What the seed corpus covers (institutional capture, controlled opposition, PSYWAR/occult, and the method), the abstract→concrete axis, and what the structure says we still need to add — modern conspiracy, alt-chronology, the reflexive sovcit layer, embeddings, scale.

Topics & Roadmap

What the seed corpus already covers, and what the structure predicts we should add next. The seed material is the three compressed knowledge-graph artifacts plus the design docs that shipped in the original project archive.

1. The seed topic set

The originating material clusters into three substantive cases plus one method.

Case A — institutional capture of moral traditions ("Script Hypothesis")

Two neighbour-love traditions (Judaism, Christianity) repeatedly borrowed by state, imperial, financial, and intelligence-adjacent actors. The reusable object is not a verdict but a pattern: a tradition teaches anti-dehumanization → an apparatus borrows its legitimacy → belonging is redirected into identity conflict → the conflict masks extraction. Distinctions held as first-class: a tradition ≠ its institutions ≠ a political movement ≠ an intelligence claim.

Case B — controlled opposition & ritual inversion ("Saturn both-sides")

The claim that the same Saturn/Saturnalia pattern (order plus managed revolt) appears on both political wings. Documented nodes (Saturnalia inversion; Theosophy/Lucifer; Bailey → Lucis; MKUltra; Wasson's CIA-front expedition; Jonestown; SLA; Process Church; O9A insight-roles) are kept separate from the symbolic overlay that would unify them — the overlay may interpret history but may not cause it unless an actor demonstrably used the symbol.

Case C — PSYWAR, occult, and manufactured criminality ("PSYWAR of Set")

Aquino as a high-salience bridge node (PSYOP officer and Temple of Set founder; From PSYOP to MindWar; named-and-contested investigations). Strong claims (continued-Nazi-occult-research, extortion-protection, cult/gang infiltration) encoded as allegations/hypotheses, never facts, until independent subgraphs corroborate.

The method (the real payload)

Source-as-observation; Cromwell rule (no claim at probability 0 or 1); confidence separated from salience (a wild claim can be low-confidence and high-salience and is kept, not deleted); graphlet recurrence across independent families as the truth-emergence signal; anti-overfit rules.

The six graphlet templates the seeds name

whispering adversary · controlled inversion · manufactured criminality · legitimacy laundering · source suppression · purification / remediation. These are the hypotheses-as-structure the experiment harness tests for recurrence.

2. The abstract → concrete axis

Every node now carries a coordinate from abstract (symbol, metaphysical claim) to concrete (named person, dated event, document), combining its type with the epistemic layer of the assertion (metaphysical → mythic → interpretive → descriptive). This is the first real sorting dimension: it lets us ask at what altitude convergence happens — do independent traditions agree on concrete facts, or on abstract structure? Early signal: recurrence concentrates in the mid/structural band, which is the interesting place (shared mechanism, not shared names).

3. Predicted additions (what the structure says we still need)

More families (each is just a new source-family; the process does not change)

• Modern internet/social conspiracy — e.g. the Britney "managed/AI-since-2021" narratives, Illuminati-control memes. High salience, very low confidence, metaphysical/mythic layer. The Cromwell rule + confidence≠salience split exist precisely so this material can be kept and watched, not discarded. (Needs a fetch adapter — reddit/news dumps; not yet in storage.)
• Alternative chronology / "missing years" — Fomenko (now ingested as alt_chronology). Tests whether a revisionist-history narrative shares structure with ancient myth or with cult mechanics.
• The reflexive layer — sovereign-citizen / own legal corpus — does the isolation → identity-rebuild → compromise-lock motif recur in the legal/sovcit material the same way it does in cult and scripture cases? (Sensitive; kept local, role-level only in any public view.)
• County tax / public records at scale — the planned large corpus; the reason the hot paths are already vectorized and the storage is tiered to B2.

More dimensions (concepts can't be sorted without them)

• Semantic embeddings for every entity/claim (local qwen3-embedding / nomic-embed-text), so concepts cluster cross-lingually and motifs match by meaning, not string. This is the Qdrant half of the architecture and the next dimension after abstract→concrete.
• Temporal (when), geographic (where), independence/provenance (who-copied-whom).

Better extraction

• Instance-level is in (named entities + edges). Next: stronger calibration, de-duplication of entities across chunks, and suppressing concept-bleed from the prompt. A larger model on a GPU box makes this both better and faster.

4. What "getting there" requires

The discipline never changes — new data is a new family, new context is a new dimension. That is the process; everything above is just feeding it.