Lessons

The first layer of reverse engineering — the file format on disk and the loader that maps it into memory. ELF, PE, and Mach-O as variants of the same idea; sections vs segments; symbol tables and what stripping actually removes; PLT/GOT/IAT and dynamic linking; packers like UPX; and the triage you do before opening a disassembler.

What a linter actually is, what its rules enforce, how AST-based analysis works, the cost of auto-fix, the ecosystem (ESLint, Ruff, Clippy, golangci-lint), where to wire it in (LSP / pre-commit / CI), why false-positive rate is the real budget, and when a custom rule pays for itself.

The structure of a purchase from first exposure through repeat. The classical funnel and its loyalty-loop extensions, consideration sets, post-purchase cognitive dissonance, the expectation/perception model of satisfaction, the retention vs acquisition cost asymmetry, and the actual mechanisms behind repeat buying — habit, switching cost, identity, and trust.

The motivation theories that try to explain consumer demand: Maslow's hierarchy and its critique, self-determination theory's three needs, the jobs-to-be-done reframing, hedonic vs utilitarian goods, identity signaling and Veblen demand, and the social-norm pull. Each presented as a lens with a clearly stated empirical track record.

The cognitive mechanisms behind consumer choice: dual-system thinking, the durable heuristics, anchoring, prospect theory's loss aversion and S-shaped value function, diminishing sensitivity, the contested choice-overload finding, and the default effect. Frames each as a mechanism with its replication status, not a marketing recipe.

SQLAlchemy 2.0 from the production angle. The engine/session/transaction layering, the typed declarative, the identity map, the N+1 query problem, async-only gotchas (no lazy loading), savepoint nesting, and the connection-pool knobs that decide whether the backend survives load.

Dependency injection as a pattern, FastAPI's Depends as one concrete implementation. Sub-dependencies and per-request caching, yield-based cleanup, where Pydantic v2 validation runs, lifespan-scoped resources, BackgroundTasks vs real queues, and the override trick that makes the whole thing testable.

Python's type system has two audiences — static checkers and runtime frameworks. Generics, Protocol, TypedDict, ParamSpec, type narrowing, and runtime introspection, framed as the spec that Pydantic, FastAPI, and SQLAlchemy actually execute.

Async Python from the bytecode up. Coroutines vs return values, how the event loop schedules tasks, when async beats threading or multiprocessing, structured concurrency with TaskGroup, and the cancellation rules that production backends live or die by.

How the proceeds of an acquisition or IPO are distributed across the cap table — the liquidation-preference stack from senior preferred to common, the participating-vs-non-participating choice, worked examples with multiple preferred classes, and why the same headline exit value can produce very different per-share outcomes.

How the employee option pool is sized and refreshed, what the 409A valuation determines about option strike prices, the tax-treatment differences between ISOs, NSOs, and RSUs in the US, and the parallel structures (BSPCE in France, EMI in the UK) that achieve similar incentive alignment.

What the specific clauses in a venture term sheet actually do — liquidation preferences and participation, board composition, protective provisions, pro-rata, drag-along, registration rights — and the difference between standard market terms and the 'dirty' terms that signal a stressed deal.

How convertible notes and SAFEs let early investors put in capital without setting a valuation, the math of the valuation cap and the discount, how the conversion at the next priced round actually computes, and why post-money SAFEs are now more common than pre-money.

How a priced equity financing actually closes — the price per share calculation, the relationship between pre-money valuation and ownership, the mechanics of anti-dilution provisions, and what 'full ratchet' and 'weighted average' mean for the cap table after a down round.

What a capitalization table records, the difference between issued and fully-diluted shares, how new issuance dilutes existing holders, and the math that makes 'I own 10% of the company' a more or less meaningful claim depending on which denominator you use.

Why an electrical grid is one giant synchronous machine, what frequency stability and rotational inertia actually mean, how dispatch ordering and ancillary services keep the lights on, and why variable renewable integration is fundamentally an engineering problem at the system level.

Why photovoltaics have a hard thermodynamic ceiling (Shockley-Queisser), why wind power scales with the cube of velocity (Betz), how lithium-ion chemistries actually differ, the learning-curve mathematics that produced the cost declines, and where the materials supply chains concentrate.

What it takes to fuse hydrogen isotopes — the four conditions (temperature, density, confinement time, energy gain), the three main approaches (magnetic, inertial, magnetized target), the engineering problems (tritium, neutrons, materials) that remain after the physics is in hand.

How fission releases energy from heavy nuclei, why neutron moderation determines reactor design, the structural choices behind LWR, CANDU, gas-cooled, fast, and molten-salt reactors, what the fuel cycle actually consists of, and where waste, cost, and safety arguments sit.

How hydrocarbon combustion releases energy as heat, the four canonical thermal cycles that convert that heat into mechanical or electrical work (Rankine, Brayton, Otto, Diesel), why combined-cycle gas plants reach 60% efficiency, and how carbon intensity scales with cycle and fuel choice.

The handful of quantitative concepts that make every energy debate readable — joules and watts, energy density, capacity factor, levelized cost of energy, exergy — and what each one is good and bad at communicating.

How to read a clinical trial result with discipline — the difference between absolute and relative risk reduction, what number-needed-to-treat captures, what confidence intervals actually mean, the hierarchy of evidence quality, and why statistical significance is not the same as clinical importance.

The phased structure of drug development from preclinical work through phase IV surveillance, what each phase actually establishes, the 90% attrition rate and where it lives, the difference between surrogate and hard clinical endpoints, and what regulatory approval pathways guarantee.

What the GLP-1 hormone does in normal physiology, why agonists of its receptor produce effects on glucose, gastric emptying, satiety, and weight, how peptide engineering achieves week-long duration, and what the clinical trial evidence shows beyond glycemic control.