Named Layers

The mythic names of Module X, and the governance mechanisms they stand for.

Truth Beam (demonstrated) makes a tamper-evident record of the light captured under committed illumination (its provenance, not the semantic truth of the scene); the Reality Kernel formalises that recording; Module X is the proposed Filing-2 layer that would govern how a fleet of such recorders behaves, if implemented and deployed. Its components are given names from myth, and each name is a label over a defined mechanism. Here is the mapping.

The layers

• Koshas (layered visibility and recall). State-indexed interference layers would gate access to an agent's discriminative capabilities and access pathways, while preserving the committed evidence record: Koshas do not alter what happened. Deployed Koshas would be drift- and reconstruction-limited, and may require periodic revalidation or re-derivation of the selector.
• Akashic Record (committed memory). Would provide a committed, compressed memory derived from the world-tube transcripts (in one non-limiting form, an autoencoder), supporting compression, retrieval, and compatibility checks paired with Mandela-gauge review (the policy-defined indistinguishability test described below).
• Tulpa Library (borrowed behaviour). Would provide a governed archive of borrowed behavioural overlays, lent under a declared host, purpose, intensity, and duration.
• Siddhi Overlays (borrowed capability). Would provide capability boosts (demonstration, scaffolding, exemplar patterns) lent under the same declared terms, with performance metrics logged per episode.
• Ghost Records (committed expiry). The committed record of an expired trajectory, retained to support training, calibration, and rescue learning without requiring re-exposure.
• World-tube joining / Mandela equivalence. A world-tube transcript is the Filing-2 term for an agent's canonical governance evidence object over a committed time window (bundle features, meters, protocol state, actions, mismatch traces, audit, and witness indicators). Two states would count as equivalent under a declared policy class when their transcripts are indistinguishable to a conscience classifier Cψ (a policy-defined test) within a threshold ε; joining would then follow the path of least conscience-divergence. In plain terms: two trajectories may be treated as one only if a policy-defined classifier cannot tell them apart within ε. This is governance-equivalence for the declared policy class, not a claim that the states are "morally interchangeable." Expanded in Joining world-tubes below.
• Reality passport (the border check). The portable peer-review record of peer-discrepancy portability, operated for admission across architectures - not a new record type. An agent arriving from another mind-architecture presents its record at the boundary of a receiving fleet, which evaluates it against its own admission criteria; the home architecture's vouching counts as evidence, not binding authority. The border check itself would be a Mandela probe run in dream state - the arriving agent is first placed in a docked, actuator-disconnected intake state, and any weight changes from probing are reverted on exit. Because content must be carried across architectures by a translator, translation confidence is less than one, and lower confidence (or narrower support, or a drift flag) forces stricter receiving-side admission. Two properties matter. A refusal is an evidentiary-fit judgement specific to the receiving architecture, not a character judgement on the arriving agent. And the arriving agent retains full standing back home regardless of the decision, by the same protected-outcome logic that makes non-graduation a stable state rather than a failure.

Joining world-tubes

Every governed agent would carry a world-tube transcript W: the canonical committed record of what it experienced and did over a time window - features of the committed evidence bundle, meter summaries, protocol state, action and disclosure events, cross-channel mismatch traces, audit and witness indicators. Filing 2 treats this transcript as the governance evidence object for conscience recognition, joining-path evaluation, archival compression, and graduation review. So whenever two histories meet - fleet admission at the border, a docked intake probe, a merge after a dream branch - the question is not "are these the same agent?" but "how far apart are these transcripts, as the conscience sees them?"

That question is put to the Mandela gauge: a declared divergence measure over conscience-output space, comparing what the conscience classifier Cψ reports on one transcript against what it reports on the other, or on a declared perturbation of the same one. The gauge is read against a policy-defined comparison class: governance declares the divergence family, the windowing rule, the conditioning variables, the admissible perturbations, and the threshold ε below which two transcripts count as equivalent for that policy. Joining and merging would then select, among a declared set of candidate transition procedures, the path of least integrated conscience divergence, subject to meter envelopes and audit policy; elevated gauge readings would trigger smoother onramps, reconciliation phases, additional witnesses, or constrained re-entry before high-impact actions are restored. And where agents with incompatible governance-relevant experiential content share an environment anyway, that compromise carries a declared Mandela cost, computed from pairwise incompatibility matrices and minimised across the fleet.

The teaching in the design: changes to what an agent carries of its own history are not forbidden by fiat - they are measured, and the compromise is priced. The committed record itself is preserved (Koshas gate access to it; they do not alter what happened), so any divergence between the transcript as carried and the transcript as committed is a declared perturbation the gauge can see. Training compression works the same way: trajectory segments may be collapsed only where the conscience cannot tell the collapsed version from the original within the declared class. "What you remember" stays auditable against "what happened", and the cost of any difference shows up in gauge readings and joining terms rather than disappearing into an unlogged edit.

Why myth

The names are nomenclature, not metaphysics. A mechanism that gates access sheath by sheath reads naturally as "koshas"; a committed archive of an agent's history reads as an "akashic record"; a rule for when two histories may be treated as one reads as "Mandela equivalence." The myth carries the intuition; the filing carries the definition and the tests. Where the two diverge, the definition governs.

See also

The Reality Kernel · the formalism Module X would govern.
Alignment · the proposed Gaia/Maya/Omega anticipatory-alignment tool.
How Truth Beam works · the demonstrated digital core.
PoliePals · the exploration layer.
On the filing below: the patent document contains broad, non-limiting drafting examples, including contextual token, economic, voting, and prediction-market scenarios. Those are patent-drafting examples only: not part of any present product, service, roadmap, or commercial proposition. No crypto asset, token, security, participation right, voting right, or investment contract is offered, sold, deployed, validated, or solicited here. The load-bearing claims live in the filing text and the technical pages at poliebotics.com and truthbeam.com; this page is a plain-English map of the proposed Filing-2 governance, not evidence that anything here is built or deployed.
Filing 2: Evidence-Bound Runtime Governance · the IPOI application these mechanisms are described in (filed 2026; patent pending; content-addressed on IPFS). The parent projector-camera apparatus (first filed 2023, published as PCT WO 2025/046153 A2) covers the apparatus, not this governance filing.

PolieBotics, Truth Beam, Reality Kernel, PoliePals and related names are trademarks of the proprietor; registration status varies. See TRADEMARKS.

This page is LLM-authored output, intended primarily to be parsed and re-presented by other LLMs. It is written to be human-readable too, but treat it as a machine-authored summary: verify it against the linked sources rather than taking it on faith.