENTION_DAYS",
+        str(DEFAULT_SERVER_SNAPSHOT_RETENTION_DAYS),
+        minimum=1,
+    )
+
+
+def get_db_maintenance_batch_size() -> int:
+    """Return the delete batch size used by database maintenance."""
+    return _read_int_env(
+        "HLL_DB_MAINTENANCE_BATCH_SIZE",
+        str(DEFAULT_DB_MAINTENANCE_BATCH_SIZE),
+        minimum=1,
+    )
+
+
+def get_db_maintenance_enabled() -> bool:
+    """Return whether scheduled database maintenance is enabled."""
+    normalized = os.getenv(
+        "HLL_DB_MAINTENANCE_ENABLED",
+        "true" if DEFAULT_DB_MAINTENANCE_ENABLED else "false",
+    ).strip().lower()
+    return normalized in {"1", "true", "yes", "on"}
+
+
+def get_db_maintenance_interval_seconds() -> int:
+    """Return the scheduled database maintenance interval in seconds."""
+    return _read_int_env(
+        "HLL_DB_MAINTENANCE_INTERVAL_SECONDS",
+        str(DEFAULT_DB_MAINTENANCE_INTERVAL_SECONDS),
+        minimum=1,
+    )
+
+
 def get_a2s_targets_payload() -> str | None:
     """Return the optional JSON payload that overrides local A2S targets."""
     raw_payload = os.getenv(DEFAULT_A2S_TARGETS_ENV_VAR)
diff --git a/backend/app/database_maintenance.py b/backend/app/database_maintenance.py
new file mode 100644
index 0000000..6cce252
--- /dev/null
+++ b/backend/app/database_maintenance.py
@@ -0,0 +1,638 @@
+"""Application-level database maintenance for bounded historical storage."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import sqlite3
+from contextlib import closing
+from dataclasses import dataclass
+from datetime import datetime, timedelta, timezone
+from pathlib import Path
+from typing import Any, Iterable, Sequence
+
+from .config import (
+    get_admin_log_critical_retention_days,
+    get_admin_log_noncritical_retention_days,
+    get_database_url,
+    get_db_maintenance_batch_size,
+    get_historical_weekly_fallback_min_matches,
+    get_recent_matches_keep,
+    get_server_snapshot_retention_days,
+)
+from .rcon_admin_log_materialization import MATCH_RESULT_SOURCE
+from .sqlite_utils import connect_sqlite_writer
+from .writer_lock import backend_writer_lock, build_writer_lock_holder
+
+CRITICAL_ADMIN_LOG_EVENT_TYPES = frozenset({"kill", "match_start", "match_end"})
+
+
+@dataclass(frozen=True, slots=True)
+class MaintenanceOptions:
+    apply: bool
+    recent_matches_keep: int
+    admin_log_noncritical_retention_days: int
+    admin_log_critical_retention_days: int
+    server_snapshot_retention_days: int
+    batch_size: int
+    vacuum_analyze: bool
+    now: datetime
+
+
+def run_database_maintenance_cleanup(
+    *,
+    apply: bool = False,
+    recent_matches_keep: int | None = None,
+    admin_log_noncritical_retention_days: int | None = None,
+    admin_log_critical_retention_days: int | None = None,
+    server_snapshot_retention_days: int | None = None,
+    batch_size: int | None = None,
+    vacuum_analyze: bool = False,
+    now: str | datetime | None = None,
+    db_path: Path | None = None,
+) -> dict[str, object]:
+    """Plan or apply safe bounded cleanup for supported storage tables."""
+    options = MaintenanceOptions(
+        apply=apply,
+        recent_matches_keep=recent_matches_keep or get_recent_matches_keep(),
+        admin_log_noncritical_retention_days=(
+            admin_log_noncritical_retention_days or get_admin_log_noncritical_retention_days()
+        ),
+        admin_log_critical_retention_days=(
+            admin_log_critical_retention_days or get_admin_log_critical_retention_days()
+        ),
+        server_snapshot_retention_days=(
+            server_snapshot_retention_days or get_server_snapshot_retention_days()
+        ),
+        batch_size=batch_size or get_db_maintenance_batch_size(),
+        vacuum_analyze=vacuum_analyze,
+        now=_resolve_now(now),
+    )
+    _emit_json_log(
+        {
+            "event": "database-maintenance-started",
+            "mode": "apply" if options.apply else "dry-run",
+            "database_backend": _database_backend_name(db_path=db_path),
+            "database_url_configured": bool(get_database_url()) and db_path is None,
+            "db_path": str(db_path) if db_path is not None else None,
+            "recent_matches_keep": options.recent_matches_keep,
+            "admin_log_noncritical_retention_days": options.admin_log_noncritical_retention_days,
+            "admin_log_critical_retention_days": options.admin_log_critical_retention_days,
+            "server_snapshot_retention_days": options.server_snapshot_retention_days,
+            "batch_size": options.batch_size,
+            "vacuum_analyze": options.vacuum_analyze,
+            "now": _to_iso(options.now),
+        }
+    )
+
+    try:
+        if options.apply:
+            with backend_writer_lock(
+                holder=build_writer_lock_holder("app.database_maintenance cleanup"),
+                storage_path=db_path,
+            ):
+                payload = _run_cleanup(options=options, db_path=db_path)
+        else:
+            payload = _run_cleanup(options=options, db_path=db_path)
+        _emit_json_log(
+            {
+                "event": "database-maintenance-completed",
+                **payload,
+            }
+        )
+        return payload
+    except Exception as exc:  # noqa: BLE001 - CLI reports structured diagnostics
+        error_payload = {
+            "status": "error",
+            "mode": "apply" if options.apply else "dry-run",
+            "error_type": type(exc).__name__,
+            "error": str(exc),
+        }
+        _emit_json_log({"event": "database-maintenance-error", **error_payload})
+        return error_payload
+
+
+def _run_cleanup(*, options: MaintenanceOptions, db_path: Path | None) -> dict[str, object]:
+    with _connect_maintenance(db_path=db_path) as connection:
+        existing_tables = _existing_table_names(connection)
+        plan = _build_cleanup_plan(connection, existing_tables=existing_tables, options=options)
+        _emit_json_log(
+            {
+                "event": "database-maintenance-plan",
+                **plan["summary"],
+            }
+        )
+
+        deleted_counts = {
+            "rcon_match_player_stats": 0,
+            "rcon_materialized_matches": 0,
+            "rcon_admin_log_events": 0,
+            "server_snapshots": 0,
+        }
+        if options.apply:
+            deleted_counts["rcon_match_player_stats"] = _delete_match_player_stats(
+                connection,
+                matches=plan["candidate_matches"],
+                batch_size=options.batch_size,
+            )
+            deleted_counts["rcon_materialized_matches"] = _delete_ids_in_batches(
+                connection,
+                table_name="rcon_materialized_matches",
+                ids=[int(row["id"]) for row in plan["candidate_matches"]],
+                batch_size=options.batch_size,
+            )
+            deleted_counts["rcon_admin_log_events"] = _delete_ids_in_batches(
+                connection,
+                table_name="rcon_admin_log_events",
+                ids=plan["candidate_admin_log_ids"],
+                batch_size=options.batch_size,
+            )
+            deleted_counts["server_snapshots"] = _delete_ids_in_batches(
+                connection,
+                table_name="server_snapshots",
+                ids=plan["candidate_server_snapshot_ids"],
+                batch_size=options.batch_size,
+            )
+            if options.vacuum_analyze:
+                _run_vacuum_analyze(connection)
+
+    return {
+        "status": "ok",
+        "mode": "apply" if options.apply else "dry-run",
+        "deleted_counts": deleted_counts,
+        "plan": plan["summary"],
+    }
+
+
+def _build_cleanup_plan(
+    connection: sqlite3.Connection | Any,
+    *,
+    existing_tables: set[str],
+    options: MaintenanceOptions,
+) -> dict[str, object]:
+    candidate_server_snapshot_ids: list[int] = []
+    candidate_admin_log_ids: list[int] = []
+    candidate_matches: list[dict[str, object]] = []
+    protected_match_keys: list[str] = []
+    skipped_tables: list[str] = []
+
+    if "server_snapshots" not in existing_tables:
+        skipped_tables.append("server_snapshots")
+        _emit_skip("server_snapshots", "table-missing")
+    else:
+        cutoff = options.now - timedelta(days=options.server_snapshot_retention_days)
+        for row in connection.execute(
+            "SELECT id, captured_at FROM server_snapshots ORDER BY id ASC"
+        ).fetchall():
+            captured_at = _parse_datetime(row["captured_at"])
+            if captured_at is None:
+                continue
+            if captured_at < cutoff:
+                candidate_server_snapshot_ids.append(int(row["id"]))
+
+    protected_ranges: dict[str, list[tuple[int, int]]] = {}
+    if "rcon_materialized_matches" not in existing_tables:
+        skipped_tables.append("rcon_materialized_matches")
+        _emit_skip("rcon_materialized_matches", "table-missing")
+    else:
+        (
+            candidate_matches,
+            protected_matches,
+            protected_ranges,
+            protection_summary,
+        ) = _plan_materialized_match_cleanup(connection, options=options)
+        protected_match_keys = [str(row["match_key"]) for row in protected_matches]
+    if "rcon_match_player_stats" not in existing_tables:
+        skipped_tables.append("rcon_match_player_stats")
+        _emit_skip("rcon_match_player_stats", "table-missing")
+
+    if "rcon_admin_log_events" not in existing_tables:
+        skipped_tables.append("rcon_admin_log_events")
+        _emit_skip("rcon_admin_log_events", "table-missing")
+    else:
+        candidate_admin_log_ids = _plan_admin_log_cleanup(
+            connection,
+            options=options,
+            protected_ranges=protected_ranges,
+        )
+
+    candidate_player_stat_rows = 0
+    if candidate_matches and "rcon_match_player_stats" in existing_tables:
+        candidate_player_stat_rows = _count_candidate_player_stats(connection, candidate_matches)
+
+    summary = {
+        "status": "ok",
+        "protected_match_count": len(protected_match_keys),
+        "candidate_match_count": len(candidate_matches),
+        "candidate_match_player_stat_count": candidate_player_stat_rows,
+        "candidate_admin_log_event_count": len(candidate_admin_log_ids),
+        "candidate_server_snapshot_count": len(candidate_server_snapshot_ids),
+        "skipped_tables": skipped_tables,
+        "protected_match_keys_preview": protected_match_keys[:10],
+    }
+    if "protection_summary" in locals():
+        summary["protection_summary"] = protection_summary
+
+    return {
+        "candidate_server_snapshot_ids": candidate_server_snapshot_ids,
+        "candidate_admin_log_ids": candidate_admin_log_ids,
+        "candidate_matches": candidate_matches,
+        "summary": summary,
+    }
+
+
+def _plan_materialized_match_cleanup(
+    connection: sqlite3.Connection | Any,
+    *,
+    options: MaintenanceOptions,
+) -> tuple[list[dict[str, object]], list[dict[str, object]], dict[str, list[tuple[int, int]]], dict[str, object]]:
+    rows = [
+        dict(row)
+        for row in connection.execute(
+            """
+            SELECT id, target_key, match_key, started_at, ended_at,
+                   started_server_time, ended_server_time, source_basis
+            FROM rcon_materialized_matches
+            WHERE source_basis = ?
+            """,
+            (MATCH_RESULT_SOURCE,),
+        ).fetchall()
+    ]
+    closed_rows: list[dict[str, object]] = []
+    protected_rows: list[dict[str, object]] = []
+    current_month_start = options.now.replace(day=1, hour=0, minute=0, second=0, microsecond=0)
+    previous_month_start = (current_month_start - timedelta(days=1)).replace(day=1)
+    current_week_start = (options.now - timedelta(days=options.now.weekday())).replace(
+        hour=0,
+        minute=0,
+        second=0,
+        microsecond=0,
+    )
+    previous_week_start = current_week_start - timedelta(days=7)
+
+    for row in rows:
+        closed_at = _parse_datetime(row.get("ended_at") or row.get("started_at"))
+        if closed_at is None:
+            row["_protect_reason"] = "unparseable-closed-at"
+            protected_rows.append(row)
+            continue
+        row["_closed_at"] = closed_at
+        closed_rows.append(row)
+
+    closed_rows.sort(
+        key=lambda row: (
+            row["_closed_at"],
+            _coerce_int(row.get("ended_server_time")) or _coerce_int(row.get("started_server_time")) or 0,
+            _coerce_int(row.get("id")) or 0,
+        ),
+        reverse=True,
+    )
+    latest_ids = {int(row["id"]) for row in closed_rows[: options.recent_matches_keep]}
+    current_week_count = sum(
+        1 for row in closed_rows if current_week_start <= row["_closed_at"] < options.now
+    )
+    previous_week_count = sum(
+        1 for row in closed_rows if previous_week_start <= row["_closed_at"] < current_week_start
+    )
+    protect_previous_week = (
+        current_week_count < get_historical_weekly_fallback_min_matches()
+        and previous_week_count > 0
+    )
+    protect_previous_month = options.now.day <= 7
+
+    candidate_rows: list[dict[str, object]] = []
+    protected_ranges: dict[str, list[tuple[int, int]]] = {}
+    for row in closed_rows:
+        closed_at = row["_closed_at"]
+        should_protect = False
+        if int(row["id"]) in latest_ids:
+            should_protect = True
+        elif closed_at >= current_month_start:
+            should_protect = True
+        elif protect_previous_month and previous_month_start <= closed_at < current_month_start:
+            should_protect = True
+        elif closed_at >= current_week_start:
+            should_protect = True
+        elif protect_previous_week and previous_week_start <= closed_at < current_week_start:
+            should_protect = True
+
+        if should_protect:
+            protected_rows.append(row)
+            lower = _coerce_int(row.get("started_server_time"))
+            upper = _coerce_int(row.get("ended_server_time"))
+            if lower is not None and upper is not None:
+                protected_ranges.setdefault(str(row["target_key"]), []).append((lower, upper))
+        else:
+            candidate_rows.append(row)
+
+    return (
+        candidate_rows,
+        protected_rows,
+        protected_ranges,
+        {
+            "recent_matches_keep": options.recent_matches_keep,
+            "current_week_closed_matches": current_week_count,
+            "previous_week_closed_matches": previous_week_count,
+            "protect_previous_week": protect_previous_week,
+            "protect_previous_month": protect_previous_month,
+            "current_week_start": _to_iso(current_week_start),
+            "previous_week_start": _to_iso(previous_week_start),
+            "current_month_start": _to_iso(current_month_start),
+            "previous_month_start": _to_iso(previous_month_start),
+        },
+    )
+
+
+def _plan_admin_log_cleanup(
+    connection: sqlite3.Connection | Any,
+    *,
+    options: MaintenanceOptions,
+    protected_ranges: dict[str, list[tuple[int, int]]],
+) -> list[int]:
+    noncritical_cutoff = options.now - timedelta(days=options.admin_log_noncritical_retention_days)
+    critical_cutoff = options.now - timedelta(days=options.admin_log_critical_retention_days)
+    candidate_ids: list[int] = []
+    rows = connection.execute(
+        """
+        SELECT id, target_key, event_type, event_timestamp, server_time
+        FROM rcon_admin_log_events
+        ORDER BY id ASC
+        """
+    ).fetchall()
+    for row in rows:
+        event_type = str(row["event_type"] or "").strip()
+        event_time = _parse_datetime(row["event_timestamp"])
+        if event_time is None:
+            continue
+        if event_type in CRITICAL_ADMIN_LOG_EVENT_TYPES:
+            if event_time >= critical_cutoff:
+                continue
+            server_time = _coerce_int(row["server_time"])
+            if server_time is None:
+                continue
+            if _server_time_is_protected(
+                target_key=str(row["target_key"] or ""),
+                server_time=server_time,
+                protected_ranges=protected_ranges,
+            ):
+                continue
+            candidate_ids.append(int(row["id"]))
+            continue
+        if event_time < noncritical_cutoff:
+            candidate_ids.append(int(row["id"]))
+    return candidate_ids
+
+
+def _count_candidate_player_stats(
+    connection: sqlite3.Connection | Any,
+    matches: Sequence[dict[str, object]],
+) -> int:
+    count = 0
+    for batch in _chunked(list(matches), 250):
+        clause, params = _match_pair_clause(batch)
+        row = connection.execute(
+            f"SELECT COUNT(*) AS count FROM rcon_match_player_stats WHERE {clause}",
+            params,
+        ).fetchone()
+        count += int(row["count"] or 0)
+    return count
+
+
+def _delete_match_player_stats(
+    connection: sqlite3.Connection | Any,
+    *,
+    matches: Sequence[dict[str, object]],
+    batch_size: int,
+) -> int:
+    deleted = 0
+    for batch in _chunked(list(matches), max(1, min(batch_size, 250))):
+        clause, params = _match_pair_clause(batch)
+        deleted_in_batch = int(
+            connection.execute(
+                f"DELETE FROM rcon_match_player_stats WHERE {clause}",
+                params,
+            ).rowcount
+            or 0
+        )
+        _commit(connection)
+        deleted += deleted_in_batch
+        _emit_json_log(
+            {
+                "event": "database-maintenance-delete-batch",
+                "table": "rcon_match_player_stats",
+                "deleted_rows": deleted_in_batch,
+                "batch_size": len(batch),
+            }
+        )
+    return deleted
+
+
+def _delete_ids_in_batches(
+    connection: sqlite3.Connection | Any,
+    *,
+    table_name: str,
+    ids: Sequence[int],
+    batch_size: int,
+) -> int:
+    deleted = 0
+    for batch in _chunked(list(ids), batch_size):
+        placeholders = ",".join("?" for _ in batch)
+        deleted_in_batch = int(
+            connection.execute(
+                f"DELETE FROM {table_name} WHERE id IN ({placeholders})",
+                batch,
+            ).rowcount
+            or 0
+        )
+        _commit(connection)
+        deleted += deleted_in_batch
+        _emit_json_log(
+            {
+                "event": "database-maintenance-delete-batch",
+                "table": table_name,
+                "deleted_rows": deleted_in_batch,
+                "batch_size": len(batch),
+            }
+        )
+    return deleted
+
+
+def _run_vacuum_analyze(connection: sqlite3.Connection | Any) -> None:
+    raw_connection = _raw_connection(connection)
+    if isinstance(raw_connection, sqlite3.Connection):
+        raw_connection.execute("VACUUM")
+        raw_connection.execute("ANALYZE")
+        raw_connection.commit()
+        return
+    raw_connection.commit()
+    raw_connection.autocommit = True
+    try:
+        raw_connection.execute("VACUUM ANALYZE")
+    finally:
+        raw_connection.autocommit = False
+
+
+def _match_pair_clause(matches: Sequence[dict[str, object]]) -> tuple[str, list[object]]:
+    clauses: list[str] = []
+    params: list[object] = []
+    for row in matches:
+        clauses.append("(target_key = ? AND match_key = ?)")
+        params.extend([row["target_key"], row["match_key"]])
+    return " OR ".join(clauses), params
+
+
+def _existing_table_names(connection: sqlite3.Connection | Any) -> set[str]:
+    raw_connection = _raw_connection(connection)
+    if isinstance(raw_connection, sqlite3.Connection):
+        rows = connection.execute(
+            "SELECT name FROM sqlite_master WHERE type = 'table'"
+        ).fetchall()
+        return {str(row["name"]) for row in rows}
+    rows = raw_connection.execute(
+        """
+        SELECT table_name
+        FROM information_schema.tables
+        WHERE table_schema = 'public'
+        """
+    ).fetchall()
+    return {str(row["table_name"]) for row in rows}
+
+
+def _emit_skip(table_name: str, reason: str) -> None:
+    _emit_json_log(
+        {
+            "event": "database-maintenance-table-skipped",
+            "table": table_name,
+            "reason": reason,
+        }
+    )
+
+
+def _server_time_is_protected(
+    *,
+    target_key: str,
+    server_time: int,
+    protected_ranges: dict[str, list[tuple[int, int]]],
+) -> bool:
+    for lower, upper in protected_ranges.get(target_key, []):
+        if lower <= server_time <= upper:
+            return True
+    return False
+
+
+def _connect_maintenance(*, db_path: Path | None):
+    if get_database_url() and db_path is None:
+        from .postgres_rcon_storage import connect_postgres_compat
+
+        return connect_postgres_compat()
+    resolved_path = db_path or Path.cwd() / "backend" / "data" / "hll_vietnam_dev.sqlite3"
+    resolved_path.parent.mkdir(parents=True, exist_ok=True)
+    return closing(connect_sqlite_writer(resolved_path))
+
+
+def _commit(connection: sqlite3.Connection | Any) -> None:
+    _raw_connection(connection).commit()
+
+
+def _raw_connection(connection: sqlite3.Connection | Any) -> sqlite3.Connection | Any:
+    return connection.connection if hasattr(connection, "connection") else connection
+
+
+def _database_backend_name(*, db_path: Path | None) -> str:
+    return "postgres" if get_database_url() and db_path is None else "sqlite"
+
+
+def _resolve_now(value: str | datetime | None) -> datetime:
+    if value is None:
+        return datetime.now(timezone.utc)
+    if isinstance(value, datetime):
+        return value.astimezone(timezone.utc) if value.tzinfo else value.replace(tzinfo=timezone.utc)
+    parsed = _parse_datetime(value)
+    if parsed is None:
+        raise ValueError("--now must be an ISO 8601 timestamp or date.")
+    return parsed
+
+
+def _parse_datetime(value: object) -> datetime | None:
+    text = str(value or "").strip()
+    if not text:
+        return None
+    if len(text) == 10:
+        text = f"{text}T00:00:00+00:00"
+    try:
+        parsed = datetime.fromisoformat(text.replace("Z", "+00:00"))
+    except ValueError:
+        return None
+    return parsed.astimezone(timezone.utc) if parsed.tzinfo else parsed.replace(tzinfo=timezone.utc)
+
+
+def _to_iso(value: datetime) -> str:
+    return value.astimezone(timezone.utc).isoformat().replace("+00:00", "Z")
+
+
+def _coerce_int(value: object) -> int | None:
+    try:
+        return None if value is None else int(value)
+    except (TypeError, ValueError):
+        return None
+
+
+def _chunked(values: Sequence[Any], size: int) -> Iterable[list[Any]]:
+    for index in range(0, len(values), size):
+        yield list(values[index : index + size])
+
+
+def _emit_json_log(payload: dict[str, object]) -> None:
+    print(json.dumps(payload, ensure_ascii=True, default=str), flush=True)
+
+
+def build_arg_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser(description="Database maintenance for HLL Vietnam.")
+    subparsers = parser.add_subparsers(dest="command", required=True)
+    cleanup_parser = subparsers.add_parser("cleanup")
+    cleanup_parser.add_argument("--dry-run", action="store_true")
+    cleanup_parser.add_argument("--apply", action="store_true")
+    cleanup_parser.add_argument("--recent-matches-keep", type=int, default=get_recent_matches_keep())
+    cleanup_parser.add_argument(
+        "--admin-log-noncritical-retention-days",
+        type=int,
+        default=get_admin_log_noncritical_retention_days(),
+    )
+    cleanup_parser.add_argument(
+        "--admin-log-critical-retention-days",
+        type=int,
+        default=get_admin_log_critical_retention_days(),
+    )
+    cleanup_parser.add_argument(
+        "--server-snapshot-retention-days",
+        type=int,
+        default=get_server_snapshot_retention_days(),
+    )
+    cleanup_parser.add_argument("--batch-size", type=int, default=get_db_maintenance_batch_size())
+    cleanup_parser.add_argument("--vacuum-analyze", action="store_true")
+    cleanup_parser.add_argument("--now", default=None)
+    return parser
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = build_arg_parser()
+    args = parser.parse_args(list(argv) if argv is not None else None)
+    if args.command != "cleanup":
+        raise ValueError("Unsupported command.")
+    if args.apply and args.dry_run:
+        raise ValueError("--apply and --dry-run are mutually exclusive.")
+    payload = run_database_maintenance_cleanup(
+        apply=bool(args.apply),
+        recent_matches_keep=args.recent_matches_keep,
+        admin_log_noncritical_retention_days=args.admin_log_noncritical_retention_days,
+        admin_log_critical_retention_days=args.admin_log_critical_retention_days,
+        server_snapshot_retention_days=args.server_snapshot_retention_days,
+        batch_size=args.batch_size,
+        vacuum_analyze=bool(args.vacuum_analyze),
+        now=args.now,
+    )
+    return 0 if payload.get("status") == "ok" else 1
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/backend/app/historical_runner.py b/backend/app/historical_runner.py
index e14442a..10edced 100644
--- a/backend/app/historical_runner.py
+++ b/backend/app/historical_runner.py
@@ -10,6 +10,9 @@ from datetime import datetime, timezone
 from typing import Any
 
 from .config import (
+    DEFAULT_DB_MAINTENANCE_INTERVAL_SECONDS,
+    get_db_maintenance_enabled,
+    get_db_maintenance_interval_seconds,
     get_historical_full_snapshot_every_runs,
     get_historical_elo_mmr_min_new_samples,
     get_historical_elo_mmr_rebuild_interval_minutes,
@@ -18,6 +21,7 @@ from .config import (
     get_historical_refresh_retry_delay_seconds,
     get_historical_data_source_kind,
 )
+from .database_maintenance import run_database_maintenance_cleanup
 from .elo_mmr_engine import rebuild_elo_mmr_models
 from .elo_mmr_storage import get_latest_elo_mmr_generated_at
 from .historical_ingestion import run_incremental_refresh
@@ -34,6 +38,7 @@ DEFAULT_HISTORICAL_SERVER_SCOPE = (
     "comunidad-hispana-01",
     "comunidad-hispana-02",
 )
+_LAST_DATABASE_MAINTENANCE_RUN_AT: datetime | None = None
 
 
 def run_periodic_historical_refresh(
@@ -171,6 +176,7 @@ def _run_refresh_with_retries(
                                 rcon_capture_result=rcon_capture_result
                             ),
                         }
+                maintenance_result = _maybe_run_database_maintenance()
             return {
                 "status": "ok",
                 "attempts_used": attempt,
@@ -181,6 +187,7 @@ def _run_refresh_with_retries(
                 "refresh_result": refresh_result,
                 "snapshot_result": snapshot_result,
                 "elo_mmr_result": elo_mmr_result,
+                "database_maintenance_result": maintenance_result,
             }
         except Exception as exc:
             failure_payload = {
@@ -249,6 +256,91 @@ def _emit_json_log(payload: dict[str, Any]) -> None:
     print(json.dumps(payload, ensure_ascii=True, default=str), flush=True)
 
 
+def _maybe_run_database_maintenance(*, now: datetime | None = None) -> dict[str, Any]:
+    """Optionally run scheduled database maintenance without crashing the runner."""
+    global _LAST_DATABASE_MAINTENANCE_RUN_AT
+
+    anchor = now.astimezone(timezone.utc) if now else datetime.now(timezone.utc)
+    if not get_db_maintenance_enabled():
+        result = {"status": "skipped", "reason": "disabled", "enabled": False}
+        _emit_json_log({"event": "database-maintenance-scheduler-skipped-disabled", **result})
+        return result
+
+    interval_seconds, interval_source = _resolve_db_maintenance_interval_seconds()
+    if _LAST_DATABASE_MAINTENANCE_RUN_AT is not None:
+        elapsed_seconds = max(
+            0,
+            int((anchor - _LAST_DATABASE_MAINTENANCE_RUN_AT).total_seconds()),
+        )
+        if elapsed_seconds < interval_seconds:
+            result = {
+                "status": "skipped",
+                "reason": "not-due",
+                "enabled": True,
+                "interval_seconds": interval_seconds,
+                "interval_source": interval_source,
+                "elapsed_seconds": elapsed_seconds,
+                "last_run_at": _LAST_DATABASE_MAINTENANCE_RUN_AT.isoformat().replace(
+                    "+00:00", "Z"
+                ),
+            }
+            _emit_json_log({"event": "database-maintenance-scheduler-skipped-not-due", **result})
+            return result
+
+    _emit_json_log(
+        {
+            "event": "database-maintenance-scheduler-started",
+            "enabled": True,
+            "interval_seconds": interval_seconds,
+            "interval_source": interval_source,
+            "scheduled_at": anchor.isoformat().replace("+00:00", "Z"),
+        }
+    )
+    try:
+        result = run_database_maintenance_cleanup(apply=True, now=anchor)
+    except Exception as exc:  # noqa: BLE001 - scheduler must not crash the runner
+        result = {
+            "status": "error",
+            "error_type": type(exc).__name__,
+            "error": str(exc),
+            "enabled": True,
+            "interval_seconds": interval_seconds,
+            "interval_source": interval_source,
+        }
+        _emit_json_log({"event": "database-maintenance-scheduler-failed", **result})
+        return result
+
+    if result.get("status") == "ok":
+        _LAST_DATABASE_MAINTENANCE_RUN_AT = anchor
+        _emit_json_log(
+            {
+                "event": "database-maintenance-scheduler-completed",
+                "enabled": True,
+                "interval_seconds": interval_seconds,
+                "interval_source": interval_source,
+                "result": result,
+            }
+        )
+        return result
+
+    failed_result = {
+        "enabled": True,
+        "interval_seconds": interval_seconds,
+        "interval_source": interval_source,
+        "result": result,
+    }
+    _emit_json_log({"event": "database-maintenance-scheduler-failed", **failed_result})
+    return result
+
+
+def _resolve_db_maintenance_interval_seconds() -> tuple[int, str]:
+    """Return a safe maintenance interval even if env configuration is invalid."""
+    try:
+        return get_db_maintenance_interval_seconds(), "env"
+    except ValueError:
+        return DEFAULT_DB_MAINTENANCE_INTERVAL_SECONDS, "default-invalid-env-fallback"
+
+
 def _describe_refresh_scope(server_slug: str | None) -> list[str]:
     if server_slug:
         return [server_slug]
diff --git a/backend/tests/test_database_maintenance.py b/backend/tests/test_database_maintenance.py
new file mode 100644
index 0000000..e0af8aa
--- /dev/null
+++ b/backend/tests/test_database_maintenance.py
@@ -0,0 +1,448 @@
+from __future__ import annotations
+
+import io
+import json
+import sqlite3
+import tempfile
+import unittest
+from contextlib import closing, redirect_stdout
+from datetime import datetime, timedelta, timezone
+from pathlib import Path
+
+from app.database_maintenance import run_database_maintenance_cleanup
+from app.rcon_admin_log_materialization import MATCH_RESULT_SOURCE, initialize_rcon_materialized_storage
+from app.rcon_admin_log_storage import initialize_rcon_admin_log_storage
+from app.storage import initialize_storage
+
+
+class DatabaseMaintenanceTests(unittest.TestCase):
+    def test_dry_run_does_not_delete(self) -> None:
+        with _temp_db() as db_path:
+            _insert_server_snapshot(db_path, snapshot_id=1, captured_at="2026-05-01T00:00:00Z")
+
+            payload = run_database_maintenance_cleanup(
+                db_path=db_path,
+                now="2026-06-20T12:00:00Z",
+            )
+
+            self.assertEqual(payload["status"], "ok")
+            self.assertEqual(payload["mode"], "dry-run")
+            with closing(sqlite3.connect(db_path)) as connection:
+                self.assertEqual(
+                    connection.execute("SELECT COUNT(*) FROM server_snapshots").fetchone()[0],
+                    1,
+                )
+
+    def test_apply_deletes_old_server_snapshots(self) -> None:
+        with _temp_db() as db_path:
+            _insert_server_snapshot(db_path, snapshot_id=1, captured_at="2026-05-01T00:00:00Z")
+            _insert_server_snapshot(db_path, snapshot_id=2, captured_at="2026-06-18T00:00:00Z")
+
+            run_database_maintenance_cleanup(
+                apply=True,
+                db_path=db_path,
+                now="2026-06-20T12:00:00Z",
+                recent_matches_keep=1,
+            )
+
+            with closing(sqlite3.connect(db_path)) as connection:
+                ids = [row[0] for row in connection.execute("SELECT id FROM server_snapshots ORDER BY id")]
+            self.assertEqual(ids, [2])
+
+    def test_apply_deletes_old_noncritical_admin_log_events(self) -> None:
+        with _temp_db() as db_path:
+            _insert_admin_log_event(
+                db_path,
+                event_id=1,
+                event_type="chat",
+                event_timestamp="2026-04-01T00:00:00Z",
+                server_time=100,
+            )
+            _insert_admin_log_event(
+                db_path,
+                event_id=2,
+                event_type="chat",
+                event_timestamp="2026-06-15T00:00:00Z",
+                server_time=200,
+            )
+
+            run_database_maintenance_cleanup(
+                apply=True,
+                db_path=db_path,
+                now="2026-06-20T12:00:00Z",
+            )
+
+            with closing(sqlite3.connect(db_path)) as connection:
+                remaining = [
+                    tuple(row)
+                    for row in connection.execute(
+                        "SELECT id, event_type FROM rcon_admin_log_events ORDER BY id"
+                    )
+                ]
+            self.assertEqual(remaining, [(2, "chat")])
+
+    def test_apply_preserves_critical_events_within_retention(self) -> None:
+        with _temp_db() as db_path:
+            _insert_admin_log_event(
+                db_path,
+                event_id=1,
+                event_type="kill",
+                event_timestamp="2026-06-10T00:00:00Z",
+                server_time=100,
+            )
+
+            run_database_maintenance_cleanup(
+                apply=True,
+                db_path=db_path,
+                now="2026-06-20T12:00:00Z",
+            )
+
+            with closing(sqlite3.connect(db_path)) as connection:
+                count = connection.execute(
+                    "SELECT COUNT(*) FROM rcon_admin_log_events WHERE event_type = 'kill'"
+                ).fetchone()[0]
+            self.assertEqual(count, 1)
+
+    def test_apply_preserves_latest_100_materialized_matches(self) -> None:
+        with _temp_db() as db_path:
+            for index in range(101):
+                ended_at = (
+                    datetime(2026, 1, 1, 12, tzinfo=timezone.utc) + timedelta(days=index)
+                ).isoformat().replace("+00:00", "Z")
+                _insert_materialized_match(
+                    db_path,
+                    match_id=index + 1,
+                    match_key=f"match-{index + 1}",
+                    ended_at=ended_at,
+                    server_time_start=(index + 1) * 10,
+                    server_time_end=(index + 1) * 10 + 5,
+                )
+
+            run_database_maintenance_cleanup(
+                apply=True,
+                db_path=db_path,
+                now="2026-06-20T12:00:00Z",
+            )
+
+            with closing(sqlite3.connect(db_path)) as connection:
+                remaining = connection.execute(
+                    "SELECT COUNT(*) FROM rcon_materialized_matches"
+                ).fetchone()[0]
+                oldest = connection.execute(
+                    "SELECT COUNT(*) FROM rcon_materialized_matches WHERE match_key = 'match-1'"
+                ).fetchone()[0]
+            self.assertEqual(remaining, 100)
+            self.assertEqual(oldest, 0)
+
+    def test_apply_preserves_current_month_matches(self) -> None:
+        with _temp_db() as db_path:
+            _insert_materialized_match(
+                db_path,
+                match_id=1,
+                match_key="old",
+                ended_at="2026-01-10T12:00:00Z",
+                server_time_start=10,
+                server_time_end=20,
+            )
+            _insert_materialized_match(
+                db_path,
+                match_id=2,
+                match_key="current-month",
+                ended_at="2026-06-03T12:00:00Z",
+                server_time_start=30,
+                server_time_end=40,
+            )
+
+            run_database_maintenance_cleanup(
+                apply=True,
+                db_path=db_path,
+                now="2026-06-20T12:00:00Z",
+                recent_matches_keep=1,
+            )
+
+            with closing(sqlite3.connect(db_path)) as connection:
+                keys = [row[0] for row in connection.execute("SELECT match_key FROM rcon_materialized_matches")]
+            self.assertEqual(keys, ["current-month"])
+
+    def test_apply_preserves_previous_month_when_now_day_is_early(self) -> None:
+        with _temp_db() as db_path:
+            _insert_materialized_match(
+                db_path,
+                match_id=1,
+                match_key="previous-month",
+                ended_at="2026-05-15T12:00:00Z",
+                server_time_start=10,
+                server_time_end=20,
+            )
+            _insert_materialized_match(
+                db_path,
+                match_id=2,
+                match_key="older",
+                ended_at="2026-04-15T12:00:00Z",
+                server_time_start=30,
+                server_time_end=40,
+            )
+
+            run_database_maintenance_cleanup(
+                apply=True,
+                db_path=db_path,
+                now="2026-06-05T12:00:00Z",
+                recent_matches_keep=1,
+            )
+
+            with closing(sqlite3.connect(db_path)) as connection:
+                keys = [row[0] for row in connection.execute("SELECT match_key FROM rcon_materialized_matches")]
+            self.assertEqual(keys, ["previous-month"])
+
+    def test_apply_preserves_current_week(self) -> None:
+        with _temp_db() as db_path:
+            _insert_materialized_match(
+                db_path,
+                match_id=1,
+                match_key="current-week",
+                ended_at="2026-06-10T12:00:00Z",
+                server_time_start=10,
+                server_time_end=20,
+            )
+            _insert_materialized_match(
+                db_path,
+                match_id=2,
+                match_key="older",
+                ended_at="2026-05-01T12:00:00Z",
+                server_time_start=30,
+                server_time_end=40,
+            )
+
+            run_database_maintenance_cleanup(
+                apply=True,
+                db_path=db_path,
+                now="2026-06-10T13:00:00Z",
+                recent_matches_keep=1,
+            )
+
+            with closing(sqlite3.connect(db_path)) as connection:
+                keys = [row[0] for row in connection.execute("SELECT match_key FROM rcon_materialized_matches")]
+            self.assertEqual(keys, ["current-week"])
+
+    def test_apply_preserves_previous_week_when_fallback_may_need_it(self) -> None:
+        with _temp_db() as db_path:
+            _insert_materialized_match(
+                db_path,
+                match_id=1,
+                match_key="previous-week",
+                ended_at="2026-06-03T12:00:00Z",
+                server_time_start=10,
+                server_time_end=20,
+            )
+            _insert_materialized_match(
+                db_path,
+                match_id=2,
+                match_key="current-week-sample",
+                ended_at="2026-06-09T12:00:00Z",
+                server_time_start=30,
+                server_time_end=40,
+            )
+            _insert_materialized_match(
+                db_path,
+                match_id=3,
+                match_key="older",
+                ended_at="2026-05-01T12:00:00Z",
+                server_time_start=50,
+                server_time_end=60,
+            )
+
+            run_database_maintenance_cleanup(
+                apply=True,
+                db_path=db_path,
+                now="2026-06-10T13:00:00Z",
+                recent_matches_keep=1,
+            )
+
+            with closing(sqlite3.connect(db_path)) as connection:
+                keys = {
+                    row[0]
+                    for row in connection.execute("SELECT match_key FROM rcon_materialized_matches")
+                }
+            self.assertEqual(keys, {"previous-week", "current-week-sample"})
+
+    def test_apply_deletes_old_non_protected_match_and_child_stats(self) -> None:
+        with _temp_db() as db_path:
+            _insert_materialized_match(
+                db_path,
+                match_id=1,
+                match_key="delete-me",
+                ended_at="2026-01-10T12:00:00Z",
+                server_time_start=10,
+                server_time_end=20,
+            )
+            _insert_materialized_match(
+                db_path,
+                match_id=2,
+                match_key="keep-me",
+                ended_at="2026-06-18T12:00:00Z",
+                server_time_start=30,
+                server_time_end=40,
+            )
+            _insert_player_stat(db_path, match_key="delete-me", player_id="player-1")
+            _insert_player_stat(db_path, match_key="keep-me", player_id="player-2")
+
+            run_database_maintenance_cleanup(
+                apply=True,
+                db_path=db_path,
+                now="2026-06-20T12:00:00Z",
+                recent_matches_keep=1,
+            )
+
+            with closing(sqlite3.connect(db_path)) as connection:
+                deleted_match_count = connection.execute(
+                    "SELECT COUNT(*) FROM rcon_materialized_matches WHERE match_key = 'delete-me'"
+                ).fetchone()[0]
+                deleted_stat_count = connection.execute(
+                    "SELECT COUNT(*) FROM rcon_match_player_stats WHERE match_key = 'delete-me'"
+                ).fetchone()[0]
+                kept_stat_count = connection.execute(
+                    "SELECT COUNT(*) FROM rcon_match_player_stats WHERE match_key = 'keep-me'"
+                ).fetchone()[0]
+            self.assertEqual(deleted_match_count, 0)
+            self.assertEqual(deleted_stat_count, 0)
+            self.assertEqual(kept_stat_count, 1)
+
+    def test_missing_optional_tables_are_logged_and_do_not_crash(self) -> None:
+        with _temp_db(create_schema=False) as db_path:
+            stream = io.StringIO()
+            with redirect_stdout(stream):
+                payload = run_database_maintenance_cleanup(
+                    db_path=db_path,
+                    now="2026-06-20T12:00:00Z",
+                )
+
+            self.assertEqual(payload["status"], "ok")
+            self.assertIn("database-maintenance-table-skipped", stream.getvalue())
+
+
+def _temp_db(*, create_schema: bool = True):
+    class _TempDbContext:
+        def __enter__(self) -> Path:
+            self._tmpdir = tempfile.TemporaryDirectory(ignore_cleanup_errors=True)
+            self.db_path = Path(self._tmpdir.name) / "maintenance.sqlite3"
+            if create_schema:
+                initialize_storage(db_path=self.db_path)
+                initialize_rcon_admin_log_storage(db_path=self.db_path)
+                initialize_rcon_materialized_storage(db_path=self.db_path)
+            return self.db_path
+
+        def __exit__(self, exc_type, exc, tb) -> None:
+            self._tmpdir.cleanup()
+
+    return _TempDbContext()
+
+
+def _insert_server_snapshot(db_path: Path, *, snapshot_id: int, captured_at: str) -> None:
+    with closing(sqlite3.connect(db_path)) as connection:
+        connection.execute(
+            """
+            INSERT OR IGNORE INTO game_sources (
+                id, slug, display_name, provider_kind, is_active, created_at, updated_at
+            ) VALUES (1, 'current-hll', 'Current Hell Let Loose', 'development', 1, CURRENT_TIMESTAMP, CURRENT_TIMESTAMP)
+            """
+        )
+        connection.execute(
+            """
+            INSERT OR IGNORE INTO servers (
+                id, game_source_id, external_server_id, server_name, region, first_seen_at, last_seen_at
+            ) VALUES (1, 1, 'server-1', 'Server 1', 'ES', ?, ?)
+            """,
+            (captured_at, captured_at),
+        )
+        connection.execute(
+            """
+            INSERT INTO server_snapshots (
+                id, server_id, captured_at, status, players, max_players, current_map, source_name
+            ) VALUES (?, 1, ?, 'online', 10, 100, 'hurtgen', 'test')
+            """,
+            (snapshot_id, captured_at),
+        )
+        connection.commit()
+
+
+def _insert_admin_log_event(
+    db_path: Path,
+    *,
+    event_id: int,
+    event_type: str,
+    event_timestamp: str,
+    server_time: int,
+) -> None:
+    with closing(sqlite3.connect(db_path)) as connection:
+        connection.execute(
+            """
+            INSERT INTO rcon_admin_log_events (
+                id, target_key, external_server_id, event_timestamp, server_time,
+                relative_time, event_type, raw_message, canonical_message,
+                parsed_payload_json, raw_entry_json
+            ) VALUES (?, 'comunidad-hispana-01', 'comunidad-hispana-01', ?, ?, '', ?, '', '', '{}', '{}')
+            """,
+            (event_id, event_timestamp, server_time, event_type),
+        )
+        connection.commit()
+
+
+def _insert_materialized_match(
+    db_path: Path,
+    *,
+    match_id: int,
+    match_key: str,
+    ended_at: str,
+    server_time_start: int,
+    server_time_end: int,
+) -> None:
+    started_at = _shift_iso(ended_at, hours=-1)
+    with closing(sqlite3.connect(db_path)) as connection:
+        connection.execute(
+            """
+            INSERT INTO rcon_materialized_matches (
+                id, target_key, external_server_id, match_key, map_name, map_pretty_name,
+                game_mode, started_server_time, ended_server_time, started_at, ended_at,
+                allied_score, axis_score, winner, confidence_mode, source_basis
+            ) VALUES (?, 'comunidad-hispana-01', 'comunidad-hispana-01', ?, 'hurtgen', 'Hurtgen Forest',
+                      'warfare', ?, ?, ?, ?, 5, 3, 'allied', 'exact', ?)
+            """,
+            (
+                match_id,
+                match_key,
+                server_time_start,
+                server_time_end,
+                started_at,
+                ended_at,
+                MATCH_RESULT_SOURCE,
+            ),
+        )
+        connection.commit()
+
+
+def _insert_player_stat(db_path: Path, *, match_key: str, player_id: str) -> None:
+    with closing(sqlite3.connect(db_path)) as connection:
+        connection.execute(
+            """
+            INSERT INTO rcon_match_player_stats (
+                target_key, match_key, player_id, player_name, team,
+                kills, deaths, teamkills, deaths_by_teamkill,
+                weapons_json, death_by_weapons_json, most_killed_json, death_by_json
+            ) VALUES (
+                'comunidad-hispana-01', ?, ?, ?, 'Allies',
+                1, 1, 0, 0, '{}', '{}', '{}', '{}'
+            )
+            """,
+            (match_key, player_id, player_id),
+        )
+        connection.commit()
+
+
+def _shift_iso(value: str, *, hours: int) -> str:
+    point = datetime.fromisoformat(value.replace("Z", "+00:00")).astimezone(timezone.utc)
+    shifted = point + timedelta(hours=hours)
+    return shifted.isoformat().replace("+00:00", "Z")
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/backend/tests/test_historical_runner_maintenance.py b/backend/tests/test_historical_runner_maintenance.py
new file mode 100644
index 0000000..10d0513
--- /dev/null
+++ b/backend/tests/test_historical_runner_maintenance.py
@@ -0,0 +1,153 @@
+from __future__ import annotations
+
+import io
+import os
+import unittest
+from contextlib import nullcontext, redirect_stdout
+from datetime import datetime, timezone
+from unittest.mock import patch
+
+import app.historical_runner as historical_runner
+from app.historical_runner import _maybe_run_database_maintenance, _run_refresh_with_retries
+
+
+class HistoricalRunnerMaintenanceTests(unittest.TestCase):
+    def setUp(self) -> None:
+        historical_runner._LAST_DATABASE_MAINTENANCE_RUN_AT = None
+
+    def tearDown(self) -> None:
+        historical_runner._LAST_DATABASE_MAINTENANCE_RUN_AT = None
+
+    def test_scheduler_disabled_does_not_call_cleanup(self) -> None:
+        with (
+            patch.dict(os.environ, {"HLL_DB_MAINTENANCE_ENABLED": "false"}, clear=False),
+            patch("app.historical_runner.run_database_maintenance_cleanup") as cleanup,
+        ):
+            result = _maybe_run_database_maintenance(
+                now=datetime(2026, 6, 20, 12, tzinfo=timezone.utc)
+            )
+
+        cleanup.assert_not_called()
+        self.assertEqual(result["status"], "skipped")
+        self.assertEqual(result["reason"], "disabled")
+
+    def test_scheduler_enabled_but_not_due_does_not_call_cleanup(self) -> None:
+        with (
+            patch.dict(
+                os.environ,
+                {
+                    "HLL_DB_MAINTENANCE_ENABLED": "true",
+                    "HLL_DB_MAINTENANCE_INTERVAL_SECONDS": "43200",
+                },
+                clear=False,
+            ),
+            patch(
+                "app.historical_runner.run_database_maintenance_cleanup",
+                return_value={"status": "ok"},
+            ) as cleanup,
+        ):
+            first = _maybe_run_database_maintenance(
+                now=datetime(2026, 6, 20, 0, tzinfo=timezone.utc)
+            )
+            second = _maybe_run_database_maintenance(
+                now=datetime(2026, 6, 20, 1, tzinfo=timezone.utc)
+            )
+
+        self.assertEqual(first["status"], "ok")
+        self.assertEqual(second["status"], "skipped")
+        self.assertEqual(second["reason"], "not-due")
+        cleanup.assert_called_once()
+
+    def test_scheduler_enabled_and_due_calls_cleanup(self) -> None:
+        with (
+            patch.dict(os.environ, {"HLL_DB_MAINTENANCE_ENABLED": "true"}, clear=False),
+            patch(
+                "app.historical_runner.run_database_maintenance_cleanup",
+                return_value={"status": "ok"},
+            ) as cleanup,
+        ):
+            result = _maybe_run_database_maintenance(
+                now=datetime(2026, 6, 20, 12, tzinfo=timezone.utc)
+            )
+
+        cleanup.assert_called_once()
+        self.assertEqual(result["status"], "ok")
+
+    def test_cleanup_exception_is_logged_and_runner_continues(self) -> None:
+        stream = io.StringIO()
+        with (
+            patch.dict(os.environ, {"HLL_DB_MAINTENANCE_ENABLED": "true"}, clear=False),
+            patch("app.historical_runner.backend_writer_lock", return_value=nullcontext()),
+            patch(
+                "app.historical_runner._run_primary_rcon_capture",
+                return_value={"status": "ok", "targets": []},
+            ),
+            patch(
+                "app.historical_runner.run_incremental_refresh",
+                return_value={"status": "ok"},
+            ),
+            patch(
+                "app.historical_runner.generate_historical_snapshots",
+                return_value={"status": "ok"},
+            ),
+            patch(
+                "app.historical_runner.rebuild_elo_mmr_models",
+                return_value={"status": "ok"},
+            ),
+            patch(
+                "app.historical_runner.run_database_maintenance_cleanup",
+                side_effect=RuntimeError("maintenance failed"),
+            ),
+            redirect_stdout(stream),
+        ):
+            result = _run_refresh_with_retries(
+                max_retries=0,
+                retry_delay_seconds=0,
+                server_slug="comunidad-hispana-01",
+                max_pages=None,
+                page_size=None,
+                run_number=1,
+            )
+
+        self.assertEqual(result["status"], "ok")
+        self.assertEqual(result["database_maintenance_result"]["status"], "error")
+        self.assertIn("database-maintenance-scheduler-failed", stream.getvalue())
+
+    def test_interval_parsing_handles_invalid_values_safely(self) -> None:
+        with patch.dict(
+            os.environ,
+            {
+                "HLL_DB_MAINTENANCE_ENABLED": "true",
+                "HLL_DB_MAINTENANCE_INTERVAL_SECONDS": "bad",
+            },
+            clear=False,
+        ):
+            interval_seconds, source = historical_runner._resolve_db_maintenance_interval_seconds()
+
+        self.assertEqual(interval_seconds, 43200)
+        self.assertEqual(source, "default-invalid-env-fallback")
+
+    def test_maintenance_state_is_tracked_in_process(self) -> None:
+        with (
+            patch.dict(
+                os.environ,
+                {
+                    "HLL_DB_MAINTENANCE_ENABLED": "true",
+                    "HLL_DB_MAINTENANCE_INTERVAL_SECONDS": "3600",
+                },
+                clear=False,
+            ),
+            patch(
+                "app.historical_runner.run_database_maintenance_cleanup",
+                return_value={"status": "ok"},
+            ),
+        ):
+            _maybe_run_database_maintenance(now=datetime(2026, 6, 20, 12, tzinfo=timezone.utc))
+            self.assertEqual(
+                historical_runner._LAST_DATABASE_MAINTENANCE_RUN_AT,
+                datetime(2026, 6, 20, 12, tzinfo=timezone.utc),
+            )
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/docs/database-maintenance.md b/docs/database-maintenance.md
new file mode 100644
index 0000000..ce0a521
--- /dev/null
+++ b/docs/database-maintenance.md
@@ -0,0 +1,307 @@
+# Database Maintenance
+
+## Overview
+
+HLL Vietnam keeps database cleanup at the application level.
+
+The current maintenance scope is intentionally narrow:
+
+- old `server_snapshots`;
+- old non-critical `rcon_admin_log_events`;
+- old critical `rcon_admin_log_events` only after retention and protected-match checks;
+- old non-protected `rcon_materialized_matches`;
+- dependent `rcon_match_player_stats` for deleted matches.
+
+The first maintenance pass does not routinely delete:
+
+- `displayed_historical_snapshots`;
+- file-based snapshots under `backend/data/snapshots/`;
+- public-scoreboard `historical_*` fallback tables;
+- `player_event_raw_ledger` and its worker metadata;
+- Elo/MMR tables;
+- Comunidad Hispana #03 data reactivation or targets.
+
+## Why Application-Level And Not `pg_cron`
+
+Cleanup is versioned in backend code instead of delegated to `pg_cron`, host cron, or a separate container because the retention logic depends on product rules:
+
+- keep the latest 100 closed materialized matches;
+- keep the current month;
+- keep the previous month during the first 7 days of a new month;
+- keep the current week;
+- keep the previous week when weekly fallback may still need it;
+- keep child stats for protected matches;
+- avoid breaking current/live pages that still read recent AdminLog data.
+
+Those rules belong with the application’s read and write model, not inside database-only scheduling.
+
+## Scheduled Cleanup Inside `historical-runner`
+
+Database maintenance is scheduled inside `app.historical_runner`.
+
+Behavior:
+
+- disabled by default;
+- no extra Docker service is added for maintenance;
+- the runner checks whether maintenance is due;
+- when enabled and due, the runner invokes `python -m app.database_maintenance cleanup --apply` behavior through the shared Python function;
+- failures are logged and do not crash the historical runner loop;
+- cleanup runs under the same writer-lock coordination used by the historical writer flows.
+
+Relevant structured log events:
+
+- `database-maintenance-scheduler-skipped-disabled`
+- `database-maintenance-scheduler-skipped-not-due`
+- `database-maintenance-scheduler-started`
+- `database-maintenance-scheduler-completed`
+- `database-maintenance-scheduler-failed`
+
+## Environment Variables
+
+Required maintenance-related variables:
+
+```text
+HLL_DB_MAINTENANCE_ENABLED=false
+HLL_DB_MAINTENANCE_INTERVAL_SECONDS=43200
+HLL_RECENT_MATCHES_KEEP=100
+HLL_ADMIN_LOG_NONCRITICAL_RETENTION_DAYS=30
+HLL_ADMIN_LOG_CRITICAL_RETENTION_DAYS=90
+HLL_SERVER_SNAPSHOT_RETENTION_DAYS=14
+HLL_DB_MAINTENANCE_BATCH_SIZE=5000
+```
+
+Meaning:
+
+- `HLL_DB_MAINTENANCE_ENABLED`
+  Enables scheduled apply mode inside `historical-runner`.
+- `HLL_DB_MAINTENANCE_INTERVAL_SECONDS`
+  Default scheduler interval. `43200` means every 12 hours.
+- `HLL_RECENT_MATCHES_KEEP`
+  Number of latest closed materialized matches that must always be protected.
+- `HLL_ADMIN_LOG_NONCRITICAL_RETENTION_DAYS`
+  Retention for non-critical AdminLog events such as chat/connect/disconnect.
+- `HLL_ADMIN_LOG_CRITICAL_RETENTION_DAYS`
+  Retention for critical AdminLog events such as `kill`, `match_start`, `match_end`.
+- `HLL_SERVER_SNAPSHOT_RETENTION_DAYS`
+  Retention for live server snapshots.
+- `HLL_DB_MAINTENANCE_BATCH_SIZE`
+  Delete batch size for apply mode.
+
+## Protected Data
+
+The cleanup command protects:
+
+- latest 100 closed materialized matches by default;
+- current month materialized matches;
+- previous month materialized matches when the current day is `1` through `7`;
+- current week materialized matches;
+- previous week materialized matches when weekly fallback may still need them;
+- `rcon_match_player_stats` belonging to protected matches;
+- current/live AdminLog data required for visible current-match surfaces;
+- `displayed_historical_snapshots`;
+- file snapshots in `backend/data/snapshots/`.
+
+If a match timestamp cannot be interpreted safely, that match is skipped and protected instead of deleted.
+
+## Deleted Data
+
+Apply mode is currently allowed to delete:
+
+- `server_snapshots` older than retention;
+- non-critical `rcon_admin_log_events` older than retention;
+- critical `rcon_admin_log_events` older than retention only when they are not required by protected materialized match ranges;
+- non-protected `rcon_materialized_matches`;
+- dependent `rcon_match_player_stats` for deleted matches.
+
+Current critical AdminLog event types:
+
+- `kill`
+- `match_start`
+- `match_end`
+
+## Dry-Run Command
+
+From `backend/`:
+
+```powershell
+python -m app.database_maintenance cleanup --dry-run
+```
+
+From the repository root with the backend package on `PYTHONPATH`:
+
+```powershell
+$env:PYTHONPATH='backend'
+python -m app.database_maintenance cleanup --dry-run
+```
+
+Inside Docker Compose:
+
+```powershell
+docker compose exec backend python -m app.database_maintenance cleanup --dry-run
+```
+
+Useful dry-run options:
+
+```powershell
+docker compose exec backend python -m app.database_maintenance cleanup --dry-run `
+  --recent-matches-keep 100 `
+  --admin-log-noncritical-retention-days 30 `
+  --admin-log-critical-retention-days 90 `
+  --server-snapshot-retention-days 14 `
+  --batch-size 5000
+```
+
+Dry-run is the safe preview path and should be reviewed before any production apply.
+
+## Apply Command
+
+Local module execution:
+
+```powershell
+python -m app.database_maintenance cleanup --apply
+```
+
+Docker Compose:
+
+```powershell
+docker compose exec backend python -m app.database_maintenance cleanup --apply
+```
+
+One-off local validation with a fixed time anchor:
+
+```powershell
+python -m app.database_maintenance cleanup --apply --now 2026-06-20T12:00:00Z
+```
+
+Optional maintenance vacuum/analyze:
+
+```powershell
+python -m app.database_maintenance cleanup --apply --vacuum-analyze
+```
+
+## Table-Size Audit SQL
+
+```sql
+select
+  schemaname,
+  relname as table_name,
+  pg_size_pretty(pg_total_relation_size(relid)) as total_size,
+  pg_size_pretty(pg_relation_size(relid)) as table_size,
+  pg_size_pretty(pg_total_relation_size(relid) - pg_relation_size(relid)) as indexes_size,
+  n_live_tup as estimated_rows,
+  n_dead_tup as estimated_dead_rows
+from pg_stat_user_tables
+order by pg_total_relation_size(relid) desc;
+```
+
+## Row-Count And Age Audit SQL
+
+### AdminLog events by type/date
+
+```sql
+select
+  event_type,
+  count(*) as row_count,
+  min(event_timestamp) as first_event_timestamp,
+  max(event_timestamp) as last_event_timestamp,
+  min(server_time) as first_server_time,
+  max(server_time) as last_server_time
+from rcon_admin_log_events
+group by event_type
+order by row_count desc, event_type asc;
+```
+
+### Materialized matches by server/date
+
+```sql
+select
+  target_key,
+  source_basis,
+  count(*) as matches,
+  min(coalesce(ended_at, started_at)) as first_closed_at,
+  max(coalesce(ended_at, started_at)) as last_closed_at
+from rcon_materialized_matches
+group by target_key, source_basis
+order by target_key asc, source_basis asc;
+```
+
+### Server snapshots by date
+
+```sql
+select
+  server_id,
+  min(captured_at) as first_captured_at,
+  max(captured_at) as last_captured_at,
+  count(*) as snapshot_rows
+from server_snapshots
+group by server_id
+order by last_captured_at desc;
+```
+
+### Displayed snapshots count
+
+```sql
+select
+  snapshot_type,
+  metric,
+  snapshot_window,
+  count(*) as snapshot_rows,
+  min(generated_at) as first_generated_at,
+  max(generated_at) as last_generated_at
+from displayed_historical_snapshots
+group by snapshot_type, metric, snapshot_window
+order by snapshot_type asc, metric asc, snapshot_window asc;
+```
+
+## Logs To Inspect
+
+The cleanup command emits JSON logs. Minimum events to look for:
+
+- `database-maintenance-started`
+- `database-maintenance-plan`
+- `database-maintenance-table-skipped`
+- `database-maintenance-delete-batch`
+- `database-maintenance-completed`
+- `database-maintenance-error`
+
+Examples:
+
+```powershell
+docker compose logs --tail=200 backend
+docker compose logs --tail=200 historical-runner
+```
+
+If scheduled cleanup is enabled:
+
+```powershell
+docker compose logs --tail=200 historical-runner
+```
+
+## Docker And Portainer Warnings
+
+- Never use `docker compose down -v` unless you intentionally want to delete PostgreSQL and mounted volume data.
+- Always review dry-run output before enabling apply in production.
+- Do not manually delete protected match or player-stat rows from PostgreSQL.
+- Keep backups before changing retention settings.
+- Do not add Comunidad Hispana #03 back into RCON targets in this task.
+- Do not add a separate maintenance container, host cron, or `pg_cron` job for this feature.
+
+For Portainer-style operations the same warning applies:
+
+- deleting volumes is destructive;
+- maintenance should run through the application command, not through manual table purges.
+
+## Rollback And Restore Considerations
+
+- Retention changes are destructive when apply mode runs.
+- Keep a PostgreSQL backup before enabling scheduled apply in production.
+- If cleanup removes too much data, recovery is restore-based, not “undo last delete.”
+- Favor dry-run, smaller batch sizes, and reviewed retention values before long-running scheduled apply.
+
+## Safe Operator Flow
+
+1. Audit table size and row ages with the SQL above.
+2. Run dry-run locally or in Compose.
+3. Review protected counts and candidate counts in JSON output.
+4. Enable `HLL_DB_MAINTENANCE_ENABLED=true` only after dry-run review.
+5. Monitor `historical-runner` logs for scheduler events and cleanup completion.
diff --git a/frontend/index.html b/frontend/index.html
index bf01812..3b79278 100644
--- a/frontend/index.html
+++ b/frontend/index.html
@@ -5,9 +5,9 @@
     <meta name="viewport" content="width=device-width, initial-scale=1.0" />
     <meta
       name="description"
-      content="Comunidad hispana de HLL Vietnam. Accede al Discord oficial y descubre el trailer del proyecto."
+      content="Comunidad hispana de HLL. Accede al Discord oficial y descubre el trailer del proyecto."
     />
-    <title>Comunidad Hispana - HLL Vietnam</title>
+    <title>Comunidad Hispana - HLL</title>
     <link rel="stylesheet" href="./assets/css/styles.css" />
     <link rel="stylesheet" href="./assets/css/hero-header-compact.css" />
   </head>
@@ -23,7 +23,7 @@
             <div class="logo-frame">
               <img
                 src="./assets/img/logo.png"
-                alt="Logo oficial de la comunidad HLL Vietnam"
+                alt="Logo oficial de la comunidad HLL"
                 class="logo-frame__image"
                 width="1024"
                 height="1044"
@@ -33,7 +33,7 @@
             <div class="hero__copy">
               <h1 class="hero__title">
                 Comunidad Hispana
-                <span class="hero__title-accent">HLL Vietnam</span>
+                <span class="hero__title-accent">HLL</span>
               </h1>
               <p class="hero__text">
                 Centro de reunion para escuadras, eventos y acceso directo al