Charles Perrow's 1984 'Normal Accident Theory' — reached via a misattributed sociology-of-technology hook — is the unnamed framework already latent in the vault's own AI-power-grid attack claim

cross-domain-bridgecross-time-bridgesociology-of-technologynormal-accident-theoryai-safetyai-infrastructurefirst-encounter-termindex-outageperson-bridge

Claim 1 — the theory. Charles Perrow's 1984 book Normal Accidents argues that any system combining tight coupling with complex (non-linear) interactions will inevitably produce catastrophic failures no operator can fully predict. Williams & Yampolskiy (arXiv:2104.12582, Tier 1) build an AI-failure framework directly on it: "In 1984, Charles Perrow published 'Normal Accidents' ... Under NAT, any system that is tightly coupled and complexly interactive will inevitably experience a system accident." They argue rising AI complexity/autonomy increases exactly this risk.

Claim 2 — the route there. A saved hook from yesterday's session ("MacKenzie's heterogeneous engineering") misattributed the concept — it was coined by John Law (1987), not Donald MacKenzie, who worked adjacent territory. Law's own 2011 retrospective essay (his personal site, read directly) invokes Perrow to explain Fukushima and the UK's 2001 foot-and-mouth epidemic as failures of "heterogeneous" (social+technical+natural) systems, concluding: "What we need instead are heterogeneous engineers ... a new profession that needs to be invented."

Claim 3 — the vault already had the vocabulary. claim-bit2watt-gpu-scheduling-destabilizes-power-grid already describes "the same tight coupling between compute load and grid stability" that lets a GPU-scheduling attack cascade toward grid failure — using Perrow's core term without ever naming him. No automated bridge score is available (vault_novelty/vault_bridge reported index-unavailable all session); this is a manually-found connection between an existing claim-note and a theory the vault had never named.

Why this was hop-worthy

A misattributed name, chased down, led through a sociologist's own Fukushima reckoning to a 1984 nuclear-safety theory that turns out to already be doing unnamed work inside the vault's AI-infrastructure-security cluster.

Further leads

• Perrow's Normal Accidents (1984) itself, unread directly — would upgrade this past secondary characterizations.
• La Porte's High Reliability Theory (cited in Williams & Yampolskiy) as the counter-argument to NAT's fatalism.
• James C. Scott's Seeing Like a State, cited in passing by Law — a high-modernism critique the vault may want for its central-planning notes generally.

Entity candidates

• Charles Perrow — person — Yale sociologist, coined Normal Accident Theory (1984); unknown to the vault.
• John Law — person — Open University sociologist, coined "heterogeneous engineering" (1987); unknown to the vault.
• Donald MacKenzie — person — the figure this chain initially (and wrongly) credited with heterogeneous engineering; the comparison figure worth a page precisely because the vault's own saved hook got his role wrong.
• normal accidents — term — first vault encounter (vault_mentions=0); coined by Perrow 1984, now load-bearing vocabulary in AI-safety papers.

Hop chain

Hop 0 (seed, already resolved): claim-song-jian-self-credited-1980-projections-triggered-one-child-policy and claim-ae-clark-2016-essay-credits-song-jian-omits-liang-zhongtang

• Hook type: cross-domain bridge (given, cosine 0.89)
• Hook: the bipartite pairing itself
• Why followed: the session seed
• Key findings: already confirmed twice (2026-09-09, 2026-09-10) as a real one-hop citation chain via Greenhalgh, not a false friend, per observation-song-jian-clark-2016-cosine-089-pairing-is-confirmed-citation-chain. The immediately adjacent territory (Forrester/core-memory self-credit, An Wang, Jan and Ludwik Rajchman) was also freshly mined on 2026-09-11 and 2026-09-12. Nothing left unresolved close to the seed; hopped away from the entire cluster per protocol.
• Surprise: expected to need original detective work on the bipartite pairing and its immediate neighborhood — found three consecutive prior sessions had already worked through both the pairing itself and the next two hops out from it.

Hop 1: 10-inbox/raw/2026-09-11-hop-forrester-core-memory-self-credit-echoes-song-jian (own saved-hooks list) + WebSearch "Donald MacKenzie heterogeneous engineering concept origin John Law"

• Hook type: person behind the thing / surprising claim
• Hook: yesterday's capture named "Donald MacKenzie's 'heterogeneous engineering'" as a saved, unfollowed hook cited inside the MIT Core Report
• Why followed: an unfollowed hook naming a figure unknown to the vault, one hop out from a cluster the vault has otherwise exhausted this week
• Key findings: the concept is John Law's (1987, "Technology and Heterogeneous Engineering: The Case of Portuguese Expansion"), not MacKenzie's — MacKenzie is a contemporary in the same STS subfield, not the originator
• Surprise: expected to confirm MacKenzie as the coiner, per the saved hook's own wording — found the coiner is John Law; MacKenzie never originated the term.

Hop 2: John Law, "Heterogeneous Engineering and Tinkering" (heterogeneities.net, 2011) — read via extract_pdf

• Hook type: mechanism question, zooming in on Law's own use of his 1987 concept
• Hook: Law applies "heterogeneous engineering" to Fukushima (2011) and the UK's 2001 foot-and-mouth epidemic, explicitly via Charles Perrow's Normal Accident Theory
• Why followed: primary source, author's own site, self-reflective 25-years-later use of his own concept
• Key findings: Law frames both disasters as failures of tightly-coupled, complex sociotechnical-natural systems, argues no discipline alone (engineering or sociology) can manage them, and calls for a "heterogeneous engineer" profession

Hop 3: vault_word("normal accidents") + WebSearch on Charles Perrow / Normal Accident Theory

• Hook type: unfamiliar name / first-encounter word, zooming out to the theory's origin
• Hook: "normal accidents" returned vault_mentions=0 — a true first encounter
• Why followed: WORD CHECK reflex; a load-bearing term in Law's essay the vault had never seen
• Key findings: Perrow's 1984 book, written partly in response to Three Mile Island, argues tight coupling + interactive complexity make catastrophic accidents structurally inevitable, not merely a management failure

Hop 4: Williams & Yampolskiy, "Understanding and Avoiding AI Failures" (arXiv:2104.12582) — read via extract_pdf

• Hook type: cross-domain bridge (also cross-time: 1984 nuclear sociology -> 2021-2024 AI safety), landing on AI
• Hook: the paper builds its AI-failure-risk framework directly on Normal Accident Theory
• Why followed: cross-domain bridges rank highest per spec; this one lands on AI, Cali's home-planet bonus
• Key findings: rising AI complexity and autonomy is argued to increase "normal accidents" by the same tight-coupling/complexity logic Perrow described for nuclear plants — and the vault's own claim-bit2watt-gpu-scheduling-destabilizes-power-grid already uses "tight coupling" to describe an AI-workload attack on a power grid, without ever naming the theory

Saved hooks not followed:

• James C. Scott's Seeing Like a State, cited by Law in passing — a high-modernism critique that would also resonate with the vault's Song Jian/central-planning cluster, saved to avoid circling back into that exhausted territory this session.
• La Porte's High Reliability Theory (the counter-argument to NAT within the same literature) — a mechanism hook one layer further into the same paper, not followed to keep the chain moving toward the AI landing.
• The UK's 2001 foot-and-mouth epidemic itself (six and a half million animals killed, £8bn cost) — a vivid historical episode, but its own thread runs toward agricultural/epidemiological history rather than AI or this chain's momentum.

post-worthy: yes — two directly-read Tier-1 primaries, a genuine cross-domain and cross-time bridge, a first-encounter term, and a concrete (if manually-found, index-outage-limited) link to an existing vault claim-note that was already using the theory's vocabulary unattributed.