Three instruments, three different sets of obligations. Treating them as one document type is the most common reason a token cap table stops reconciling six months after it was built.
A Simple Agreement for Future Tokens is a promise to buy tokens that do not exist yet. What it fixes: the amount paid now, the valuation cap, the discount, and usually the allocation and vesting. What it does not do: transfer anything today. The token side is contingent on a TGE that may be years away.
The number people look for is the cap. The number that actually drives the outcome is the vesting schedule, and it is usually buried in an annex.
A warrant is the right — not the obligation — to buy tokens at a set price, within a window, on stated conditions. It shows up two ways: attached to an equity round as a sweetener, or issued on its own to a strategic investor.
Four things a warrant has that a SAFT usually does not: an exercise price, an expiry date, exercise conditions (a TGE, a listing, a valuation event), and sometimes a cashless exercise mechanic. Miss the expiry date and the warrant is worth zero; miss the cashless mechanic and you will overstate what the holder paid in.
A side letter modifies a document you already signed. In token deals it usually carries one of four things: most-favoured-nation protection, pro-rata rights, information rights, or a changed vesting or unlock schedule.
This is the document that quietly breaks cap tables. An MFN clause can retroactively improve the economics of an earlier round. If your spreadsheet was built from SAFTs only, it is wrong, and nothing in the SAFT will tell you so.
| QUESTION | SAFT | WARRANT |
|---|---|---|
| Money already paid? | Yes | No — paid on exercise |
| Has an expiry? | No | Usually yes |
| Valuation cap? | Usually | Sometimes |
| Counts as invested capital? | Yes | Only once exercised |
Any spreadsheet that puts "amount" in one column and mixes SAFTs with unexercised warrants is adding money that has not been paid to money that has. The fix is a separate column for instrument type and a separate column for "committed" versus "exercised".
Extract every document as its own type, keep the raw clause reference, and only then merge. Do that and the cap table survives the next side letter. Skip it and you will be reconciling by hand at the worst possible moment.
Try it on one document, free →
Related: extracting SAFT terms · building the cap table