Answer
The wiki is doing its job at the content level — clusters are coherent, entity/concept pages are substantive, and the overview synthesis is sharp. But it is failing at the structural plumbing in ways that will compound. The two biggest problems are (1) the wiki/syntheses/ directory does not exist, so every query output since project start has either evaporated into chat or been misfiled, and (2) wiki/overview.md is the single largest source of broken wikilinks in the vault — it liberally references named actors (Miran, Lisa Cook, Spanberger, Sherrill, Mamdani-as-surname, World Liberty Financial, Israel, Strategic Bitcoin Reserve) that have no backing page. The wiki’s best-written document is also its biggest schema violator.
The emerging argument in overview.md largely holds up — the infrastructure-chokepoint spine is real and evidenced — but it papers over the Iran cluster’s thinness (6 of the 8 Iran-related concept pages have sources: 1) and the mental health cluster’s unfinished connective tissue.
Priority below. Numbers in [brackets] are the counts I verified against the filesystem today, not what frontmatter claims.
Supporting Evidence
P0 — Fix These Before The Next Ingest
P0-1. wiki/syntheses/ directory does not exist.
- Verified by
Glob wiki/syntheses/**/*→ no files. - CLAUDE.md (core rule #7) says “If a query produces a useful synthesis, save it to
wiki/syntheses/.” There is nowhere to save them. This report is the first file in the directory. - Action: This file creates the directory. But it also means every prior query output since the wiki’s inception has either been lost to chat scrollback or misfiled into
wiki/concepts/(which is why some “concept” pages read more like one-off analyses — see the LLM Wiki Agent page). Recommend a one-time sweep ofwiki/concepts/to identify query-output pages that should move tosyntheses/.
P0-2. wiki/index.md frontmatter counts are wrong by large margins.
wiki/index.md:5claimstotal_pages: 594andtotal_sources: 385.- Actual filesystem counts (from Glob + Grep):
wiki/sources/: ~384 files (Glob returned 384; one source listed in index probably doesn’t have a corresponding file, or vice versa — small drift)wiki/entities/: 79 fileswiki/concepts/: 58 fileswiki/articles/(includingepisodes/): 82 fileswiki/syntheses/: 0 files (directory absent)- Plus
overview.md,index.md,log.md= 3 - Total: ~606 pages, not 594. And
total_sourcesshould be ~384, not 385.
- The drift is small, but the principle is important: the index is load-bearing metadata and its counts should match the filesystem. This is also a lint item that can be automated.
- Action: Recompute both numbers at the end of each ingest; consider adding a section-level count check (Sources: 384, Entities: 79, …) inside
wiki/index.md:9so drift is visible.
P0-3. wiki/overview.md has at least 10 named-entity broken wikilinks.
I searched the entire vault for each of these bracketed links and confirmed the target page does not exist:
| Broken link | Where it’s referenced | Severity |
|---|---|---|
[[Stephen Miran]] | overview.md L40 + Fed section | P0 — central actor in Fed pressure campaign |
[[Lisa Cook]] | overview.md L40 | P0 — Trump attempted firing is a landmark event |
[[Sherrill]] | overview.md L46 | P0 — NJ governor, 2025 elections centerpiece |
[[Spanberger]] | overview.md L46 (but Abigail Spanberger exists) | P1 — link format mismatch |
[[Mamdani]] | overview.md L46 (but Zohran Mamdani exists) | P1 — link format mismatch |
[[Strategic Bitcoin Reserve]] | overview.md L30 | P1 — referenced in Trump entity, but no standalone page |
[[World Liberty Financial]] | implied throughout crypto section, not bracketed but should be an entity | P1 — $57M conflict of interest, no page |
[[Israel]] | entities/Donald Trump.md L93 | P1 — co-belligerent in the active war, no page |
[[Jake Auchincloss]], [[Rick Crawford]], [[Tim Kaine]] | Donald Trump.md Connections | P2 — supporting actors, stubs acceptable |
Additional broken links found elsewhere:
[[Federal Immunity Above Constitutional Law]]— referenced in concepts/Retroactive Executive Protection.md L74 and overview.md L130’s cross-cutting patterns. No file.[[Institutional Capture]]— referenced in concepts/Retroactive Executive Protection.md L76. No file.[[International Humanitarian Law]]— referenced in concepts/Infrastructure Warfare.md L41. No file.[[Trump blames the radical left for Charlie Kirk's killing]]— referenced in entities/Donald Trump.md L86. Not inwiki/sources/(confirmed via Glob).
Action: Create stubs for Stephen Miran, Lisa Cook, Mikie Sherrill, World Liberty Financial, Israel (minimum). Fix the Spanberger/Mamdani link format (use [[Abigail Spanberger|Spanberger]] or rename the target). This is a 45-minute pass.
P1 — Schema Compliance Drift
P1-1. Concept-page sources: counts are stale across the entire Iran cluster and most “recent-ingest” concepts.
The overview says Iran/geopolitics has “~10 sources.” Actual counts from concept frontmatter:
concepts/War-Driven Inflation.md:7→sources: 1concepts/Chokepoint Control.md:7→sources: 1concepts/Infrastructure Warfare.md:7→sources: 1concepts/Coercive Diplomacy.md→sources: 1concepts/Oil Seizure as Coercion.md→sources: 1concepts/Tech-State Conflict.md→sources: 1concepts/AI Sovereignty.md→sources: 1concepts/Retroactive Executive Protection.md:7→sources: 1concepts/PKM Failure Pattern.md→sources: 1concepts/LLM Wiki Agent.md→sources: 1concepts/Shadow Docket.md→sources: 1
Every one of these has a “Key Sources” section with exactly one entry — so the frontmatter is technically accurate, but that means the concept pages have not been updated as newer sources landed. For example, the Iran war has at least: Trump threatens hell on Iran infrastructure if Strait remains blocked, Will blow up everything, take over Iran’s oil — Trump says can reach deal by Monday, Britain woos Anthropic expansion after US defence clash (DoD angle), and the Marjorie Taylor Greene / coalition sources — but none of that has flowed down into the concept pages.
This is the single biggest process failure in the ingest workflow: Step 4 of the INGEST workflow in CLAUDE.md says “If a wiki page already exists: update it, add source to its list, increment sources: count.” That’s not happening for concepts — only for entities (entity counts on Donald Trump = 40, Federal Reserve = 23 look reasonable).
Action: One-time reconciliation pass — for each concept, search for [[Concept Name]] across wiki/sources/ and rebuild the Key Sources list. This is the highest-leverage hygiene task in the audit.
P1-2. Source count in the wiki/index.md Federal Reserve section already exceeds the overview’s “~30 sources” claim.
Counting just the Federal Reserve subsection of wiki/index.md (lines 23–78): ~35 source entries visible in the first screen, and the section continues further. Not a correctness problem, but the overview numbers are aging.
P1-3. Concept stubs that are actually fine vs. concept stubs that need work.
Thin concepts (sources: 1–2) that need expansion now because the material exists:
- Chokepoint Control — should cite Iran sources + NFL Macdonald piece (pre-snap ambiguity is an information chokepoint) + GENIUS Act T-bill backing (captive Treasury chokepoint). The overview explicitly calls out that these are all the same concept (L126), but the concept page doesn’t reflect it.
- Infrastructure Warfare — should cite Russia/Ukraine if that’s in the wiki, and definitely the Gulf energy infrastructure strikes across the Iran cluster.
- Coercive Diplomacy — should absorb the Trump dual-track messaging thread that’s all over the Iran cluster.
- Tech-State Conflict — Anthropic blacklisting + UK recruitment is evidenced across 2+ sources but the concept shows
sources: 1.
Thin concepts that are correctly thin (genuine 1-source observations):
- PKM Failure Pattern, LLM Wiki Agent — these are the user’s own reflection pages; leave alone.
P1-4. Entity frontmatter is clean.
Spot-checked Donald Trump (sources: 40), Jerome Powell (sources: 16), Mechanical Turk concept (sources: 13, matches ~13 evidence citations). Entity and active-cluster concept frontmatter is disciplined. The drift is specifically in the concept pages that haven’t been revisited.
P2 — Cross-Linking Integrity Spot Checks
I pulled 10 entity/concept pages and verified a few key ones against their actual citation footprint (files that contain [[Page Name]]):
| Page | Claimed sources: | Observed citations in wiki/sources/ | Assessment |
|---|---|---|---|
| Donald Trump | 40 | Not recounted (too many), but Source Appearances list is substantive | Looks accurate |
| Jerome Powell | 16 | Heavy and consistent | Accurate |
| Federal Reserve | 23 | Heavy | Accurate |
| Mechanical Turk Pattern | 13 | Key Sources list has 7+; full text has 10+ inline citations | Roughly accurate |
| Chokepoint Control | 1 | Only cites Iran source | Stale — should be ~4-6 |
| War-Driven Inflation | 1 | Only cites Iran source + The Great Inflation | Stale — should be 2-3 |
| Retroactive Executive Protection | 1 | Lists 3 Key Sources | Frontmatter wrong (should be 3) |
| Infrastructure Warfare | 1 | Cites only Iran | Stale |
| GENIUS Act | 18 | Index shows ~15 GENIUS-titled sources + many that cite it | Roughly accurate |
| Fed Independence | 20 | Strong alignment | Accurate |
Finding: Entity pages are maintained; concept pages from the Iran cluster and the older “first draft” concepts are not. The ingest workflow treats entities as the active index but concepts as write-once.
Orphan check: Did not do a full orphan sweep (would require reading every page), but based on the ~15 files that reference any of Miran/Cook/Sherrill/Spanberger/Mamdani/etc., the overview.md is carrying the wiki’s best cross-link density by a wide margin. Most concept pages have 3-6 outgoing links, most entity pages 5-10. The bottleneck is missing target pages, not missing links.
The Emerging Argument — Does It Hold?
Yes, mostly, with two caveats.
The overview.md (182 lines, thoughtful, tight) argues that the wiki’s theoretical spine is infrastructure-chokepoint + concealment + state-power-without-accountability + conflict-of-interest, with a Schelling layer underneath. Against the material I spot-checked, that holds:
- Chokepoint control genuinely recurs across Strait of Hormuz, DEA quotas (ADHD shortage), FOMC independence, App Store, stablecoin reserves, and NFL salary cap. The cross-cutting pattern at
wiki/overview.md:126is well-earned. - Mechanical Turk Pattern is the cleanest, best-sourced concept in the vault (13 sources, tight concept page, clean entity links to Amazon/Cruise/Google). This is the model for what every concept page should look like.
- Crypto cluster is dense and internally consistent — GENIUS Act, CLARITY Act, Anti-CBDC, Crypto Week all have substantial dedicated pages.
- Fed Independence cluster is similarly dense and the Burns/Volcker historical framing is well-integrated.
Caveats:
-
The Iran cluster is doing heavy lifting for the “infrastructure warfare” thesis but it is only 2 sources of primary coverage (per the index). The overview says “~10 sources” but half of those are coalition-fracture reactions and tangential material. The war is the most-invoked example of the chokepoint thesis and the least-sourced cluster. If this wiki is feeding a newsletter, the Iran coverage is a risk point — one Reuters story is doing most of the analytical work. Action: flag this explicitly in overview.md’s Open Questions rather than obscuring it.
-
The mental health / ADHD / autism cluster doesn’t actually hook into the rest of the argument. It sits on its own with ~30 sources and no cross-link into the chokepoint/infrastructure spine. The overview claims (L87) “DEA Schedule II quota mechanism is the chokepoint” — great argument, but neither Chokepoint Control nor ADHD Medication Shortage explicitly links the two. The wiki has the argument; the wiki doesn’t have the link.
-
The NFL cluster doesn’t hook in either. The overview (L126) says the NFL salary cap is “the same kind of object” as the Strait of Hormuz and the Fed chair. That’s a genuinely interesting claim. It doesn’t appear in Chokepoint Control, Salary Cap Optimization, or Defensive Scheme Architecture. This is the move the synthesis hasn’t made yet — if the cross-cluster spine is real, the concept pages need to cite each other across clusters. Right now each cluster is an island linked only by overview.md.
-
“Concealment as operating mode” (overview L128) is an elegant pattern claim but it has no concept page. The three examples (Mechanical Turk Pattern, CISA Jawboning, Iran dual-track messaging) sit in separate pages with no connective concept. This is the strongest synthesis-that-should-become-a-concept in the vault.
Caveats & Gaps
- I did not do a full cross-link orphan sweep. That would require reading every page and is a 2-hour job; I targeted the overview.md’s highest-leverage references instead.
- I did not verify the
raw:field in source pages against actual files inraw/. The log.md entry shows the user already did a triage pass on April 7, so this is probably in reasonable shape. - Spot-checked 5-6 frontmatter samples rather than all 84+ concept/entity pages. A full schema lint would be a separate pass.
- I could not run counts from the shell; all counts are from Glob and Grep output, which has a 250-file cap in some cases. File counts marked as “~384” etc. are within ±2 of the truth.
Newsletter Application
None of this changes the newsletter output directly. But three things here will affect newsletter quality if not fixed:
- The missing syntheses directory means every past query-driven analysis has either been lost or misfiled. Recovering those from chat history and filing them in
wiki/syntheses/would be a significant one-time content recovery. - The stale concept source counts mean that when the agent writes an article about “chokepoint control” or “infrastructure warfare,” it’s pulling from 1 source instead of 5-10. The raw material is there; the indexing isn’t. Articles are almost certainly thinner than the vault’s actual knowledge would support.
- The missing cross-cluster concept page (“Concealment as operating mode,” or similar) is a potential Substack piece in waiting. The synthesis already exists in overview.md L128 — it just needs a home page.
Follow-up Questions
- Where did the query outputs go before
wiki/syntheses/existed? Are they recoverable? - Should
wiki/index.mdsection counts be automated at the bottom of each ingest as a lint step? - Is there appetite for a one-time “concept-page reconciliation” pass — rebuild Key Sources for every concept from a
[[Concept Name]]grep? Estimated 90 minutes. - Should the overview.md’s cross-cutting patterns (L124-136) be promoted to proper concept pages (“Concealment as Operating Mode,” “State Power Without Accountability,” “Distributed Alternatives as Implicit Argument”)? These are the wiki’s highest-leverage ideas and currently live only in overview.md.
- Are Obsidian sync duplicate files (the
2.mdentries in git status) a recurring issue? They’re not on disk now but showed up in the status snapshot.
Prioritized Action List (What To Do In One Sitting)
If the user has 60 minutes, in order:
- [10 min] Create stub entity pages for Stephen Miran, Lisa Cook, Mikie Sherrill, World Liberty Financial, Israel. Even minimal (3-sentence overview + 1-2 source links) unbreaks overview.md.
- [5 min] Fix
wiki/index.mdfrontmatter counts (total_pages,total_sources) to match reality. - [10 min] Reconcile the Iran-cluster concept page source counts: open each of War-Driven Inflation, Chokepoint Control, Infrastructure Warfare, Coercive Diplomacy, add the 2-3 Iran sources that exist, bump counts.
- [15 min] Cross-link the NFL and ADHD clusters into Chokepoint Control: add a line each citing ADHD Medication Shortage (DEA quota) and Salary Cap Optimization (NFL cap) as parallel chokepoint cases. This actually makes the overview’s thesis visible at the concept level.
- [15 min] Create the “Concealment as Operating Mode” concept page that consolidates Mechanical Turk Pattern + CISA Jawboning + Iran dual-track messaging. This is the highest-leverage missing page in the vault.
- [5 min] Add one line to
wiki/overview.md’s Iran cluster noting that the cluster is thinly sourced and should be expanded.