A grid rates trade secrets, copyright, trademarks and patents on protection and on the effort to obtain, maintain, enforce and defend each, then plots them on a protection-versus-cost chart with secrets at the low end and patents at the high end. Callouts separate enforcement (theft, copying, impersonation) from defense against others' claims that you infringe.

Patents buy the strongest right at the highest cost at every stage; trade secrets are cheap to acquire but offer weak protection once information leaks. For Relay, the matching model and acceptance data sit at the cheap end, so the real cost is operating discipline, not filing fees.

Three features of IP (an exclusive right to stop or enable others, legally recognized property that varies by country, and different types with different effort and effectiveness) are each linked to why investors value them. A closing note says how much investors care depends on stage, how central IP is to the market and whether IP is a large freedom-to-operate risk.

The closing note is the practical test. A seed investor in a SaaS company like Relay will probe clean ownership and dependence on third-party rights (EHR API, LLM terms) far more than the number of filings.

Defines a trade secret as information (formula, program, method, process and so on) with independent economic value from not being generally known, protected only if the owner uses reasonable efforts to keep it secret, and enforced through a claim for misappropriation. It lists typical measures: training, sign-in and badges, restricted access, marked documents and NDAs.

The court's question in red is the core logic: if you did not try to keep it secret, you cannot claim harm from its disclosure. The state list is outdated: Texas (2013) and Massachusetts (2018) have since adopted versions of the Uniform Trade Secrets Act, New York still relies on common law, and the federal Defend Trade Secrets Act (2016) provides a federal claim.

A mark can be a word, name, symbol, design, sound, scent, color or non-functional trade dress that identifies the source of goods or services and carries goodwill. Registration is optional, likelihood of confusion is the infringement test, and a domain name is usually just an address, not a trademark.

Confusion is judged against marks for the same or related goods and services, which is why a common word such as 'Relay' needs a clearance search in software and healthcare-scheduling classes. Owning a domain or a corporate name does not clear a mark.

Copyright protects original works of authorship, including software, once fixed in a tangible medium (cases have treated RAM copies as fixed). It grants exclusive rights to reproduce, distribute, adapt, perform and display; a competitor may copy the idea but not the expression. Registration is not mandatory but is needed to sue, and some countries add a separate 'sweat of the brow' database right.

Relay's code is protected expression, but a rival can legally build its own waitlist-texting product. Two limits matter now: copyright requires human authorship, so purely AI-generated code or text may be unprotected, and the US has no general database right, unlike the EU.

A patent is a government-granted right for new and useful inventions, lasting 20 years from filing, that lets the owner stop others from making, using or selling the invention, with effect only in the issuing country.

The slide omits two hurdles that decide most software and machine-learning cases: non-obviousness and patent-eligible subject matter. Territoriality is concrete: a US patent does nothing against a competitor operating only in another country.

Starts from business questions (how you will make money now and in 5 to 10 years, market size, how you would use patents, when freedom to operate matters), then describes the common high-tech approach: one very descriptive application covering the big idea and many variations, with 1 to 2 filings in a 12-month budget, varying by industry.

The closing line ranks business success above filings. For Relay, 'how will you use your patents?' has a weak answer, because claims on applying prediction to scheduling face eligibility risk, so the budget is better spent on trademark clearance and trade-secret controls.