Skip to content

prime-tensor-stack.md

Latest commit

 

History

History
117 lines (87 loc) · 4.58 KB

prime-tensor-stack.md

File metadata and controls

117 lines (87 loc) · 4.58 KB
Raw

The Prime-Tensor Stack — role and boundary map

This document is the cross-repository role-and-boundary map for the prime-tensor compute family. It is not a proof, and it moves no theorem, empirical, measurement, or architectural status between repositories.

2026-08-17 PTCNA construction correction

The intended PTCNA is not the currently published four-layer runtime.

The intended dependency is:

functioning conventional neural network
        ↓
UCNS audit
        ↓
source-bound relational evidence
        ↓
PTCNA construction derived from that evidence
        ↓
a0 / interdependent-lib consumption

Until UCNS performs that audit and PTCNA publishes a separately identified architecture derived from the evidence, intended PTCNA construction is BLOCKED.

The existing ptcna>=0.1.1 package remains available as a historical pre-audit experimental scaffold. This aggregator may expose/install that package, but must not represent its neural/circle/seed/core structure, ring sizes, weights, whole-string input projection, or sealed evaluation as the intended PTCNA architecture.

For the intended PTCNA language path, the primitive input object is a UCNS Unicode-character gonol. Conventional tokenizer ids, subword ids, whole-string cryptographic fingerprints, and opaque external embeddings are not substitutes for that primitive.

Historical scaffold map

The currently published ptcna package consolidates the former pcna, pcta, and pcsa repositories into one executable research scaffold. PCEA remains a separate orthogonal repository.

Module Historical scaffold layer Tensor kind Back-propagation
ptcna.neural neural neural yes in the scaffold
ptcna.circle circle auditing / timing no
ptcna.seed seed auditing / timing no
ptcna.core core auditing / timing no

The candidate-state compatibility pair previously recorded between UCNS and PTCNA remains historical construction evidence for that scaffold only. It does not satisfy the restored requirement that UCNS first audit a functioning conventional neural network.

The scaffold's frozen role-acquisition experiment also remains historical sealed evidence: ptcna.experimental.v1 scored 0.3333333333 versus 0.9444444444 for its hashed-linear fallback, falsifying both declared usefulness and superiority claims for that exact in-sample scope. This does not evaluate the not-yet-built intended PTCNA.

Authority boundaries

UCNS owns the neural-audit prerequisite, the relational evidence it produces, and the Unicode-character gonols used as intended language-input primitives.

PTCNA owns preservation of its historical scaffold and, after the audit, construction and evaluation of a new candidate derived from the exact UCNS evidence.

a0 may preserve or execute explicitly typed historical scaffold state, but a future intended PTCNA region requires a new producer identity. a0 does not inherit PTCNA architecture authority through integration.

PCEA remains orthogonal privacy/guardian infrastructure and is not promoted into a PTCNA layer by this map.

EDCM remains an external measurement authority. PTCNA/a0 consumption of an EDCM result transfers no measurement validity.

Packaging status

Member Package status Aggregator status
ptcna published (ptcna>=0.1.1) installable historical scaffold via ptcna extra
pcea published pcea extra (+ all)
ucns published ucns extra (+ all)
aimmh published as aimmh-lib aimmh extra (+ all)
zfae repository/runtime authority remains separate registry/runtime relation as declared by its consumers

The prior ptca-lib core-layer distribution remains superseded by the ptcna consolidation. That packaging fact does not promote the historical scaffold into the intended architecture.

Next integration boundary

interdependent-lib should not encode the unknown future PTCNA topology. When PTCNA eventually publishes an audit-derived candidate, integration must pin its new exact producer identity and preserve the historical package identity rather than silently treating an upgrade as continuity.

hmmm

The architecture recovered by the UCNS neural audit, the future PTCNA package identity/API, and the exact compatibility boundary between the historical scaffold and the audit-derived candidate remain unresolved. No missing detail in this aggregator is permission to assume the result.