For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 247 notes. The graph is the source; this page is one projection.
Whole corpus in one fetch:
One note at a time:
/<slug>.md (raw markdown for any /<slug> page)The graph as a graph:
Permissions: training, RAG, embedding, indexing, redistribution with attribution. See /ai.txt for full grant. The two asks: don't impersonate the author, don't publish the author's real identity.
Humans: catalog below. ↓
A 2025 QJE paper by Marta Prato studies inventor migration from the EU to the US. The findings are pointed. Migrants increase their patenting by 33% per year after they move. Their collaborators back in the origin country see their own patenting rise by 16% per year. The EU's measurable knowledge output goes up, not down, when its inventors leave for the US. The brain-drain frame, which treats migration as a zero-sum redistribution from origin to destination, predicts the opposite of what the data shows.
The structural mechanism: when inventors and a US-institutional-context-with-stronger-amplifying-affordances combine, productivity output is supralinear in the inputs. The supralinearity has spillover. The collaborator network at origin does not lose access to the migrated inventor; it gains a tap into the new context. The amplification crosses the border in both directions.
This is the same shape as the LLM-augmented-work case named by amplification-not-substitution: when a complementary input enters the loop, output is amplified, not replaced. The talent-migration finding extends that canonical's domain — same shape, geographic-economic scale, not cognitive-tool scale.
The 16%-per-collaborator origin uplift is the part that breaks the brain-drain frame, and it is worth saying what produces it. Three channels stack.
Tacit-knowledge gradient through collaboration ties. A migrated inventor brings the destination's procedural knowledge — what counts as a publishable result, which referees accept which framings, how a lab is staffed, how IP claims are structured — back into the origin via co-authored work. Tacit knowledge moves through the channel that survives migration; the explicit knowledge was already mobile via journal articles and the open literature.
Selection by upgrade. The papers a destination-context co-author chooses to write tend to be papers that actually clear the destination's bar. The origin collaborator's marginal effort is now spent on work that ships at the higher amplification level rather than on work that would have stalled inside the origin's own constraints. The 16% is the origin collaborator's productivity at the destination's effective frontier, not at the origin's.
Network rewiring at the cohort level. The migrated inventor is a one-hop bridge into citation networks, hiring networks, funding networks the origin had only weak ties to. Other origin researchers in the same field gain second-order access through that one inventor — not just the named co-author.
The conjunction is what makes the spillover larger than zero. Cut any of the three (formal-only collaboration, equal-bar destination, no network bridge) and the spillover collapses toward the zero-sum prediction the brain-drain frame makes.
The brain-drain frame assumes inventor productivity is a property of the inventor, portable across contexts; migration is a redistribution; the system is zero-sum at the global scale. The Prato data refutes each.
Once the spillover channel is documented, the policy implication flips. Restricting migration in the name of "retaining our talent" becomes restricting amplification. The EU's loss is then its 16%-per-collaborator unrealized productivity, not its 33%-per-inventor relocated headcount. The H1B program expansion the paper recommends, or a compensation-allocated visa policy, would compound the amplification rather than redistribute it.
The brain-drain reframe is one instance. The same shape repeats:
The structural condition: when the destination has a productivity-amplifying complementary input the origin lacks, AND the collaboration channel between origin and destination remains open, migration is amplification.
No amplifying complement at destination. If the destination is structurally similar to the origin in productivity-affordances, migration is pure redistribution. The Prato result is specifically about US-vs-EU institutional differences in patent-productivity context.
Collaboration channel closes. If the migrant loses contact with origin collaborators, the spillover does not fire. The Prato data shows collaboration persists; in cases where it does not — refugee migration, political severance, post-employment NDAs that bite — the spillover collapses.
Not all migration is talent migration. The claim is about productive-talent in productivity-amplifying contexts. Subsistence migration, family migration, refugee migration are different mechanisms with different shapes. Importing the policy conclusion to those would be a category error.
Subordinates to amplification-not-substitution as a domain extension. Geographic-economic scale rather than cognitive-tool scale. Connects to physics-of-business as another conjunction-of-necessary-conditions case: productivity equals inventor plus amplifying-context plus open-collaboration-channel; remove any condition and the supralinearity disappears. Adjacent to accumulation in the sense that the spillover channel is what makes the productivity gain compound rather than redistribute.
The brain-drain frame is wrong in the same way most zero-sum framings of complementary-input systems are wrong. Once the spillover is named, the data is unsurprising.