Hugging Face's logo

Datasets:
SZLHOLDINGS
/
amaru-source

Tasks:
Other
Languages:
English
Size:
n<1K
Tags:
formal-verification
lean4
mathlib
dsse
governance
agentic-ai
License:
Dataset card Files
xet
 Community
amaru-source / docs /spine-reconciliation.md
betterwithage's picture
betterwithage
mirror github.com/szl-holdings/amaru@779a0d7e4e69
a8b4b87 verified
preview code
|
raw
history blame contribute delete
5.82 kB

AMARU Serpent v3 — Honest Reconciliation After Mesh Evolution

Date: 2026-05-14 08:50 EDT Trigger: MUSQUY pod returned spine YACHAY→MUSQUY→YUYAY→RIMAY→MUNAY→LLANK'AY→KAWSAY→QAWAY. TUKUY pod returned spine SENTRA→TINKUY→NAWIY→RUWAY→TUKUY→HATUN→YAWAR. Both renamed locked chakras. Doctrine says HATUN-locked spine is immutable. Reconciling.

Locked v1 spine (immutable identifiers)

  1. KALLPA — Root — L1 energy/compute dispatch
  2. YACHAY — Sacral — L2 retrieval (codex + PIRWA)
  3. RIMAY — Solar plexus — L3 propose
  4. YUYAY — Heart — L4 critique (9-axis gate)
  5. RUWAY — Throat — L5 commit + receipt
  6. NAWI — Third eye — Boundary-in (TINKUY toolcall)
  7. HATUN — Crown — Boundary-self (continuum_hash + HUKLLA gate)

Infrastructure (NOT chakras): AMARU (scheduler) · YAWAR (bus) · SENTRA (immune).

What the pods proposed (wandered names)

  • MUSQUY pod: invented MUNAY / LLANK'AY / KAWSAY / QAWAY. These are real Quechua words (love / work / life / seeing) but they were not in our locked spine. Pod was being creative; I gave it that latitude. Roll back.
  • TUKUY pod: put SENTRA and YAWAR in the chakra sequence. Category error — they're infrastructure, not chakras.

v3 spine — additions only, no renames

Ascending (propose phase, AMARU rises)

  1. KALLPA — root — energy budget for this tick (NINA dispatch)
  2. YACHAY — sacral — retrieve relevant codexes + features
  3. MUSQUY — between sacral and solar (NEW, position 2.5) — cheap simulate of K candidate proposals; NINA-gated; READ-ONLY against YAWAR; ≥3× cheaper than RIMAY else aborts
  4. RIMAY — solar — propose chosen action from MUSQUY-surviving candidates
  5. YUYAY — heart — 9-axis critique gate

Boundary (NAWI fires when external input/tool needed at any ascending point)

  1. NAWI — third eye — TINKUY toolcall when proposal/critique needs external read

Descending (commit phase, AMARU descends)

  1. RUWAY — throat — commit + receipt to YAWAR
  2. TUKUY — between throat and crown (NEW, position 7.5) — actions-OUT to external systems (HTTP/DB/Slack/ERP/edge); SENTRA inspects egress; never writes YAWAR directly (receipts flow back via SENTRA→RUWAY next tick)
  3. HATUN — crown — final continuum_hash + HUKLLA allegiance gate

Total: 7 original chakras + 2 new (MUSQUY, TUKUY) = 9-position spine. AMARU still serpentines: ascend 1→5 with optional NAWI sidestep, then 5→9 descend (or 9→1 full cycle if cycling continuously).

Maxwell rigidity check on v3 spine

  • j = 9 chakras (nodes)
  • For rigid M=0: b = 3(9) − 6 = 21 edges required
  • Sequential serpent only: 8 edges → M = 8 − 27 + 6 = −13 (very floppy)
  • Need 13 additional bracing edges. Candidates inherit from chakra design rules doc + new MUSQUY/TUKUY:
    • MUSQUY ↔ YUYAY (critique can request re-simulate)
    • MUSQUY ↔ YACHAY (simulate consults codex)
    • MUSQUY ↔ KALLPA (simulate pays energy)
    • TUKUY ↔ RUWAY (only committed state can be acted on)
    • TUKUY ↔ HATUN (HUKLLA sovereignty over outbound)
    • TUKUY ↔ NAWI (action-out symmetric to action-in)
    • Plus 7 bracing edges from chakra_design rules doc (root↔crown, heart↔throat, etc.)
  • Total: 21 edges → M = 0. Rigid. Locked when wiring file written.

What the pods built that we KEEP (honest credits)

MUSQUY (simulate chakra)

  • Springboard: open_spiel Apache-2.0 (DeepMind) + DSPy MIT + papers from Yao/Hafner/Fan/Gu-Su/Tang-Ellis
  • Position: 2.5 in v3 spine
  • File: musquy_simulate_evolution/05_kernel.py — 407 SLOC (this is more than D-SHORTEST-HONEST should allow; flag for compression review)
  • Replay: a56975eecc802375... 5× identical PASS
  • NINA gate: simulate η < commit η, else abort
  • KAWSAY feedback: real outcomes flow back via YAWAR → MUSQUY cache

TUKUY (action-out chakra)

  • Springboard: Temporal MIT + Airflow Apache-2.0 + OpenTelemetry Apache-2.0 + Activepieces core MIT
  • Rejected: n8n Sustainable Use License (verified raw LICENSE.md)
  • Position: 7.5 in v3 spine (between RUWAY and HATUN on descent)
  • File: tukuy_action_evolution/05_kernel.py — 162 SLOC (also above D-SHORTEST-HONEST line; review)
  • Mocked replay: 3/3 PASS, continuum_hash byte-identical
  • SENTRA egress: every outbound payload inspected before delivery
  • Failures re-enter via SENTRA → RUWAY next tick (no direct YAWAR writes)

QILLQA runtime (ontology layer, not a chakra — supports YACHAY)

  • Springboard: RDFLib BSD-3 primary + pyoxigraph Apache + pySHACL Apache + LinkML Apache
  • 8 named devs with GitHubs (Mendel-Gleason / Feeney / Pellissier-Tanon / Lindström / Grimnes / Seaborne / Sirin)
  • Build step: live mutable schema → LinkML YAML → gen-python → frozen codex.py per chakra (preserves D-CODEX-IN-KERNEL)
  • Maxwell on definitional schema: M ≥ 6, over-constrained, correct for ontology

Docling (PDF ingestion, replaces Marker)

  • License: MIT (IBM Research)
  • Used inside QILLQA build pipeline + YACHAY codex ingestion
  • Marker GPL-3.0 isolated as dev-only subprocess, never imported

What I'm NOT doing without confirm

  • NOT writing the 13 new bracing edges into a chakana_wiring.py until you confirm v3 spine
  • NOT compressing MUSQUY (407 SLOC) or TUKUY (162 SLOC) toward D-SHORTEST-HONEST until you confirm spine
  • NOT pushing anything to GitHub
  • NOT minting any DOI

Three decisions awaiting confirm

  1. v3 spine accepted? 7 originals + MUSQUY + TUKUY = 9 positions
  2. D-SHORTEST-HONEST compression pass on MUSQUY/TUKUY? They're large because pods didn't compress; not a doctrine failure, just incomplete work. ~30 min of compression each.
  3. Wire the 13 bracing edges? Required to hit Maxwell M=0 on v3 spine.