Research milestones

Every gate, in the open.

What each milestone has shown, what it has not, and the next result that moves it.

Last reviewed 2026-10-20.

Program

Machine Cognition

Building AI designed to answer only from knowledge it can point to, and to say "I don't know" rather than guess.

Governed request handling

Internally tested

Requests are handled through a governed path, with checks at defined points.

Demonstrated

Governed request handling passes internal automated tests.

Not demonstrated

Complete governance. It has not been demonstrated, and it is not presented as complete.

Next observable result: Complete governance demonstrated and accepted under internal review.
Evidence: Internal evidence only
Limitations: Internal tests only. Not available to the public.
Last reviewed: 2026-10-20

Canonical knowledge record handling

Internally tested

Knowledge is held as explicit records that each carry a record of their stated source.

Demonstrated

Knowledge records keep their stated source under internal automated tests.

Not demonstrated

Final rules for how stated sources are trusted.
Operation at public scale.
Independent review.

Next observable result: The same behaviour confirmed under a broader internal test set.
Evidence: Internal evidence only
Limitations: Internal tests only.
Last reviewed: 2026-10-20

Frozen knowledge generations

Internally tested

Knowledge is captured in fixed, versioned generations, and each request is handled against one fixed generation.

Demonstrated

Versioned knowledge snapshots pass internal automated tests.

Not demonstrated

Full verification of the newest form of these generations.

Next observable result: The newest form of knowledge generations verified internally.
Evidence: Internal evidence only
Limitations: Internal tests only.
Last reviewed: 2026-10-20

Explicit operator teaching

Internally tested

New knowledge is added through explicit operator steps, never from conversation.

Demonstrated

Explicit operator teaching passes internal automated tests.

Not demonstrated

Conversational teaching.
Verification that taught knowledge is understood rather than merely stored.
Final rules for which knowledge sources are trusted.

Next observable result: A defined acceptance test for taught knowledge, passed internally.
Evidence: Internal evidence only
Limitations: The public cannot teach the system.
Last reviewed: 2026-10-20

Task-scoped conceptual representation

Implemented, unverified

Research into representing the concepts involved in each individual request. This work is under development.

Demonstrated

A first version has been implemented. It is not yet verified.

Not demonstrated

Accepted verification of that first version.
Relational understanding of a request.
Use of the representation to answer questions.

Next observable result: The first version verified and accepted internally.
Evidence: Internal evidence only
Limitations: Recent work. Not verified.
Last reviewed: 2026-10-20

Structural input relationships

Not yet demonstrated

Identifying how the parts of a request relate to one another.

Demonstrated

Not demonstrated. This is the next stage of the input work.

Not demonstrated

Relationships extracted from a request and used by the system.

Next observable result: Relationships within a request identified and checked by tests.
Evidence: No evidence claimed
Limitations: Planned.
Last reviewed: 2026-10-20

Canonical knowledge correspondence

Not yet demonstrated

Matching what a request refers to against established knowledge, and showing where no justified match exists.

Demonstrated

Not demonstrated. This is a stated objective.

Not demonstrated

Correspondence between a request and established knowledge.

Next observable result: A request whose referenced concepts are matched, or reported as unmatched, under test.
Evidence: No evidence claimed
Limitations: Depends on the structural input relationships milestone.
Last reviewed: 2026-10-20

Precise missing-knowledge discovery

Not yet demonstrated

Determining exactly which knowledge a task needs that the system does not yet have, instead of guessing.

Demonstrated

In tested cases, the system states plainly when it does not know about a concept.

Not demonstrated

A precise, structured account of what is missing and why the task needs it.

Next observable result: Structured missing-knowledge reports produced for test tasks.
Evidence: Internal evidence only
Limitations: Only basic missing-knowledge reporting exists today.
Last reviewed: 2026-10-20

Task-dependent dependency completion

Research concept

Working out the full set of supporting knowledge a task depends on.

Demonstrated

Not demonstrated. This is a research objective.

Not demonstrated

Any working implementation.

Next observable result: A first implementation demonstrated on a small test task.
Evidence: No evidence claimed
Limitations: Research stage.
Last reviewed: 2026-10-20

Governed continuing knowledge acquisition

Active development

Adding knowledge over time without losing the record of where each piece came from and who approved it.

Demonstrated

Distribution of knowledge packages is internally tested.

Not demonstrated

Continuing acquisition of new knowledge under complete governance.

Next observable result: Governed acquisition demonstrated end to end under internal tests.
Evidence: Internal evidence only
Limitations: The system does not learn from conversations.
Last reviewed: 2026-10-20

Multimodal conceptual integration

Research concept

Connecting concepts derived from text, images and sound into one consistent body of knowledge.

Demonstrated

Not demonstrated. This is a long-term objective.

Not demonstrated

Any integration between the cognition research and the visual or audio research.

Next observable result: A defined, testable exchange between one sensory program and the cognition program.
Evidence: No evidence claimed
Limitations: The visual and audio programs are at early stages.
Last reviewed: 2026-10-20

Independent evaluation and production qualification

Not yet demonstrated

Evaluation by people outside ByteLite LLC, followed by qualification for real use.

Demonstrated

Not demonstrated.

Not demonstrated

Any independent evaluation.
Production qualification.
Public availability.

Next observable result: A documented evaluation plan agreed with an outside evaluator.
Evidence: No evidence claimed
Limitations: Public-facing use stays disabled until governance is independently verified.
Last reviewed: 2026-10-20

Program

Data Representation

Representing data so it rebuilds exactly, byte for byte, every time, with every byte accounted for.

Exact reconstruction

Internally tested

The original data is rebuilt byte for byte from its representation.

Demonstrated

Every case in the internal test set is decoded and compared with the original byte for byte before the representation is accepted.

Not demonstrated

Exact reconstruction verified by anyone outside ByteLite LLC.
Detection of corruption that happens to a representation after it was created.

Next observable result: The same check repeated on a broad public test collection.
Evidence: Internal evidence only
Limitations: Internal test set only. Exactness is checked when a representation is created.
Last reviewed: 2026-10-20

Complete artifact accounting

Implemented, unverified

Every size figure should count the complete representation, including everything needed to rebuild the original.

Demonstrated

For the current single-round format, reported sizes are measured from the complete artifact file.

Not demonstrated

Complete accounting for every output form.
Independent audit of the accounting.

Next observable result: Accounting rules published in a form an outside reviewer can check.
Evidence: Internal evidence only
Limitations: Internal evidence only.
Last reviewed: 2026-10-20

Deterministic operation

Internally tested

The same input and the same software always produce the same representation.

Demonstrated

Repeated runs on one machine produce identical output in internal tests.

Not demonstrated

Identical output across separate machines.
Independent confirmation.

Next observable result: Determinism confirmed across separate machines by an outside party.
Evidence: Internal evidence only
Limitations: Internal evidence only.
Last reviewed: 2026-10-20

Measured net size reduction

Not yet demonstrated

A complete, self-contained representation that is smaller than the original for real inputs, measured and reported per input.

Demonstrated

Not demonstrated. The current development representation is larger than its input.

Not demonstrated

Net reduction on any input class.
Any compression ratio.

Next observable result: A measured, complete representation smaller than its input on a declared test collection.
Evidence: No evidence claimed
Limitations: No reduction figure is claimed anywhere on this site.
Last reviewed: 2026-10-20

Independent evaluation and production qualification

Not yet demonstrated

Reproduction of results by an outside party, followed by qualification for real workloads.

Demonstrated

Not demonstrated.

Not demonstrated

Independent reproduction.
Production qualification.

Next observable result: An agreed evaluation protocol with an outside reviewer.
Evidence: No evidence claimed
Limitations: Depends on measured net reduction first.
Last reviewed: 2026-10-20

Program

Visual and Audio

Describing images and sound in exact, repeatable structures that other systems can rely on.

Deterministic image-structure components

Internally tested

Software components that describe the structure of an image in a repeatable, deterministic way.

Demonstrated

Core components pass internal automated tests on synthetic test images.
A deterministic image stylization component passes internal tests.

Not demonstrated

Visual understanding of any kind.
Results on real-world image collections.

Next observable result: The remaining defined behaviours implemented and tested.
Evidence: Internal evidence only
Limitations: Internal tests on synthetic inputs only.
Last reviewed: 2026-10-20

Audio and speech research architecture

Active development

The defined structure of a future audio and speech research system.

Demonstrated

An architectural skeleton exists and passes internal structural tests.

Not demonstrated

Speech recognition or comprehension of any kind.
Audio analysis results.

Next observable result: The remaining undefined parts of the architecture specified.
Evidence: Internal evidence only
Limitations: Early stage. Most behaviour is not yet defined.
Last reviewed: 2026-10-20

Definitions

What each status means.

Internal testing is never presented as independent verification.

Status: Research concept: A defined research question or intended capability. Nothing has been built for it yet.
Status: Active development: Engineering work is under way. Any internal results so far are listed with the milestone; the milestone itself is not complete.
Status: Implemented, unverified: Software exists in the company source tree, but its required checks have not yet been completed and accepted.
Status: Internally tested: Automated tests written by ByteLite LLC pass for the stated scope. Nobody outside the company has checked the result.
Status: Internally validated: Tested, and the result has also been reviewed against written acceptance criteria inside the company. Still not independent.
Status: Independently verified: Someone outside ByteLite LLC has reproduced the result under conditions they controlled.
Status: Production qualified: Shown to be dependable enough to run real workloads for real users.
Status: Not yet demonstrated: An intended capability that has not been shown to work. It is a goal, not a result. Any partial internal results are listed with the milestone.
Status: Blocked / under review: Work is paused until an open engineering, governance or business decision is resolved.

Update discipline

How this page changes.

A milestone changes only through a separately approved evidence-status update. Internal engineering changes never advance a public milestone on their own.
Every milestone records what has been demonstrated and what has not. Neither list may be empty.
There are no percentage-complete figures. Progress is shown as observable gates that are either met or not met.
Internal evidence is labelled internal. Only a reproduction by someone outside ByteLite LLC may be labelled independent.