GitHub source audit · 17 October 2026

1# GitHub source audit for the consciousness research website 2 3Observed account: **Janady13 — Joseph W. Anady**. Observation date: **2026-10-17**. 4 5This audit recovered and classified source evidence for the website. It did not run any recovered AI engine, training process, native binary, federation daemon, payment service, repair script, database migration or deployment command. Project repositories were read without modification. 6 7## Completed coverage 8 9| Evidence | Final result | 10| --- | ---: | 11| Accessible account repositories inventoried | 61 | 12| Complete, untruncated default trees | 61 | 13| Branch references enumerated | 223 | 14| Selected default and alternate trees recovered | 128 | 15| Complete source paths recovered | 9,457 | 16| Independent Git blob integrity matches | 9,457 of 9,457 | 17| Distinct verified source blobs | 4,935 | 18| Additional branch-path mappings to identical verified blobs | 15,299 | 19| Selected readable files still unavailable | 1 | 20| Website build/configuration records | 216 | 21 22All selected readable text in the recovered trees is available except the explicitly listed peripheral SEO reference HTML. Shared Git blobs are reused only after calculating the Git object hash over the exact downloaded bytes. Content identity across two paths therefore has byte-level evidence; it does not imply identical repository histories or runtime behavior. 23 24The 216 website records consist of 33 repository builds or identity checkpoints, 67 branch generations, six separately explained subsystems, ten Generator Core families and 100 local constructor configurations. The latter are local registry settings that share family implementations. They are not a claim that one hundred official model architectures, pretrained checkpoints or independently trained models were recovered. 25 26## Integration files 27 28| File | Purpose | 29| --- | --- | 30| `github_builds.json` | Complete build and configuration catalogue with explanations, exact constructor text, components, equation links, pinned commits and source references. | 31| `github_generator_configurations.json` | The 100 named Go constructor configurations, with IDs 1 through 100, exact source ranges and parent-family links. | 32| `github_source_manifest.json` | Full recovered source paths, stable source IDs, local paths, pinned URLs, byte counts, Git blob SHA, SHA-256 and integrity status. | 33| `github_coverage.json` | Repository and branch coverage, selected-text counts, explicit exclusions, one retrieval failure and equation-classification counts. | 34| `github_branch_source_aliases.json` | Exact alternate-branch paths whose complete bytes are available under a canonical source ID. | 35| `github_equation_candidates.json` | Accepted mathematical source occurrences for equation pages. | 36| `github_equation_candidate_ledger.json` | Classification and disposition of every originally detected unique expression, including rejected non-equations and replacement links. | 37| `github_equation_candidates_raw.json` | Every original candidate occurrence and its source context, retained even when unsuitable for an equation page. | 38| `github_equation_recovered_spans.json` | Complete adjacent source-comment spans that repair incomplete line fragments without adding mathematical terms. | 39| `source_fetch_failures.json` | Exact metadata and retrieval outcome for the unavailable peripheral file. | 40 41The authoritative source tuple is `(repository, commit, path)`. The stable source ID is derived from that tuple; a source with a different pinned commit retains a distinct identity. All source references in the final build and accepted-equation catalogues resolve to the source manifest. All equation links in the build catalogue resolve to the accepted equation file. 42 43The complete source bytes are under `sources/`. Default snapshots use `sources/<repo>/<path>`. Recovered differing alternate snapshots use `sources/<repo>@<commit>/<path>`. The `trees/`, `alternate_trees/`, `branches/` and `commit_metadata/` directories preserve the inventory and commit evidence. Original transcripts can contain operational code and private project information, so the website should use the intended sanitized reading copies and maintain the private audience chosen for this edition. 44 45## Equation classification 46 47The extractor recovered **3,016 candidate occurrences**, grouped into **1,212 exact unique expression strings**. A bounded content classification and context review identified **364 non-equation or incomplete-fragment groups**. Commands, protocol attributes, error-handling examples, decorative headings, callback types, field assignments, operator-log prose and qualitative labels were retained in the candidate ledger and full source transcripts. 48 49The accepted output contains **1,300 source occurrences representing 858 exact unique expression strings**. Eighteen complete comment spans repair ten groups of split expressions. These include the RDP conversion, survival-product definition, calibrated score, coefficient sum, free-energy continuation, arithmetic working and an integral continuation. The repaired records identify the original candidate IDs they replace. Their expressions use adjacent source text, with source ranges updated to the full span. 50 51The classification preserves mathematical parameter definitions, constraints, pseudocode and documented model analogies even when their scientific meaning or implementation fidelity is disputed. In particular, the uncertainty formula, neuronal updates, potential and coupling updates, Euclidean norm, FNV pseudocode, Shannon entropy, diffusion sampling formula, Athena argmax, Dionysus exponential and Zeus product remain present with their original source identities. 52 53An exact-string count is not a count of independent mathematical laws or original discoveries. Prefix wording and source presentation can produce different strings for related formulas. Scientific interpretation, mathematical corrections, attribution and implementation mismatches belong beside the preserved expression, using the separate scientific review. 54 55## Important recovered lineages 56 57**Complete AGI Core has substantial material outside its default branch.** The alternate branch `claude/improve-math-logic-ehIW1`, pinned at `33aa84f2aa61cedbe010240ada7f4efbdfe795a7`, adds the Go Generator Core, the C++ Mind Farm and a native Unified consciousness component. Restricting recovery to `main` would have omitted these systems and their equations. 58 59**The original ThatAIGuy, original Seferim and Unified v3 remain separate.** Their directory prefixes, rosters, orchestration scripts and numerical claims describe different snapshots. The original seventeen-name Pantheon, sixteen-family G16 work and twenty-four-name Unified roster must not be collapsed into one roster or assumed to share a state format. 60 61**AI_Brain preserves materially different branches.** Its default native Tired AGI tree contains DNA, helix, FGTP, ingestion, bridges and hardware-oriented source. Older branches preserve Python and C# seeding and marketing-ANI work. Branch names and pinned file differences are evidence of distinct development snapshots; they are not runtime verification. 62 63**Missionmatch contains applied AI development across many branches.** The recovered trees include workforce ranking, graph retrieval, live-brain integration, a MAG substrate, teaching loops, differential-privacy accounting, survival models, vendor scoring and supporting infrastructure. Its mathematical examples are labelled as application and test evidence, separately from consciousness claims. 64 65**AI Music System advanced on the observation date.** The pinned default source constructs arrangement plans using genre/mood corpus profiles, counts, tempo averages and weighted choices. The recovered source does not establish an audio waveform model merely because it is named a music generator. Earlier music branches and the standalone autonomous-music HTML lineages remain separately recorded. 66 67**Some named identities have very small current repositories.** Healer-AI retains a short README; skynet-backend retains a funding configuration. These are explicit identity checkpoints rather than inferred full implementations. The current tree cannot establish what may once have existed in local, deleted or uncommitted work. 68 69**Affiliate Agents preserves one thousand numbered HTML entries.** AGENT4001 through AGENT5000 are fully recovered as publishing records. Their directory count does not demonstrate one thousand independently running AI engines. 70 71**The Hive endpoint is a small API-state scaffold.** In the recovered `feedthejoe-hive/sync.py`, start and stop change an in-memory count. The source does not launch a separate worker for each increment. 72 73**General application names can hide relevant integrations.** `seo-sidecar` contains the MEGAMIND Go SEO feedback, experiment and analytics subsystem. The two newer illustrated-website repositories contain creative pipelines, publication and authority projection, and AI-discovery provider interfaces. Their authored text was screened along with all other default repositories. 74 75**Astra has more than one meaning in the recovered evidence.** The ASTRA plans under `thatstupidcomputer/docs/` concern the illustrated workshop website and its mobile character rigs. They are not evidence for the earlier proposed Astra mobile consciousness node. Historical conversation plans for that node must remain separately sourced unless matching implementation is recovered. 76 77## Static implementation findings handed to scientific review 78 79The recovered Generator Core exposes concrete limits that should appear beside its claims. Its autoregressive family fixes a hidden dimension of 256 but tries to slice a hidden-width linear-layer result to much larger vocabulary lengths; this is a static out-of-bounds condition for the named constructors. The inspected family `Train` methods return loss values without making an optimizer update inside those methods. Architecture-name flags do not establish implementation of the corresponding published architecture. 80 81Scientific review also identified the zero denominator in the Go DNA family coupling initialization, the scalar behavior of CompleteThought relative to the documented temporal integral, and the distinction between returning a maximum utility value and the argmax location in Athena. Source originals were preserved; no remote source fixes or engine execution were performed. 82 83## Explicit limits 84 85One selected readable file remains unavailable: `Janady13/seo-aio-26-discipline-library/seo-aio-26-discipline-library.html`, pinned at `7dd3f9fb313ee3298f08e28c0435cf7415d839b8`, expected size 1,767,268 bytes and blob SHA `0930c74b99ef0e8dc0760c18944733449e231422`. The connector returned an empty body for both requested encodings; its unsupported blob/raw routes did not recover the text. Its metadata is preserved, and it is not represented as a complete recovered source. 86 87Credential and environment-secret filenames, dependencies, caches, embedded Git object backups, binary databases, model weights, large generated visual inspection dumps and collected Python knowledge records remain represented by tree metadata where present. They were not executed or treated as authored research equations. A bounded literal-pattern scan of recovered text found six private-key marker examples in marketing-intelligence test fixtures; it did not establish a valid private-key payload. This is a scoped pattern check, not a security certification. 88 89All 61 repositories received selected default-source screening. The coverage file enumerates every branch whose tree was outside the selected research scope, including peripheral/reference-repository branches. Full Git object history, deleted or unreachable commits, private repositories outside the connected account, device-only files and local uncommitted work are not mirrored by this audit. The website should therefore describe this edition as a comprehensive catalogue of recovered evidence with an explicit coverage ledger, rather than asserting proof that no historical build or equation can exist elsewhere. 90