10 KiB
Lint Report — 2026-07-28
#lint-report
First lint pass on this vault. Scope: all 41 pages in wiki/ plus index.md, checked against the operational contract in CLAUDE.md.
Vault size at time of check: 10 sources · 24 concepts · 14 entities · 4 queries · 1 comparison · 1 timeline · 1 overview.
Headline: structurally the vault is in good shape — zero broken links, zero orphans, zero template violations, zero tag errors. The real defects are all staleness of synthesis: five references to files that moved, four contradictions recorded on only one of the two pages they concern, and two synthesis pages that predate the last two ingests. Twelve findings; nine fixed in this pass, three left as recommendations because they are scope decisions rather than defects.
1. Clean — verified, no action
| Check | Result |
|---|---|
| Page-type tags present, correct for folder, on line 3 | 41/41 pass |
| H1 heading on line 1 | 41/41 pass |
| Source template sections (6 required) | 10/10 complete |
| Entity/concept template sections (6 required) | 38/38 complete |
Evidence section cites ≥1 wiki/sources/* page |
38/38 pass |
Broken [[wiki links]] |
0 |
| Pages with zero outbound links | 0 |
| Strict orphans (zero inbound) | 0 |
index.md concept list ↔ wiki/concepts/ files |
24/24 in sync, no drift |
raw/ paths referenced from wiki resolve |
5 failures — see L1 |
Link-graph health is good. Most-referenced pages: eugene (33 inbound), skills-as-memory (29), context-as-scarce-resource (27), claude-code (24), code-as-throwaway and harness (22 each). Source fan-out median ≈ 17 citing pages.
2. Fixed in this pass
L1 — Stale raw/ paths after the notes move · fixed
The four webinar deliverable docs moved from raw/sources/ to raw/notes/ (commit 1361dd7), but three wiki pages still pointed at the old location. Five references, all now corrected and reworded to say authored deliverable, not not yet ingested — the distinction the move was making.
wiki/concepts/levels-of-ai-usage.md→raw/notes/Webinar script.mdwiki/queries/2026-07-22-webinar-theses.md→ three pathswiki/sources/2026-07-14-sebastian-eugene-interview.md→raw/notes/Ideas for webinar.md
L2 — Stale source count on claude-code · fixed
Summary read "cited across all four ingested sources" — written when the vault had four. Now rewritten to state the actual situation, which is more useful than a number that will rot again: cited across most of the corpus, and used by every practitioner in it except thorsten-ball, who runs amp.
L3 — Dead reference in overview open questions · fixed
The question "should Ideas for webinar / HR Contacts / Webinar Plan / Webinar script be ingested?" named a file that does not exist in the vault. The 2026-07-21 ingest caught this and corrected index.md, but the fix never propagated to overview.md. The question is also now answered — those docs are authored deliverables in raw/notes/. Marked resolved, with the phantom file noted.
L4–L7 — Contradictions recorded on only one side · all four fixed
Rule 5 requires contradictions be recorded explicitly. Four were logged on one page but not on the page holding the opposing view, so a reader arriving from the other direction would see an uncontested claim.
| # | Contradiction | Was recorded on | Was missing from |
|---|---|---|---|
| L4 | Wide latitude vs tight specs | leave-less-room-for-imagination (×2) + overview | think-wider-not-bigger |
| L5 | The skills dissent | skills-as-memory, evolution-of-agent-tooling, thorsten-ball, overview | levels-of-ai-usage — whose top rung is the thing contested |
| L6 | Memory as anti-feature | skills-as-memory, allie-miller | personal-ai-operating-system — whose layer 1 is the thing contested |
| L7 | Skills vs RAG | context-as-scarce-resource | skills-as-memory — which asserts the winner in its Summary as settled |
L5 and L6 are the consequential ones: in both cases the page that makes the contested claim was the page not carrying the objection.
L8 — 2026-07-22-webinar-theses is two ingests out of date · staleness note added
Synthesized from 7 sources; 10 now exist. Preserved rather than rewritten (Update Policy: no silent large rewrites), with a note recording two specific gaps:
- Thesis 3 — "Skills are the new memory" — is load-bearing for the talk and now has a live counter-example absent from its own "Honest tensions" list.
- Nothing from the last two ingests appears as a candidate: make-more-cheap-code, shedding-weight, and especially explosion-of-internal-software, which is the corpus's strongest outside validation of thesis 5 ("you build it, one small tool at a time").
L9 — theo-konstantin-allie predates Theo's second source and Thorsten · staleness note added
Two specific problems, both noted on the page: its reading of Theo's position on code omits the verification discipline he later supplied; and its closing line — "None of the three directly contradicts another; the disagreements in this corpus are elsewhere" — now misleads, because the skills convergence it calls "the vault's strongest cross-source thread" is exactly what the newest source contradicts.
3. Open findings — recommendations, not defects
L10 — "Taste" is the vault's largest uncovered concept · recommend a page
Appears in 12 wiki pages / 17 occurrences, including three entity pages and the vault's one-sentence through-line ("judgment, ownership, taste, and in-person relationships"). It is named as the meta-skill by allie-miller ("knowing what good looks like — you don't need to do the graphic design to judge whether the ad is good") and is thorsten-ball's entire answer to the slop objection ("taste at AI speed" — 15 generated icon variants, one chosen by hand; "slop = lack of ideas, lack of playfulness"). theo-browne supplies the third leg via the ship/no-ship line.
It currently has no page, split across product-ownership and make-more-cheap-code. Given that it is one of four nouns in the through-line and the other three all have pages, this is the clearest coverage gap in the vault.
L11 — Query pages are a weakly-linked class · recommend backlinks
Durable Q&A outputs are not cited back from the concept pages they inform:
| Query page | Inbound (excl. index/overview) |
|---|---|
| 2026-07-22-webinar-theses | 0 |
| 2026-07-14-best-first-skill-for-beginner | 1 |
| 2026-07-14-network-from-standing-start | 2 |
| 2026-07-24-non-engineer-throwaway-verification | 2 |
The theses page is effectively orphaned — reachable only from the catalog — despite being the most directly webinar-relevant page in the vault. Rule 7 asks for at least one inbound link; it has none from any content page.
L12 — 2026-07-22-ai-is-stupid is thinly integrated · watch
Cited by 2 pages against a source median of ~17. Consistent with the 2026-07-22 ingest note that it restates the machine-side spine rather than adding to it — so this may be correct rather than a defect. Worth revisiting only if it stays at 2 after the next ingest.
L13 — Possible future merge · watch
emacsification-of-software and explosion-of-internal-software were created in the same ingest, from the same source, as "sibling mechanisms" of one shift (software becomes personal — by remixing vs by building). Both currently carry distinct evidence and distinct open questions. If neither gains independent support from a second source by the next lint, they should merge.
Also considered and rejected as a new page: trust calibration / what to verify. It touches 15 pages, but is substantively covered by make-more-cheap-code (tier discipline), code-as-throwaway (the auth/payments carve-out) and seniority-and-the-junior-squeeze (the "read what you approve" habit). Creating it would duplicate rather than consolidate.
4. Contradiction inventory
The vault's live disagreements after this pass, all now recorded on both sides:
- Skills as memory vs no skills at all — Konstantin/Allie/Eugene vs thorsten-ball. Three competing readings on skills-as-memory; unresolved, and the evidential asymmetry favours the dissent. The most consequential open question in the vault, because the webinar's central promise rests on it.
- Personal vs company-managed vs vendor-managed harness — eugene vs sebastian vs amp.
- Built-in memory as anti-feature — Eugene vs Allie's untroubled persistent context docs.
- Tight specs vs wide latitude — leave-less-room-for-imagination vs think-wider-not-bigger.
- Diff summaries as sufficient review vs invisible drift — Theo/Dax vs Eugene.
- Model choice as a real lever vs a distraction — Eugene (4.7 over 4.8) vs Thorsten (stop tuning).
- Local consolidated workspace vs local dev disappearing — Eugene vs Thorsten.
- Skills vs RAG as the context mechanism — Konstantin vs the business-facing short.
- Online vs in-person networking — Eugene vs Sebastian.
Items 6 and 7 arrived with the newest source and were recorded at ingest; 1, 3, 4 and 8 were the ones needing reciprocal entries in this pass.
5. Recommended next actions
- Refresh 2026-07-22-webinar-theses against all 10 sources — highest value, since it is the page closest to the actual deliverable and it is both stale and orphaned. Fixes L8 and most of L11 in one operation.
- Create a
tasteconcept page (L10), consolidating Allie's meta-skill, Thorsten's taste-at-AI-speed, and Theo's ship/no-ship discipline. - Extend or supersede theo-konstantin-allie with Thorsten as a fourth lens or explicit dissent column (L9).
- Consider the falsification test already flagged on skills-as-memory: same task, with and without a skill, in a non-engineer's hands. It is the cheapest experiment that would move contradiction #1, and nobody in the corpus has run it.