The Git mental model

Git has four object types, all content-addressed by the SHA of their contents. A blob is file contents. A tree is a directory listing of blobs and other trees. A commit points at one tree plus zero or more parent commits, with author and message attached. A tag points at a commit with a name. Because a commit's hash covers its tree and its parents, changing anything about history necessarily produces new hashes for everything downstream — this is why rewriting history is never in-place.

Commits form a directed acyclic graph. A branch is a 40-character file under .git/refs/heads/ containing one commit hash — nothing more. HEAD is a pointer to the branch you're on. Committing writes a new commit whose parent is the current one, then advances the branch label. Merging creates a commit with two parents. Rebasing copies commits onto a new base and moves the label, leaving the originals orphaned until garbage collection.

Between your editor and the repository sit two more places: the working tree (actual files on disk), and the index — the staging area, which is a proposed next tree. git add copies working-tree content into the index; git commit turns the index into a tree object and a commit. Most confusion about reset dissolves once you see it as choosing how many of those three to move: --soft moves only the branch, --mixed also resets the index, --hard also overwrites the working tree.

Remotes are just other repositories with their own refs. origin/main is a local cache of where main was on the remote last time you talked to it; fetch updates that cache and touches nothing else, while pull is fetch plus a merge or rebase. Separating those two is what makes "my branch diverged" legible instead of scary.

Deep-dive rounds don't ask you to recite Git internals, but they do ask what you did when a rebase went wrong or how you'd untangle a shared branch — and the answers are trivially derivable from the model and pure memorisation without it. The graph model is also what makes reflog, bisect, and cherry-pick obvious tools rather than incantations.

• Commits are immutable snapshots, not diffs; Git computes diffs on demand between two trees.
• A branch is a movable pointer to a commit, so creating and deleting branches is nearly free.
• Rewriting history always creates new commit objects, because the hash covers the parent chain.
• The index is a real third state between working tree and repository, and reset is best understood as choosing which of the three to move.
• origin/main is a local cache of a remote ref, which is why fetch and pull behave differently.
• Nothing you commit is lost by moving labels — the old commits remain reachable through the reflog until garbage collection.
• Detached HEAD just means HEAD points directly at a commit instead of at a branch label.