MiniMongoDB ↔ MongoDB mapping

The Level column is part of the teaching contract:

• Equivalent means the local mechanism preserves the named core invariant, not that its performance or implementation is production-equivalent.
• Intentionally simplified means the direction is the same but the surface or machinery is smaller.
• Semantically opposite means the miniature deliberately reverses an important real-system choice; never transfer that behavior by analogy.

MiniMongoDB module	Real MongoDB concept/subsystem	Level	What to carry forward
bson.types.ObjectId	BSON ObjectId identity value	Semantically opposite	Both are 12-byte-shaped identities, but real ObjectIds encode time/process/random/counter material; this project bans those sources and uses one injected counter.
bson.types dict/list model	Binary BSON values and comparison order	Intentionally simplified	Values remain typed and documents/arrays remain distinct. The supported types and cross-type order are much smaller.
bson.path	MQL dotted field paths	Intentionally simplified	Nested fields and explicit numeric array indexes share one path notation; ambiguous sparse-array creation is rejected here.
query.matcher scalar-vs-array equality	MQL multikey matching behavior	Equivalent	A scalar predicate can match an element of a stored array without explicit iteration.
query.matcher literal document equality	BSON embedded-document equality	Equivalent	A literal embedded document is compared as a complete value, including field order; dotted paths select individual nested fields.
query.matcher range comparisons	MQL comparison predicates	Semantically opposite	MongoDB normally applies type bracketing to range predicates; MiniMongoDB exposes its global teaching order across supported types.
query.matcher top-level $not	field-level $not query operator	Semantically opposite	MiniMongoDB accepts a project-specific top-level logical $not; real MongoDB supports $not only as a field operator and rejects the top-level form.
update.operators	MQL update modifiers	Intentionally simplified	Replacement and operator update are separate; path changes are single-document atomic. Operator options and array filters are absent.
index.IdIndex	automatically-created unique _id_ index	Equivalent	Every collection has identity uniqueness under BSON-typed equality before a write becomes visible. Canonical tagged keys back a Python hash map, not a B-tree.
index.SecondaryIndex	ascending secondary, compound, unique, and multikey indexes	Intentionally simplified	Dotted fields share _id's canonical BSON-tagged equality; an array-bearing document owns multiple de-duplicated keys. Compound scans require a leftmost prefix. The local ordered map is not a B-tree and does not implement descending, sparse, partial, or specialized indexes.
collection.Collection	collection CRUD layer	Intentionally simplified	Documents cross a copy boundary and publish only after their journal entry is durable. Batch methods commit a prefix rather than acting as multi-document transactions.
oplog.OplogEntry	replica-set oplog post-image/idempotence discipline	Intentionally simplified	Action updates become repeat-safe final assignments. Real oplog formats and version-specific update encodings are richer.
storage.journal carrying oplog frames	WiredTiger journal plus replica oplog	Semantically opposite	Real MongoDB separates storage-engine recovery records from the replication oplog; M1 reuses logical oplog entries as its local durability journal.
storage.checkpoint	WiredTiger checkpoint	Intentionally simplified	Restart begins from a snapshot and applies newer durable records. The snapshot is whole-database tagged JSON, without pages or MVCC.
storage.recovery	startup recovery	Equivalent	Only a CRC-valid journal prefix is replayed, and replaying already-applied post-images is harmless.
plan.choose_plan	query planner and explain	Intentionally simplified	Prefix-compatible indexes estimate selectivity from candidate ownership. IXSCAN wins only when it examines fewer documents than COLLSCAN; explain reports the winning Mongo-named stage and actual key/document counts.
aggregate.execute_pipeline	aggregation pipeline execution	Intentionally simplified	$match/$project/$group/$sort/$limit compose as document operators; group supports $sum/$avg/$min/$max/$push. Streaming and blocking stage boundaries are explicit, without MongoDB's optimizer or distributed pipeline.
oplog.capped docstring	capped local.oplog.rs retention	Intentionally simplified	The interface direction is recorded, but bounded retention is an explicit M3 item.

One write through the miniature

update_one({"_id": 1}, {"$inc": {"visits": 1}})
  → matcher selects one document
  → update engine mutates an isolated copy
  → immutable _id is validated
  → oplog prepares {"$set": {"visits": <final value>}}
  → journal appends length | payload | CRC and fsyncs
  → oplog publishes its sequence and in-memory entry
  → collection swaps the copy and all indexes atomically

The crucial ownership boundary is between the user command and the durable entry. The command says what action to attempt; the entry says which final state to converge on.

Relation operators versus document pipelines

MiniPostgres turns parsed SQL over schema-bound rows into a plan tree whose Volcano operators pull tuples from children. MiniMongoDB starts with an already-shaped query document, chooses IXSCAN or COLLSCAN, then passes self-shaped nested documents through aggregation stages. Both teach operator composition and blocking boundaries (Sort/Aggregate versus $sort/$group); only the relational side binds columns to a schema, while the document side keeps dotted paths and multikey fan-out as runtime value semantics.