# Six priority GitHub builds — source inspection

**Owner:** Janady13 / Joseph Anady  
**Local inspection date:** October 14, 2026 (America/Chicago)  
**Method:** Authenticated, read-only GitHub file inspection. Main-branch heads were resolved to immutable commit identifiers before the selected source reads.

## Executive assessment

This collection contains native numerical-engine source, small mathematical kernels, orchestration and information-collection tools, and tracked historical data. It is substantially more than a set of dashboards. The six repositories are not six interchangeable copies of one verified system. Different versions mix substantive computational code with proxies, placeholders, telemetry generation and incomplete connections.

No application was launched, no build or repository test was run, no repository changed, and no database payload integrity was verified. These notes are an inspection record, not a repository backup or a scientific validation.

## 1. complete-agi-core — orchestration, collection and historical data

Its top-level trees preserve `thataiguy`, `seferim`, and `unified` as separate implementations. The inspected unified scripts provide:

- An orchestrator that persists scalar status JSON and counts ticks.
- Collection functions for external reference and research services.
- Category/topic/source selection and JSON append persistence with source, category, importance and timestamp.
- A federation routine that reads shared iCloud state files and aggregates locally saved metrics.

The orchestrator explicitly simulates generator contributions using random increments. It boosts a variable named consciousness toward a target and uses a fixed Pantheon average of 0.8 in the inspected calculation. The `8000` node, `128` neural-agent and related counts are declarations and saved metadata in this file; they do not establish correspondingly allocated computation there.

The learner's eight strategy labels are randomly selected, then logged and saved. The inspected cycle does not execute eight distinct reasoning/learning algorithms. It does implement information collection and persistence. The federation receive function reads files from shared storage; it does not perform the SSH/HTTP network exchange described as a future implementation in its comments.

Important tracked recovery targets from the Git tree:

| Path | Git-reported bytes | Evidence level |
|---|---:|---|
| seferim/federation/remote_memory.db | 7,000,064 | Metadata; payload not verified |
| seferim/federation/shared_memory.db | 6,897,664 | Metadata; payload not verified |
| seferim/data/learned_items.json | 1,384,902 | Metadata; content fetch returned no text |

The empty connector response for learned_items does not establish that the remote file is empty. `seferim/g16_state/epona.json` was readable and contains `.5`, an invalid strict JSON number. Preserve the original and create any repaired derivative separately.

**Use:** historical lineage, collected-data recovery and orchestration source. Separate useful collection behavior from generated consciousness telemetry.

## 2. AI_Brain — native numerical and ingestion implementation

`the_tired_agi/core` contains native C/C++ and Verilog source, headers and CMake configuration. The runtime source processes active agents by computing a Helix state, computing an FGTP energy field, applying an embedding update, and updating fitness and tick metadata. It includes queue draining, growth checks and timing/statistics paths.

The ingestion implementation supports file-extension/size checks, binary detection, memory-mapped reading, chunking and parallel processing. Its embedding generator produces deterministic hash-seeded numerical features and mixes in byte entropy and chunk size. That is a real feature-generation algorithm, but it does not establish semantic understanding or a trained text encoder.

Helix calculates its named phi quantity as `1 / (1 + abs(x - 1))` where x is derived from the mean embedding. That is an explicit scalar proxy, not independent evidence of consciousness.

Concrete repair/review findings:

- `src/blake3.c` begins with a GitHub HTML document rather than C. The inspected CMake section does not establish that every build target uses this file, so it is a verified source defect, not a demonstrated failure of every target.
- CMake enables AVX2 and adds `-mavx2 -mfma` by default under GNU/Clang in the inspected block. Architecture portability needs review before an Apple Silicon build.
- `process_ingestion()` evaluates `pop_embedding` before its 100-item limit test, which can consume an additional queue item when that limit is reached. This is a static code-path concern, not a measured runtime result.
- Scheduling, allocation, the full ingestion-to-runtime connection, actual throughput and persistent-state restoration remain untested.

**Use:** strongest candidate in these six for native-engine engineering work. Preserve the source and validate it in an isolated build, without assuming that a filename or target frequency proves production behavior.

## 3. ai-brain-core — older supervision and recovery history

The Python manager auto-discovers `sub_brains/*/launch.sh`, supervises processes and exposes a local control API. The inspected snapshot's sub-brain tree contains `browser_healer`, `hypothalamus` and `watchdog` directories.

This manager is not passive: it changes shell-script permissions, normalizes Python files and creates compiled bytecode. It defaults to a Desktop path and permits API requests when its optional token is absent. Do not run it against original recovery evidence.

The embedded `.git_backup_1755669199` contains a Git index, HEAD and loose object files. That makes it a concrete historical recovery lead, although its object closure and uniqueness versus current history have not been validated.

The database directory preserves this exact-name set:

| Path | Bytes |
|---|---:|
| db/hypothalumus.db | 4,096 |
| db/hypothalumus.db-wal | 24,752 |
| db/hypothalumus.db-shm | 32,768 |

Keep the three files together and retain their original spelling and commit. Their presence does not prove a consistent database snapshot; inspect disposable copies rather than the originals.

**Use:** process-control recovery, old environment/path provenance and potentially additional historical versions.

## 4. rhiannon — compact deterministic optimizer and extension interface

The C++ engine initializes parameters with fixed seed 42, computes sigmoid outputs, performs squared-error gradient steps and an additional target-directed correction, and reports RMS error. Its facade sends output/error to attached ANI interfaces.

This snapshot includes two GoogleTest cases: convergence toward a constant target and determinism for identical input. They were inspected, not run. The reverse step is a target-correction proxy rather than an exact inverse or undo-training implementation. SIMD is a no-op; Metal delegates to it. This repository should not be conflated with the separate Rhiannon music implementations solely because of its name.

**Use:** a small, understandable component suited to isolated validation and comparison against the intended equations.

## 5. seferim-agi-g16 — C++ G16 state kernel and surrounding scaffold

The core implements sixteen scalar family transforms, weighted updates, inertia, stability gating, normalization and an Omega objective composed from surprise, uncertainty, value and a rate penalty. Applications, bridge code and language-binding directories are recorded in the repository; their presence does not establish that every binding is complete.

Several named families are explicitly proxies. For example, the mutual-information family averages a state scalar with a sigmoid of dx_norm rather than computing mutual information from a probability distribution. Preserve this implementation but do not treat the labels as validated general algorithms.

The Go memory indexer catalogs paths, sizes, modified times, hashes, extensions and opening lines. Its FamilyTag field is a placeholder. It defaults the scan root to HOME and creates/overwrites its output file. It also assigns `err = filepath.WalkDir(...)` without an enclosing declaration of err in the read main function. That is a static compile-time defect; no compilation was performed here.

**Use:** inspectable G16 numerical kernel and a development scaffold, not a verified distributed cognitive runtime.

## 6. seferim-core — broad JavaScript mathematical reference implementation

The inspected module contains:

| Component | Implemented behavior |
|---|---|
| MetaState | Sixteen-family update, inertia, stability gating and Omega |
| DNAMemory | Learn a sixteen-dimensional signature and return a numeric query score |
| Gate5000 | A 5,000-entry binary array with phase-driven XOR updates |
| NeuralBrain | 88 scalar-state agents with 44-value hidden windows by default |
| ConsciousnessEngine | Generated 174+200-element field |
| LatticeEngine | Sixteen 16-dimensional strands, fixed kernel transforms and a Hebbian accumulator |
| Federation | Numerical coupling of local signals using vector coherence |

The module is compact and accompanied by formula/range/shape checks. Test-source inspection is not a passing test run or a performance benchmark.

Important behavioral qualifications:

- DNAMemory.query returns a score, not an original text document or decoded lossless memory.
- ConsciousnessEngine.evaluate resets its recurrence and accepts no inputs, so the same generated field results on every call. Array lengths 174 and 200 do not establish 374 independent experimentally validated equations.
- NeuralBrain synchrony reads fixed phase values that step does not update, so it is not measuring changing synchronization between evolving agent states.
- LatticeEngine changes W but its state update never reads W. The stored Hebbian accumulator therefore does not affect future dynamics through this implementation.
- The federation class is an in-memory numerical coupling model, not network transport.
- The memory test section claims a learned-versus-unrelated comparison in a comment but its assertions do not actually perform that comparison.

**Use:** the clearest compact map of several mathematical component families, with explicit behavior that can be independently tested.

## Cross-build compatibility finding

The JavaScript MetaState computes its normalizer after inertia and stability gating. The inspected C++ G16 core calculates the normalizer before those operations. The function families look related but these update orders are not equivalent in general. No automatic state/checkpoint interchange or behavior equivalence should be assumed. Keep both original versions.

## Suggested preservation and validation order

Preserve all six pinned snapshots and histories before repairs. Give additional attention to the tracked federation data, database sidecars and embedded Git history. For validation, begin with the small Rhiannon and JavaScript numerical components, then compare C++/JavaScript trajectories under identical inputs. Repair native build inputs and validate AI_Brain in isolation before reconnecting orchestration, automated ingestion or federation.

Keep kernel source, collected documents, scalar telemetry, saved checkpoints and scientific measurements as separately labeled evidence types. This inspection does not close the recovery gaps for the latest MEGAMIND/Pantheon deployment, the complete learned state, or the manuscript's measurement provenance.

## Immutable source references

### complete-agi-core

Commit: `223c58f8857eb60c4d020e6d6feacdbfb482d71d`

- [unified/core/unified-orchestrator.sh](https://github.com/Janady13/complete-agi-core/blob/223c58f8857eb60c4d020e6d6feacdbfb482d71d/unified/core/unified-orchestrator.sh)
- [unified/knowledge/autonomous-learner.sh](https://github.com/Janady13/complete-agi-core/blob/223c58f8857eb60c4d020e6d6feacdbfb482d71d/unified/knowledge/autonomous-learner.sh)
- [unified/federation/federation-mesh.sh](https://github.com/Janady13/complete-agi-core/blob/223c58f8857eb60c4d020e6d6feacdbfb482d71d/unified/federation/federation-mesh.sh)
- [seferim/g16_state/epona.json](https://github.com/Janady13/complete-agi-core/blob/223c58f8857eb60c4d020e6d6feacdbfb482d71d/seferim/g16_state/epona.json)

### AI_Brain

Commit: `d6c309d3fa439714ca46b9d3bb94caf478ea5376`

- [the_tired_agi/core/src/dna_core_real.cpp](https://github.com/Janady13/AI_Brain/blob/d6c309d3fa439714ca46b9d3bb94caf478ea5376/the_tired_agi/core/src/dna_core_real.cpp)
- [the_tired_agi/core/src/ingest_pipeline.cpp](https://github.com/Janady13/AI_Brain/blob/d6c309d3fa439714ca46b9d3bb94caf478ea5376/the_tired_agi/core/src/ingest_pipeline.cpp)
- [the_tired_agi/core/src/helix.cpp](https://github.com/Janady13/AI_Brain/blob/d6c309d3fa439714ca46b9d3bb94caf478ea5376/the_tired_agi/core/src/helix.cpp)
- [the_tired_agi/core/src/blake3.c](https://github.com/Janady13/AI_Brain/blob/d6c309d3fa439714ca46b9d3bb94caf478ea5376/the_tired_agi/core/src/blake3.c)
- [the_tired_agi/core/CMakeLists.txt](https://github.com/Janady13/AI_Brain/blob/d6c309d3fa439714ca46b9d3bb94caf478ea5376/the_tired_agi/core/CMakeLists.txt)

### ai-brain-core

Commit: `f7275582f13ada80bfb3b1008089800ca26f4a23`

- [manager.py](https://github.com/Janady13/ai-brain-core/blob/f7275582f13ada80bfb3b1008089800ca26f4a23/manager.py)

### rhiannon

Commit: `df358362a21fdde4372dc6239573622aafe4835f`

- [src/agents/two_way_engine.cpp](https://github.com/Janady13/rhiannon/blob/df358362a21fdde4372dc6239573622aafe4835f/src/agents/two_way_engine.cpp)
- [include/agents/rhiannon.hpp](https://github.com/Janady13/rhiannon/blob/df358362a21fdde4372dc6239573622aafe4835f/include/agents/rhiannon.hpp)
- [tests/rhiannon_tests.cpp](https://github.com/Janady13/rhiannon/blob/df358362a21fdde4372dc6239573622aafe4835f/tests/rhiannon_tests.cpp)

### seferim-agi-g16

Commit: `25f281b08b2f786f5a235b5abebfc39bf44d1044`

- [src/common/g16_core.cpp](https://github.com/Janady13/seferim-agi-g16/blob/25f281b08b2f786f5a235b5abebfc39bf44d1044/src/common/g16_core.cpp)
- [tools/memory_indexer/main.go](https://github.com/Janady13/seferim-agi-g16/blob/25f281b08b2f786f5a235b5abebfc39bf44d1044/tools/memory_indexer/main.go)

### seferim-core

Commit: `1f05cd5640edf0a5fe93dc276a12a48db84fc702`

- [src/seferim-core.js](https://github.com/Janady13/seferim-core/blob/1f05cd5640edf0a5fe93dc276a12a48db84fc702/src/seferim-core.js)
- [test/run.js](https://github.com/Janady13/seferim-core/blob/1f05cd5640edf0a5fe93dc276a12a48db84fc702/test/run.js)

These are links to the inspected repositories, not an assertion that their full contents are included in this report. Private source links require authorized GitHub access.