Three offers, three shapes. The one with the highest number is usually the one you should read last, not first. Rank them by what each does to you in the state where you miss, because that is the state the document was actually written for.
The headline number governs one branch. The instruments govern all the others. Valuation only pays out in the good exit. Liquidation preference, participation, anti-dilution, redemption, board control and milestone tranches pay out everywhere else, and everywhere else is where most companies live.
So here is the reframe. You are not negotiating a price. You are negotiating what happens when you miss.
Why these three models
A term sheet is a machine that produces a different owner of the company in each future state. That is a mechanism-design object, not a price tag. The distressed branch inside that machine does not behave in a straight line: one missed round can trip a chain of clauses that ends somewhere you did not model. That is a threshold-and-cascade object. And the reason the two sides value the same clause so differently is that the investor holds a portfolio of independent bets while you hold exactly one. That is an aggregation object.
Three models, three outcome types: an equilibrium the contract designs, a complex cascade it can trigger, and a random draw you cannot diversify. Two lenses I am folding in rather than shipping as their own cards. The first is the option-value reading, that every downside clause is an option you write to the investor and it is never free even when the price looks generous. It lives inside the mechanism-design card because the mechanism already prices it. The second is founder optimism, the habit of weighting the branch where you win. It explains why the mistake persists but hands you no lever the other three do not, so it sits in the blind spot where it belongs.
The framework
1. The path: the term sheet is a machine with a different setting for each state
The defining feature of a venture financing is that it splits cash-flow rights, voting rights, board rights and liquidation rights apart and hands each one out separately, contingent on how the company performs.1 The same signed document gives you one company if you clear the cap, a different company if you stall, and a third if you break a covenant. Kaplan and Strömberg found the pattern in real contracts: when the company does well the founder gains control and cash-flow rights, and when it does poorly the investor takes full control.1
The clause menu is public. The NVCA model term sheet spells out the standard set: a liquidation preference, weighted-average or full-ratchet anti-dilution, pay-to-play, protective provisions.2 Even the seed-stage post-money SAFE, which reads as founder-friendly, carries downside instructions: on a dissolution the holder takes their money back ahead of you, and on a sale before conversion they take the greater of their cash or their converted stake, ranking as non-participating preferred.3
Read one term through this lens. A participating preference lets the investor take their preference first and then share the rest as if they had converted, the double dip, and the McCarter term-sheet anatomy calls the liquidation preference the single most important economic provision for exactly this reason.4, 2 On a paper you were pricing at a billion, it is invisible. On the exit you will probably get, it decides whether you and your team see anything at all.
Mechanism design.
Assumes: the contract is enforceable and each clause fires in the state it names.
Fits because: VC terms are empirically state-contingent allocations of control and cash, not a single price.
Breaks when: the investor never invokes a right it holds, so the paper machine and the real one diverge.
Counteracts: optimism that reads the good branch as the only branch.
May reinforce: over-lawyering a deal you should walk from on the number alone.
2. The path: the distressed branch is a cliff, not a slope
Founders model the bad state as a smaller version of the good one. It is not. It is a sequence of triggers, and each trigger can move you past a point of no return.
Watch the chain. You raise a down round. Full-ratchet anti-dilution reprices the earlier investor’s stake all the way to the new low, so their share of the company jumps and yours collapses.5 Weighted-average would have softened it. Then pay-to-play asks every existing holder to write another cheque, and the ones who decline convert or lose their protection.2 Below a participation threshold the round does not clear at all. Cross another line and the board flips to the investor, which under a state-contingent contract is exactly what a poor-performance state is built to do.1
This is a Granovetter threshold sitting inside your cap table. Each holder’s decision to support the next round depends on how many others will, and once support falls under the line, the descent feeds itself. The reason you judge an offer on the distressed branch is that a small parameter you waved through, 1x against 1.5x, weighted-average against full ratchet, pay-to-play in or out, is what sets where the cliff edge is.
Threshold and cascade.
Assumes: clauses interact, and one trigger can pull the next.
Fits because: anti-dilution, pay-to-play and control flips are wired to fire together in a down state.
Breaks when: the round is clean and no single term can start a chain.
Counteracts: the linear read of the bad state as a milder good state.
May reinforce: paralysis, if you treat every recoverable stumble as a wipeout.
3. The path: they hold a hundred draws, you hold one
Venture returns follow a power law. A small share of a fund’s bets carries almost all of its return, and most of the rest come back at or below cost.6 The investor across the table is aggregating many independent draws. You are one of them, and you are undiversified in a way they never are.
This is why the two of you price the same downside clause so differently. To the fund, a liquidation preference is a harvest mechanism that scrapes value out of the many small and failed outcomes to lift the blended return. To you, it is the whole of your one outcome. A liquidation preference is a rounding error in their portfolio and an extinction event in your life.
The African exit map makes the point concrete. Most liquidity here arrives as a modest trade sale to a bank, a telco, an insurer or a regional buyer, with public listings almost absent.8 When the exit is small, the preference stack, not the equity percentage, decides who gets paid. And the instrument matters twice over: roughly a billion dollars of 2024 African venture funding came as debt, whose bad-state terms, acceleration, security, a director’s guarantee, behave differently again, and the lender is aggregating across a loan book while you carry the single obligation.7
Independent-source aggregation.
Assumes: the investor’s outcome is a portfolio average and yours is a single realisation.
Fits because: power-law returns make downside clauses cheap protection for a fund and total exposure for a founder.
Breaks when: your counterparty is concentrated too, such as a single-deal angel or a strategic buyer.
Counteracts: the instinct to accept a term because it looks small to the person offering it.
May reinforce: treating every investor as adversarial when incentives sometimes align.
GEER: read every offer from the bad state up
Start with the cheapest, most reversible move and work toward the costly one.
Build the waterfall before you read the cap. For each offer, compute who gets what in a small trade sale, the outcome the base rate says is most likely. Costs nothing but an afternoon. It routinely flips the ranking: the higher-priced offer with a participating preference and full-ratchet protection pays you less below the line than the lower-priced clean one.
Ask for the three standard swaps. A 1x non-participating preference over a participating one. Broad-based weighted-average over full ratchet. A redemption date pushed far out, and cumulative dividends capped or removed. Each is a normal ask against the model documents, not an act of aggression.2, 4
Name the triggers yourself. If there is a milestone tranche or a pay-to-play, write the observable that fires it in plain terms and set it where you can actually clear it. You define the line, or they will define it for you.
Then, and only then, trade on price. The expensive, less reversible move is to give up headline valuation in exchange for a clean stack. Do it deliberately, once you can see what the clean stack is worth in the branch you are likely to land in.
RADAR: what to settle before the signature
Do now (T+0 to T+3). One page per offer, showing the payout to founders and team at a small exit and at a down round. Rank the offers by that page, not by the valuation line. This is reversible, it dominates in every scenario, and it changes which deal you want.
Hedge (T+3 to T+14). Buy the cheap tail insurance while you still have leverage. Cap participation. Hold anti-dilution at weighted-average. Set the redemption clock to start years out. Get pro-rata and information rights in a side letter. Each is inexpensive today and load-bearing in the state you are underwriting.
Defer and trigger (T+14 to T+28). The irreversible concessions, a participating preference, full ratchet, a personal guarantee on venture debt, should not be accepted on hope. Pre-commit the observable that would justify each: sign only if a named condition holds, for example a credible path to clearing the cap inside a set number of months, and walk if it does not. Write the trigger down before the pressure of the closing week rewrites it for you.
CHAIN: how these terms usually read out
Match the reference class on structure, not on sector. The right comparison is not “African seed rounds.” It is financings where the founder optimised the headline number, conceded the downside terms, and then landed in a flat or falling state. That set is large.
The base rate is unkind. Most venture bets return at or below cost for the investor,6 and most exits on this continent are modest trade sales rather than the outcome the valuation implied.8 The flat and distressed branches are the modal case, not the tail. Present-state modifiers push the same way: African venture deal value fell 28 percent in 2024 and a large slice of capital now arrives as debt, so down rounds, anti-dilution and guarantees are live conditions today, not hypotheticals.7
Now subtract the counterfactual. The good exit you were picturing pays out the same under a clean stack or a punishing one, because in that branch everyone converts and the preferences fall away. So the term you conceded bought you nothing in the branch you imagined, and it costs you everything in the branch you are most likely to get. That subtraction is the whole argument.
Matrix-break flag. All of this assumes a take-it-or-leave-it. If you hold real leverage, several clean offers, live revenue, a debt alternative you can actually use, the downside terms become negotiable and the equilibrium moves in your favour. Competition rewrites the machine. Manufacture it before you sit down.
The part the contract cannot tell you
The models read the paper, not the person. A tough term sheet in the hands of an investor who never invokes its worst clauses can be gentler than a clean one held by someone who will use every lever they have. The instrument sets the boundary of what can be done to you. It cannot tell you what will be done. Reputation, relationship and the investor’s own incentives fill that gap, and none of them appear in the document.
That uncertainty does not release you from a decision, it sharpens one. Because you cannot know which investor will invoke the bad-state clauses, price every offer as if they will. Before T+28, put the distressed-branch waterfall for all three offers on a single page, and sign the one whose bad state you can still run a company inside. Rank the offers by that page. Everything above the line was already going to be fine.
Sources and notes
- Steven N. Kaplan and Per Strömberg, “Financial Contracting Theory Meets the Real World: An Empirical Analysis of Venture Capital Contracts,” NBER Working Paper 7660. The paper documents that VC financings separately allocate cash-flow, voting, board and liquidation rights contingent on performance, and that poor performance shifts full control to the investor. nber.org/papers/w7660
- National Venture Capital Association, Model Term Sheet. The standard clause menu: liquidation preference, weighted-average and full-ratchet anti-dilution, pay-to-play, protective provisions. nvca.org (Model Term Sheet, .doc)
- Y Combinator, “Primer for the Post-Money SAFE.” Sets out the dissolution and liquidity-event provisions: the holder recovers its purchase amount ahead of common on a dissolution and takes the greater of cash or converted value on a sale, ranking as non-participating preferred. ycombinator.com (post-money SAFE primer, PDF)
- McCarter & English, “Anatomy of a Term Sheet: Series A Financing.” Explains participating versus non-participating liquidation preference and names the liquidation preference the most important economic term. mccarter.com (Anatomy of a Term Sheet, PDF)
- Jonathan Lea Network, “A Startup’s Guide to Ratchets and Other Anti-Dilution Provisions.” Describes how a down round triggers anti-dilution, and how full ratchet reprices the earlier investor to the new low while weighted-average softens the adjustment. jonathanlea.net
- BIP Ventures, “Explainer: What is the Venture Capital Power Law.” Describes power-law return concentration, where a small fraction of a fund’s investments drives the majority of returns and diversification shapes the strategy. bipventures.vc
- AVCA, “2024 Venture Capital in Africa Report.” African venture deal value fell 28 percent in 2024, and venture debt deals worth roughly US$1bn formed a large share of the year’s funding. avca.africa
- TechCabal Insights, “The State of Startup Exits in Africa in 5 Charts.” Trade sales and mergers dominate African liquidity, with more than 50 M&A deals recorded in 2025 and public listings almost absent. insights.techcabal.com