drag to rotate ✥

Read against
the human brain.

braineon is an engineered memory system. It happens to follow the same encode → consolidate → retrieve → monitor arc your field has characterised for a century — so here it is, mapped honestly, term by term.

A note on the analogy. This is not a brain model and makes no mechanistic claim. Where we write “like the hippocampus,” we mean functionally analogous — a shared vocabulary for engineers and neuroscientists, not an equivalence. The divergences are the interesting part.

The pipeline

The canonical arc, made legible

The same four stages — only here every stage is inspectable state you can open, diff, and replay. Tap to separate them.

Encode
afferent intake
bind the episode
01
Consolidate
systems consolidation
stabilise · no sleep
02
Retrieve
separation · completion
the right trace, back
03
Monitor
metamemory
confidence + source
04

tap to separate ↑

The mapping

Term by term

Each construct as your field defines it, its counterpart in braineon, and where the engineered system parts ways.

Encodingstimulus → trace
In the brainAttention-gated, reconstructive; the hippocampus rapidly binds a sparse episodic trace.
In braineonConnectors ingest events from your tools; each becomes a discrete, timestamped, verbatim memory.
Diverges: encoding is lossless and explicit — not gated by attention or reshaped by expectation.
Storage substratethe engram
In the brainDistributed synaptic weights across neocortex — robust, but not directly inspectable or addressable.
In braineonPlain human-readable files on disk. The memory is the artifact — addressable, portable, versioned.
Improves: fully inspectable and exportable; no representational drift over time.
Indexinghippocampal binding
In the brainThe hippocampus indexes cortical patterns so a partial cue can reach the whole.
In braineonA derived index maps cues to the exact file and section that answers.
Diverges: the index is disposable — rebuilt from the files on demand. Yours is load-bearing.
ConsolidationHC → neocortex
In the brainSystems consolidation over sleep transfers traces to cortex; synaptic LTP stabilises them.
In braineonDerived state is rebuilt deterministically from the files — instant, repeatable, no offline phase.
Improves: consolidation is on-demand and lossless — no schema distortion, no interference.
Retrievalrecall vs recognition
In the brainCA3 pattern completion fills a partial cue; dentate pattern separation keeps similar memories distinct.
In braineonExact recall is separation-dominant — one clean trace. The meaning-aware arm behaves like completion — it fills the gist.
Diverges: retrieval is read-only. It never silently edits the memory it touches.
Metamemoryfeeling-of-knowing
In the brainPrefrontal monitoring yields a confidence signal that is often miscalibrated.
In braineonEvery answer carries an explicit confidence and the signals that produced it; below a floor it abstains.
Improves: confidence is measured and shown — not a felt sense you have to trust.
Confabulationsource monitoring
In the brainThe system fills gaps with plausible detail and misattributes sources — sharply so after frontal injury.
In braineonIt returns the source or an honest “I don’t know.” It cannot fabricate a memory it does not hold.
Improves: perfect source monitoring; zero confabulation, by construction.
ForgettingEbbinghaus decay
In the brainRetention falls on a power-law curve; interference and retrieval-induced forgetting reshape what remains.
In braineonDecay is a tunable policy — strengthen what’s used, let the trivial fade, or retain losslessly.
Improves: forgetting is deliberate, and true erasure on demand is possible — brains cannot reliably delete.
Reconsolidationlabile-on-recall
In the brainRecalling a memory makes it briefly editable — and vulnerable to distortion before it re-stabilises.
In braineonRecall never mutates the trace. Corrections are explicit, versioned, and keep their full history.
Improves: no silent rewrite — every edit has provenance you can audit.
PlasticityHebbian LTP / LTD
In the brainCo-active synapses strengthen; disuse weakens them — learning as weight change.
In braineonPrioritisation weights shift with what’s actually retrieved; the memory web reflects real use.
Diverges: every weight change is logged and reversible — plasticity you can replay.
Four that matter most

Where the analogy earns its keep

Confabulation

The gap it refuses to fill

Human memory is generative: when the trace is thin, cortex supplies a coherent story and the source tag is lost. Useful for gist; corrosive for facts.

braineon draws the opposite line. Below its confidence floor it abstains and hands back the nearest real candidates — the machine equivalent of intact source monitoring.

Forgetting curve

Decay as a dial, not a fate

Ebbinghaus made forgetting a curve; interference makes it lossy and involuntary. You cannot choose what fades.

Here, the same power-law shape is a setting. Salience strengthens on use; the trivial can decay; and when policy demands it, a memory is erased cleanly and verifiably.

Separation & completion

Two retrieval regimes, on purpose

The dentate gyrus separates near-identical episodes; CA3 completes a partial cue. The brain arbitrates between them implicitly.

braineon exposes the arbitration: exact recall for separation, meaning-aware recall for completion, with a confidence route deciding which answers — and telling you which did.

Consolidation

Cortex, without the night shift

Systems consolidation needs offline replay; that’s partly why sleep loss degrades memory and why each replay risks reshaping the trace.

braineon’s “consolidation” is a deterministic rebuild from the source files. It can run a thousand times and return byte-identical state — replay without the reshaping.

Where it improves on biology

The affordances a brain can’t offer

Perfect provenanceEvery recall names its source. No feeling-of-knowing — a receipt.
Zero confabulationIt abstains rather than invent. Failure mode: a miss, never a fabrication.
Deterministic recallSame cue, same store, same answer — indefinitely. No state-dependent variance.
Retention is optional-losslessKeep everything verbatim, or decay by policy. You choose the curve.
True erasure on demandA memory can be removed cleanly and provably — something biological deletion never guarantees.
No interference distortionNew memories don’t overwrite or blur the old; retrieval doesn’t weaken neighbours.
Where biology still wins

The honest open questions

A memory that never distorts is not the same as a mind. These are the gaps — and our live research directions.

?
One-shot schema integrationBrains fold a single new fact into a rich existing schema instantly. Our integration is shallower.
?
Embodied, multimodal groundingHuman memory is sensory, affective, spatial. Ours is symbolic — for now.
?
Remote association & creativityThe generative gap-filling we engineer against is also the seat of insight and analogy.
?
Energy & graceful degradation~20 watts, and it fails soft. We fail loud, on purpose — but at a cost.
Research access

Work on the mapping
with us.

The full construct mapping, the abstention and decay models, and the benchmark methodology — cited, honestly qualified, and open to read right now. If you study human memory, we’d like your scrutiny.

Read the research brief →

or get new findings as we publish them

for labs & memory researchers
read against the human brain
an analogy, offered rigorouslyprivacy · © 2026 braineon