Technical Manual — 432 Atlas, Waveform Rosetta & Geometric Fiber Rosetta

Scientific posture. The reference frame is chosen. Its consequences are derived and replayed. Its external utility is tested. The ontology remains open.

W_i+(t) = M_i(t) + R_i(t)
W_i-(t) = M_i(t) - R_i(t) = 2M_i(t) - W_i+(t)
W_i(t;s) = M_i(t) + (2s - 1)R_i(t),  s in [0,1]
[W_i+(t) + W_i-(t)] / 2 = M_i(t)

Do not globalize s. The meaningful identity is (i, s). The same numeric value in two different cells does not establish a cross-cell ranking or shared human quality.

Xi = (m, i, s, x, ell, u, rho, Sigma, pi, a)

m      native modality or adapter
i      Atlas address or other bounded reference address
s      local expression coordinate in [0,1]
x      optional domain-typed geometric coordinate
ell    finite observation or refinement depth
u      (E_cpl, T_tr, E_s, lambda, mu) in [0,1]^5
rho    unexplained or off-basis residual
Sigma  uncertainty or covariance
pi     provenance and transformation lineage
a      aperture, consent, retention, and governance posture

Xi = (m, i, s, x, ell, u, rho, Sigma, pi, a)

m      native modality or adapter
i      bounded reference address
s      cell-local expression in [0,1]
x      domain-typed geometric coordinate
ell    finite observation or refinement depth
u      typed GUFT vector
rho    unexplained or off-basis residual
Sigma  uncertainty or covariance
pi     provenance and transform lineage
a      aperture, consent, retention, and governance posture

Public thesis. GUFT does not tell an AI what reality is. It gives the AI a governed language for learning how observed domains relate, where translations fail, and how vocabulary may grow after repeated residual evidence.

Grammar of not knowing. The projector may emit MAPPED, PARTIALLY_MAPPED, MULTI_CELL, CONTESTED, AMBIGUOUS, INSUFFICIENT_EVIDENCE, OUT_OF_DISTRIBUTION, NO_ADEQUATE_MAPPING, AXIS_DEGENERATE, or NEW_PATTERN_CANDIDATE. No code and decimal becomes truth merely by existing.

Core boundary. The exact CoherenceLattice core already contains a generic five-axis synthetic reference-waveform encoder. The optional Rosetta module adds the cell-specific master-preserving and geometric methods, then emits uvlm.coherence.observation.v1. The sealed core is not patched.

Set-ExecutionPolicy -Scope Process Bypass
.\INSTALL_WINDOWS.ps1

bash install.sh

Network note. pip install -r requirements-runtime.txt may contact the configured package index for jsonschema>=4.23,<5. Pre-stage that dependency before installation when the environment must remain offline.

.\RUN_DEMO_WINDOWS.ps1

bash run_demo.sh

A successful summary includes "valid": true and "authority_effect": "NONE". “Valid” means the declared integration contract passed. It is not scientific confirmation or human authorization.

.\.venv\Scripts\python.exe .\tools\verify_release.py

.venv/bin/python tools/verify_release.py

Expected high-level result: artifacts_verified: 6, source_closures_verified: 4, public_release: true, and valid: true.

.\.venv\Scripts\uvlm-atlas-432.exe verify-registry

.venv/bin/uvlm-atlas-432 verify-registry

uvlm-atlas-432 record F1-01

uvlm-coherence map /path/to/grounding_bundle   --task "Map source-grounded care and boundary-restoration patterns"   --dictionary-plugin atlas_432

Interpretation. Atlas returns at most eight deterministic competing readings. The relation score is not a probability, and the output is not a person classification.

uvlm-fiber-rosetta bridge modules/UVLM_Recursive_Geometric_Fiber_Rosetta_Module_v0.1.0-alpha.3/examples/example_bridge_input.json > bridge_receipt.json

python -c "import json; p=json.load(open('bridge_receipt.json',encoding='utf-8')); json.dump(p['observation'],open('observation.json','w',encoding='utf-8'),indent=2,sort_keys=True)"

uvlm-coherence validate-packet observation.json
uvlm-coherence project observation.json

Use the virtual-environment executable paths on systems where the commands are not globally available.

Separate research lineage. The v0.12.0-RC1 Universal Pattern Bridge candidate used to refine these explanations is not part of this exact public release. Its no-render, no-publication, no-open-source-release posture remains controlling for that candidate. Do not upload it with the eight v1.3.0 public files.

1. Record the model name, version, settings, prompt, and exact source material.
2. Quarantine the raw model output as a candidate.
3. Build authenticated source segments or typed Rosetta inputs independently.
4. Run Atlas or Rosetta as a separate bounded process.
5. Preserve refusals, unsupported claims, residuals, and uncertainty.
6. Record human edits and the final human decision.

• Define a narrow purpose and changed-file surface.
• Add expected and adversarial tests.
• Update source manifests, rights ledgers, notices, and provenance.
• Do not add person classification, diagnosis, spiritual or moral ranking, model-provider calls, silent network use, memory writes, autonomous publication, or deployment authority.
• Build deterministically where possible and obtain a new exact-hash review before public release.