Architecture of Knowing

AIEP protocol flow diagram — Figure 1. AIEP protocol flow — Publishers through to Governance.

AIEP is not only a file format or a web convention. It is an attempt to respect how knowledge actually grows.

Human knowledge evolves through:

hypothesis
criticism
dissent
evidence
revision

The web, however, often collapses these stages into a single outcome: a page that looks authoritative. AIEP separates the stages again by allowing knowledge to exist in multiple states, and by linking instructions to evidence.

The jigsaw metaphor

AIEP can be understood as a jigsaw puzzle with billions of pieces.

Each published artefact is a piece. Each piece carries provenance: who published it, when, and what evidence it rests on.

Consensus is the picture most pieces currently support. That picture is not complete — it is the best configuration possible with the pieces available today.

Outliers are pieces that do not yet fit the current picture. A different shape. A different colour in the wrong place. In most systems, these pieces are discarded — or never published at all.

AIEP does not discard pieces. It stores every piece, records where it came from, and preserves it in a retrievable state. When the surrounding picture changes — when new pieces arrive and alter the configuration — the outlier can be retrieved, re-examined, and fitted in where it could not fit before.

This is how knowledge actually grows. Plate tectonics was an outlier for decades. Heliocentrism was a radical outlier. Germ theory was dismissed. The pieces existed. The surrounding picture had not changed enough to accomodate them yet.

A system that discards outliers cannot recover from that error. A system that preserves them — with provenance intact — can.

The dissent loop

The dissent loop is the core mechanism by which AIEP allows knowledge to evolve without losing history.

The cycle

Stage	What happens
Claim	An artefact is published — an instruction, a finding, a position, linked to its evidence
Challenge	Counter-evidence emerges — a contradicting artefact is published by the same or a different issuer
Dissent	The original claim is contested — both views are recorded, neither deleted
Archive	The minority view is classified as an outlier or radical outlier — stored, not elevated
New evidence	Time passes. New evidence appears that was not previously available
Recall	The archived outlier is retrieved and re-evaluated against the new evidence
Re-evaluation	The outlier may be elevated to consensus — or the former consensus may be demoted
New cycle	The re-evaluated position becomes the new claim — and the loop continues

This loop does not resolve to a final state. It is a continuous process. AIEP does not promise to know the truth. It promises to record the evidence so the question can be asked again later.

Why the loop matters

Without the dissent loop, knowledge systems converge prematurely. The loudest claim wins. The minority view is erased. Future correction becomes expensive or impossible because the evidence chain no longer exists.

With the dissent loop:

A claim retracted five years later is still traceable — you can see what was claimed, when, and on what evidence
A position that was wrong is demoted, not deleted — future systems can understand why the correction happened
Outliers from decades ago can be re-surfaced if new evidence makes them relevant — the pieces were never thrown away

What triggers a recall

Recall is triggered by new evidence arriving that matches the domain of a stored outlier. An AIEP-compliant system can query for outliers in a domain and surface them when new evidence has arrived that warrants re-examination. This is not automatic elevation — it is a prompt to the retrieval agent or human reviewer that a stored piece may now fit.

The mechanical pathway for recall runs through two layers:

Plausibility registry update — the PlausibilityScore for an archived claim-type is updated in the versioned safety registry, via signed assessments from authorised authorities aggregated under threshold signatures. This changes what is eligible to execute.
Deterministic context reconstruction (P22) — when a previously archived branch is recalled, its context is reconstructed exclusively from admissible artefacts in the ContextRegistry, producing a ContextReconstructionHash that is bit-identical across every node in a distributed system. Recall is not approximate re-assembly — it is cryptographically verifiable reconstruction.

See: Recall for the full mechanism · Plausibility Matrix for the registry-bound gate

Dissent is not misinformation

A common misreading: if an outlier is stored, does that mean AIEP validates misinformation?

No. AIEP classifies, not endorses. An outlier is labelled an outlier — it carries its evidence record, its issuer identity, its date, and its status. A retrieval agent sees the label. The label is not neutral — it is an explicit signal that the current evidence does not support elevation. But the record exists so the question can be asked again.

How evidence flows

Stage	Component	Outcome
Input	New evidence published	Enters the AIEP Mirror surface
Discovery	AIEP Mirror (distributed sources)	Evidence collected and indexed
Retrieval	Evidence-backed knowledge retrieval	Evidence evaluated against existing records
Consensus	Supported evidence	Elevated to consensus — used with confidence
Outliers	Contradicting or unresolved evidence	Stored, not discarded — recalled when new evidence fits

Why this matters for AI

If AI is to become safer and more reliable, it needs a way to retrieve knowledge from sources that publish evidence, provenance, and governance signals. AIEP is a proposal for that layer.

It is not “bigger models”. It is better structure.

An AI trained on a snapshot of the web inherits whatever the web forgot to preserve. An AI that retrieves from AIEP-structured sources inherits the dissent loop — it can ask what the outlier position was, whether it has been re-evaluated, and whether the current consensus is stable or contested.

The future of information retrieval is not search — it is evidence-backed knowledge retrieval.