Walk me through Apex governor limits in detail and how to stay under them.

Salesforce enforces hard caps per transaction. The big ones:

| Limit | Sync | Async | What hits it | |---|---|---|---| | SOQL queries | 100 | 200 | Loops with SOQL | | SOQL rows | 50,000 | 50,000 | Massive single query | | DML statements | 150 | 150 | Loops with DML | | DML records | 10,000 | 10,000 | Single bulk DML | | CPU time | 10s | 60s | Heavy computation, recursive flow | | Heap size | 6 MB | 12 MB | Large collections | | Callouts | 100 | 100 | API integrations | | Future calls / queueable | 50 | 50 | Triggering many async jobs |

Patterns to stay under:

1. Bulkify everything. Never SOQL or DML inside loops. Build collections, do one query/DML.

1. Use Maps for lookups, not SOQL. If you need parent records: query parents into a Map by Id, then look up in-memory.

1. Selective queries. WHERE clause must hit indexed fields. Without selectivity, queries against large tables fail or time out (separate from row limits).

1. Don't load more than you need. SELECT Id, Name FROM Account not SELECT * FROM Account. Reduces heap and SOQL row count.

1. Use QueryLocator in Batch for >50k rows. Database.getQueryLocator() returns iterator, not a static list.

1. Async for heavy work. Move CPU/heap-heavy logic into Queueable or Batch — they get more headroom.

1. Limit governor-limit counter awareness. Limits.getQueries(), Limits.getDmlStatements(), etc. — read at runtime to make conditional decisions.

1. Static caching across method calls in the same transaction — store results of expensive queries in a static Map.

1. Avoid recursion. Use static flags. Each re-entry adds to limits.

Common mistakes:

• A trigger handler that re-queries [SELECT ... FROM Contact WHERE AccountId IN :accIds] from inside another method, when the parent already has it.
• A flow that loops over Get Records results and does Update inside the loop.
• An LWC @wire Apex method that does heavy work for every page render.
• A managed package that adds 30 SOQL queries per save without coordinating.

The discipline: every method that does SOQL or DML should accept and return a list, not a single record. That's the fundamental shift.