# Methodology > [[index|Index]] · [[overview|Overview]] · [[License]] The Israel Bonds Investigation wiki is built under a hybrid methodology that combines two independent contributions from recent LLM-tooling work. We call it the **Karpathy-Hegelion Pipeline** (sometimes "Methodology" in cross-references) because the two components correspond to the two operational problems a public-records wiki has to solve at scale. **Karpathy's persistent LLM-wiki pattern** solves the *accumulation* problem: how does a researcher build a knowledge artifact whose value compounds across hundreds of source documents, instead of having the model rediscover the corpus on every query. **Hegelion's three-phase dialectical harness** solves the *contestation* problem: how does the same researcher prevent an LLM with a strong sycophancy prior from flattening genuinely contested questions into the safest-sounding split-the-difference summary. Both components are open-source and publicly documented. Both have been adapted, not adopted wholesale — the adaptations are recorded below so a reader (or a future maintainer) can audit the deviations. --- ## I. The Karpathy layer — persistent compilation over immutable evidence ### The problem Karpathy names In a public gist published as ["LLM Wiki"](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f), Andrej Karpathy diagnoses the structural limit of retrieval-augmented generation: *"the LLM is rediscovering knowledge from scratch on every question. There's no accumulation."* His proposed remedy is to have the LLM *"incrementally build and maintain a persistent wiki — a structured, interlinked collection of markdown files that sits between you and the raw sources."* The wiki becomes *"a persistent, compounding artifact"* whose *"cross-references are already there. The contradictions have already been flagged. The synthesis already reflects everything you've read."* ### Karpathy's three-layer architecture Karpathy is deliberately abstract about schema and format — *"This document is intentionally abstract. It describes the idea, not a specific implementation."* What he prescribes is a strict layered separation: 1. **Raw sources.** *"Your curated collection of source documents. … These are immutable — the LLM reads from them but never modifies them. This is your source of truth."* 2. **Wiki layer.** LLM-generated markdown. The LLM *"does all the grunt work — the summarizing, cross-referencing, filing, and bookkeeping."* 3. **Schema layer.** An `AGENTS.md` carrying the operating constitution for the maintainer. Two index files coordinate the wiki layer: `index.md` (*"content-oriented … a catalog of everything in the wiki — each page listed with a link, a one-line summary"*) and `log.md` (*"append-only record of what happened and when — ingests, queries, lint passes"*). ### How this corpus instantiates the architecture For the Israel Bonds investigation, the raw layer is a single strand of Tier-1 evidence: FOIA productions and Arkansas public records produced under the Arkansas Freedom of Information Act (Ark. Code § 25-19-101 *et seq.*) and through the Arkansas Securities Department's broker-dealer registration files. The custodial structure is: - `raw/atrs/` — Arkansas Teacher Retirement System FOIA productions (board packets, board rules, staff emails, audio recordings). - `raw/apers/` — Arkansas Public Employees Retirement System FOIA productions (IFSC packets, board packets, staff communications). - `raw/ashers/` — Arkansas State Highway Employees Retirement System FOIA productions (the cross-agency Berman pitch and the control-case decline). - `raw/treasury/` — Arkansas State Treasury FOIA productions (investment policies, internal credit overview, SBOF accounting summaries, ladder schedules). - `raw/auditor/` — Auditor of State FOIA productions (Milligan, Brady, Spadoni, Peterson, residual custodian buckets). - `raw/securities/` — Arkansas Securities Department FOIA production on DCI broker-dealer registration (155 files, CRD #11148). Audio recordings (ATRS Investment Committee and Board of Trustees meetings produced in FOIA responses) are transcribed locally with an offline faster-whisper transcription tool (no API key, no data leaves the machine); the transcript file becomes the ingestible artifact and is itself a raw source. External material (statute text, news reporting, encyclopedic background, contemporaneous public statements) lives outside the immutable layer; it contextualizes claims, never establishes what an Arkansas agency, pension system, or company did. ### Two adaptations Karpathy's gist does not specify The Israel Bonds corpus adds two disciplines to Karpathy's abstract design: **The contamination firewall.** Wiki pages are LLM-compiled artifacts — useful for navigation, synthesis, and pattern recognition, but **never citeable as primary evidence in any deliverable**. Every factual claim traces, within two hops, to a raw source with file path, location, and verbatim quote. Karpathy's gist permits citing the wiki itself; the public-records context of this investigation does not. The contamination firewall is the load-bearing reason every entity, concept, and source page carries an `## Evidence` section with verbatim Tier-1 quotes — the page summary is the navigation surface; the evidence section is the citation chain. **The two-hop maximum.** Wiki pages may cite each other, but the citation chain from any factual claim to a raw source document must complete in two hops or fewer. A concept page may cite an entity page that cites a raw source (two hops); a concept page may not cite an entity page that cites another concept page that cites a raw source (three hops). The two-hop ceiling exists to prevent the wiki from compounding error: every hop introduces interpretive distance, and a three-hop chain is structurally indistinguishable from a hallucination. ### Antecedent Karpathy traces the design back to Vannevar Bush's Memex ("As We May Think," *The Atlantic*, July 1945) — *"a personal, curated knowledge store with associative trails between documents."* Bush's bottleneck was the human labor of building those trails. Karpathy's claim is that LLMs solve that bottleneck. This corpus is one test of that claim against a non-trivial multi-agency public-records record. --- ## II. The Hegelion layer — call-isolated dialectics for contested mechanisms ### The problem Hegelion names The Karpathy pattern does not, by itself, adjudicate contradictions. The "Lint" workflow flags them; it does not resolve them. When two FOIA productions or two communications record the same mechanism differently — Statement A says the ATRS Resolution 2025-22 preamble cites Aon Hewitt's "advice" as basis for the authorization; Statement B says the actual Aon "memo" in the Board packet is a header plus a DISCLAIMERS heading with no substantive body — the wiki needs more than a `> [!contradiction]` callout. It needs a procedure for staging the dispute in a form a downstream reader (a journalist, a legislator, an intervenor) can audit. [Hegelion](https://github.com/Hmbown/Hegelion) is a dialectical-reasoning harness for LLMs published as an open-source Python package by Harrison Bown. Its central architectural claim is that *"when the model must commit to a thesis, then genuinely attack it in a separate call, the synthesis surfaces insights that single-call approaches shortcut."* The harness enforces three sequential model calls: > *"[Call 1] Thesis → LLM commits to a position / [Call 2] Antithesis → LLM attacks that position (separate call, no hedging) / [Call 3] Synthesis → LLM reconciles."* The premise — that single-context "argue both sides" prompting produces hedged mush, while call-isolated oppositional construction produces genuine disagreement — is what we call **call isolation**: a term Hegelion implies architecturally without naming, and that we adopt as an explicit discipline rule. ### The empirical anchor for explicit-vs-implicit framing Hegelion's published documentation reports no quantified benchmarks for the call-isolation claim itself. The strongest empirical anchor for the design comes from a separate study — *Inducing Disagreement in Multi-Agent LLM Executive Teams: Only the Devil's Advocate Works* ([TMLR submission, January 2026](https://openreview.net/forum?id=mxBmj5LYU2)) — which measures disagreement rates across five prompting conditions in multi-agent LLM decision-making: | Condition | Disagreement rate | |---|---| | **Devil's Advocate** (explicit *"you must oppose"*) | **99.2%** | | Combined soft techniques | 63.3% | | Strong Role Framing | 61.7% | | Explicit Dissent Instructions | 55.0% | | **Baseline** (no oppositional framing) | **48.3%** | The headline finding: **explicit behavioral assignment ("you must oppose") succeeds where implicit instructions ("think critically," "consider counterarguments") fail.** The three "soft" techniques are statistically indistinguishable from baseline. The same study surfaces a counterweight: under Devil's Advocate framing, 4.9% of agents recommend options they privately rate lower, producing "inauthentic dissent." That tradeoff is the reason this wiki retains a human-in-the-loop verdict at the integration step rather than auto-promoting the synthesis-phase output into the source tension's `status:` field. The explicit-framing rule is the load-bearing reason `AGENTS.md` instructs antithesis subagent prompts to use **"you must oppose," "you must attack," "you must argue against"** — not "consider counterarguments" or "think critically." Soft framing collapses back to baseline disagreement; hard framing achieves near-total disagreement, with the understood and bounded inauthenticity tradeoff. ### How this corpus instantiates the dialectic The wiki's tension pages and dialectic phase files implement the Hegelion pattern with three specific adaptations: **1. Tensions are first-class durable objects, not transient context.** Every `> [!contradiction]` callout flagged during ingest is also filed as a `tensions/T### - <Slug>.md` page with explicit Statement A / Statement B / Resolution status fields, a `tension_type:` classifier (framing / factual / attribution / evidentiary / temporal), and frontmatter back-links to every concept page provisional on its resolution. The tension persists across dialectical rounds even if a given dialectic does not resolve it; a second dialectic may be run later when new evidence lands (a follow-up FOIA production, a board minutes release, an SBOF deposition). Hegelion treats the source question as ephemeral context for a single invocation; this corpus keeps it as a citable, durable artifact. **2. Fresh-context subagents enforce call isolation at the agent-orchestration layer.** Each of the three phases is executed by a separate `Agent` invocation with a freshly constructed prompt — no shared context window across phases. The thesis subagent is blocked from reading Statement B's sources and the parent concept page (the parent encodes a directional reading that would short-circuit independent construction). The synthesis subagent is blocked from reading both the source tension page and the parent concept page (`Glob` / `Grep` for wikilink resolution is permitted, but re-reading the framing reintroduces framing bias). What Hegelion enforces inside a single Python harness, this corpus enforces at the agent-orchestration boundary. **3. Synthesis-phase verdicts are explicit and discrete.** The synthesis subagent must recommend one of four concrete `status:` values for the source tension: - `open` — competing readings stand; new evidence pending. - `bracketed-because-<reason>` — sharp disagreement on what evidence would settle it; action deferred (e.g., `bracketed-because-followup-FOIA-pending`, `bracketed-because-board-minutes-not-yet-released`). - `resolved-via-D###-Statement-A` or `-Statement-B` — settled by this dialectic; specifies which Statement won. - `unresolvable-without-<evidence>` — no realistic evidence path exists on the available record (e.g., `unresolvable-without-deposition-of-X`). *"It's complicated"* is not a verdict. Integration of the verdict back into the tension's frontmatter happens in the main session, not in a subagent, so the human-in-the-loop retains final authority over status changes — and the inauthentic-dissent tradeoff from the Devil's Advocate study stays bounded. ### The Hegelian shape of the verdict The synthesis-phase output produces four determinations under fixed section headings: **what is resolved**, **what is sharper but unresolved**, **what is bracketed**, and **verdict on the source tension's status**. This is deliberately Hegelian. The verdict is *aufhebung* in the standard Stanford-Encyclopedia reading ([plato.stanford.edu/entries/hegel-dialectics](https://plato.stanford.edu/entries/hegel-dialectics/)): the contradiction is not collapsed into one of its sides, nor split-the-difference averaged, but **raised to a more articulated form** in which the original disagreement is preserved as a clarified question. > **Worked example.** The first dialectic in this corpus ran against the [[independent-credit-analysis-gap]] consultant-role tension on ATRS Resolution 2025-22 — Statement A: the resolution preamble cites Aon Hewitt's "advice" as basis for the authorization, and the consultant engagement produced a documented advisory product across the 5/8 directive, the 5/28 Kelly Reams forward, the 6/2 IC oral participation, and the BOT motion. Statement B: the Aon memo in the Board packet (Attachment 17, pp.149-150) is a header plus a DISCLAIMERS heading with no substantive body; the actual decisional rationale was Mark White's 5/22/2025 "pecuniary standpoint" self-attribution; the shifting framings of the consultant role across the documentary record are themselves evidence of post-hoc reconstruction. The dialectic [[D001 - Resolution 2025-22 Consultant-Role Attribution/D001 Synthesis|D001 Synthesis]] produced four determinations. **What is resolved:** the packet artifact is two pages of header and empty heading (thesis stipulated; antithesis established); the within-meeting Franklin Park comparator on Resolutions 2025-23 and 2025-24 and the April 7 baseline are documentary-record fact the thesis's licensure-scope framing cannot account for; the IC member's verbatim post-vote statement at segments 2681-2685 that the credit-risk discussion was "what I was wishing we would hear more of" stands unrebutted in the thesis. **What is sharper but unresolved:** the textual question of what the Resolution 2025-22 preamble's "advice" denotes — broad engagement-product reading versus narrow substantive-written-artifact reading — both defensible on the preamble's language. **What is bracketed:** the 6/2/2025 Investment Committee minutes (transmitted to Jennifer Lenow on 6/4/2025; not in the corpus) which would document on-record IC consultant statements; depositions of PJ Kelly or Katie Comstock on the substantive content of Aon's "advice" on Israel Bonds. **Verdict on tension:** `resolved-via-D001-Statement-B-on-evidentiary-question-with-open-BP4-A5-interpretive-question`. The dialectic does not pretend to settle what BP4 Section A.5's "written advice" language requires; it has narrowed and named that as the operative open question for a future dialectic against [[T004 - BP4 Section A5 Compliance on Israel Bonds]]. ### Antecedent Hegelion's repository cites prior multi-agent disagreement research as architectural inspiration. The deeper antecedent is Hegel's *Wissenschaft der Logik* (1812–1816) and the modern reconstruction of dialectical method in the Stanford Encyclopedia. The methodological warrant for treating contradictions as productive — rather than as errors to be averaged away — is two centuries old. Hegelion translates that warrant into LLM call topology; this wiki translates it into a public-records research procedure. --- ## III. The four workflows The schema layer (the `AGENTS.md` operating constitution at the repository root, not published on this site) defines four workflows that operate over the same persistent wiki and share its citation discipline. The first three are Karpathy's; the fourth is Hegelion's. - **Ingest** — read a raw source from `raw/<agency>/<production>/`; surface key takeaways to the user before writing; create one source page per FOIA production (or per significant document group); create or update entity pages for every named person, agency, and organization; create or update concept pages for every analytical theme invoked; create or update tension pages for every contested mechanism the source surfaces; update `index.md`; append a `log.md` entry; run a wikilink-resolution check and resolve every unresolved link before finishing. - **Query** — read `index.md` to identify relevant pages; pull those pages; synthesize an answer with Obsidian-style double-bracket wikilinks; trace every factual claim to a raw source within two hops. **Every factual claim traces to a raw source document, not a wiki page.** If a wiki page is referenced but lacks detail, pull from the underlying raw source in `raw/`. Optionally file the answer as a new concept page or append it to an existing one if it is a non-trivial insight future queries will want. - **Lint** — scan for orphans, missing pages, stale claims, concept gaps, cross-reference gaps, citation-format violations, contamination of the wiki/primary-evidence firewall, redaction violations, untracked tensions (concept or source pages with `> [!contradiction]` callouts but no corresponding tension page), undeclared tension dependencies, orphan tensions (tension pages with empty `synthesis_pages:`), orphan dialectic phases (`D### - <slug>/` folders missing one or two of Thesis / Antithesis / Synthesis), undeclared dialectical antecedents, and stale dialectic-vs-tension status. Report; do not auto-fix. - **Dialectic** — for a tagged tension (`T###`), run thesis / antithesis / synthesis as three sequential fresh-context subagents under call isolation, then integrate the verdict back into the tension's frontmatter, update any concept or synthesis pages declaring the tension in their `tensions:` (or `relies_on_tensions:`) field, and append a log entry documenting the verdict and any newly-surfaced evidence the prior concept page did not capture. Full workflow definitions, the explicit-framing prompt-discipline rules, the contamination firewall, the citation tier hierarchy, and the agency-key conventions are codified in the operating constitution at the repository root, which is operational and not published on this site. This page is the public-facing companion to that schema; both should be re-read whenever the operating constitution changes. --- ## IV. What this methodology refuses The methodology is defined as much by what it refuses to do as by what it does: - **Never modify the raw layer.** Nothing under `raw/` is ever changed by the LLM. Audio transcriptions write to `raw/<agency>/<production>/<recording>.transcript.txt` and then those transcripts are themselves frozen. - **Never republish redacted content.** Information an agency redacted in the underlying production — citizen-requester personal email and phone, bank-account numbers, personal identifiers — stays redacted in the wiki layer. Public-official names and public-official personal Gmail addresses used in their official capacity are retained as substantive procedural findings, distinct from citizen-requester PII. - **Never present inference as established fact.** "Cannot verify" is a legitimate finding. Absence is information, not evidence. The wiki shows the gap; it does not fill it. - **Never let wiki pages cite other wiki pages as primary evidence.** Every factual claim traces, within two hops, to a raw source with file path, location, and verbatim quote. The wiki is the navigation surface; the raw productions are the evidence. - **Never force a `resolved` verdict on a record that does not support one.** The Hegelion synthesis produces a verdict; when the dispositive evidence is outside the dialectic's reach (follow-up FOIA pending, board minutes not yet released, deposition not yet taken), the verdict is `bracketed-because-<reason>` or `unresolvable-without-<evidence>`, not `resolved`. Substituting the wiki's accounting for the underlying decision-maker's would be the structural error the dialectic is designed to surface, not commit. - **Never provide investment, legal, fiduciary, or regulatory advice.** The wiki documents what the records show. Whether a particular pension system board's decision satisfied its fiduciary duty under Act 498 is a legal question for counsel and ultimately the courts; the wiki carries the documentary record on which any such determination would rest. --- ## V. Further reading The two foundational documents: - **Karpathy, Andrej.** *LLM Wiki.* GitHub gist. <https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f> - **Bown, Harrison.** *Hegelion: Dialectical Reasoning Architecture for LLMs.* GitHub repository, MIT-licensed Python package. <https://github.com/Hmbown/Hegelion> The empirical anchor and the philosophical antecedent: - *Inducing Disagreement in Multi-Agent LLM Executive Teams: Only the Devil's Advocate Works.* TMLR submission, January 2026. <https://openreview.net/forum?id=mxBmj5LYU2> - *Hegel's Dialectics.* Stanford Encyclopedia of Philosophy. <https://plato.stanford.edu/entries/hegel-dialectics/> - Bush, Vannevar. "As We May Think." *The Atlantic*, July 1945. The wiki's operating constitution: - `AGENTS.md` (repository root; not published on this site) — the full Hegelion-layer schema: tension and dialectic page templates, the dialectical workflow, explicit-framing discipline rules, and lint additions. The companion Karpathy-layer conventions — raw-immutability, two-hop citation, the contamination firewall, the ingest/query/lint workflows, agency keys, and page-template fields — are enforced as corpus conventions and summarized in §I and §III above. - [[index|Index]] — the catalog of every wiki page with one-line summaries; the operational entry point for the Query workflow. - [[overview|Overview]] — the investigation framing (corpus scope, pension system inventory, FOIA campaign status, key evidence threads). External sources cited on this page contextualize the methodology's research provenance; per the corpus's source-tier discipline they do not establish what any Arkansas agency, pension system, or company did.