Durable does not mean every read waits for storage hardware. Local does not mean fake. Fast claims only mean something when the measured category is visible.
A founder sees a shocking speed claim and wants to know whether it is real. The answer is not “trust us.” The answer is the proof trail: mechanism, category, ledger row, non-claims, and comparison guardrails.Mechanism
Local durable shape, warm mapped pages, and content identity remove common read-path work.
→
Category
Warm local read, content lookup, durable batch write, or batch ceiling.
→
Boundary
Cold storage, network, every query, and every write are not silently included.
The short version
Bitfield is fast because the storage shape removes work from the read path. Durable data means the data survives a crash or restart. Warm local read means the needed local file pages are already available and Bitfield has the read address.Durable does not mean every read waits for disk
A memory map connects a durable local file to the program reading it. The operating system lets Bitfield touch local file pages like memory. The file is still durable storage.When the operating system already has the needed pages warm, Bitfield can read known local bytes without sending a request to a separate database process.If the page is cold and hardware has to fetch it, physics is slower. That is why the honest wording is durable data with a warm local read path.Content addresses remove a kind of searching
Most database reads start by asking another database process a question. That process parses the question, plans the work, checks indexes, walks structures, and sends the answer back.Bitfield stores content by identity. Content identity means bytes can be named by what they are. Once Bitfield knows the local bytes to read, the warm local read path has much less work to do.No benchmark gets to hide its category
This is not a promise that a cold disk read takes less than a nanosecond. It is not a claim that a network round trip disappears. It is not saying every vendor benchmark is comparable to every other one.The rule is simple: show the mechanism, show the tradeoff, and say what work is actually being measured.
Common failures
| Failure | Why it is wrong | Fix |
|---|---|---|
| Saying the number means every database operation | It ignores category boundaries | Name the measured job first |
| Comparing warm local read to a remote query | It mixes local work with network/server work | Use Comparison guardrails |
| Explaining private layouts | It gives away the wrong detail | Stay at public mechanism-category level |
| Dropping non-claims | Readers overread the headline | Link the Claim ledger |