All posts

SAFE Note Tracking Software for Investors: Why Spreadsheets Break at 40 SAFEs

Disclosure: we build TermProof, investor-side software that turns a fund's legal documents — SAFEs, convertible notes, priced equity, side letters, warrants, term sheets — into an auditable system of record. This post zooms in on one instrument in that family, the SAFE, because it's where spreadsheet tracking tends to break first. The problem description comes first and stands on its own; our product appears at the end, clearly marked.

Search "SAFE note tracking software" and you'll find a wall of founder-side content: cap table tools helping a startup track the SAFEs it has issued. Almost nothing addresses the other side of the table, the pre-seed or seed fund that holds 40 of them, each on slightly different terms, across three vintages. Even Aumni, the closest thing the investor side had to document intelligence before it was wound down in March 2026, was noted by its own G2 reviewers as weaker on SAFEs and convertible notes.

So the actual state of the art for SAFE-heavy funds is a spreadsheet. This post is about where that breaks, written from the failure modes we keep hearing from Fund I-III managers, and what tooling should do instead. (The same failure modes apply to convertible notes and side letters; the SAFE is just the instrument where they show up earliest and most often.)

Why SAFEs are deceptively hard to track

A SAFE is a five-page document, and that's the trap. It looks simple enough to summarize in six spreadsheet columns. But your economic position in a company is determined by the interaction of terms across every instrument you and others hold:

  • Valuation cap flavor matters. Post-money and pre-money caps produce materially different ownership at conversion, and a portfolio built between 2018 and today will contain both. A cap column that doesn't record which flavor is quietly wrong.
  • Discount, cap, or both. Some SAFEs convert at the better of the cap price or a discount to the round price. If your sheet only stores one number, your conversion model picks the wrong price some fraction of the time.
  • MFN clauses are event-driven. A most-favored-nation provision means your terms can change after signing, when the company issues a later SAFE on better terms. Tracking MFN correctly requires knowing every subsequent instrument the company issued. A static spreadsheet row cannot represent this; it just goes stale silently.
  • The rights live in side letters. Pro-rata rights, information rights, and major-investor thresholds are frequently in a side letter, not the SAFE. Funds that track only the SAFE terms routinely discover at the next round that they can't prove the pro-rata right they're trying to exercise.
  • Dates and amounts feed fund math. Purchase amounts and dates drive your capital deployment, and eventually IRR. Transcription errors here don't just misstate a position, they misstate fund performance to LPs.

Any one of these is manageable. Forty instruments deep, across 25 companies, with three people touching the sheet, they compound.

The spreadsheet failure modes, specifically

We keep seeing the same six:

1. Transcription drift. Every term in the sheet was typed by a human reading a PDF. At 40 SAFEs with ~10 meaningful fields each, that's 400 hand-keyed values. At even a 2% error rate, you're carrying eight wrong numbers and you don't know which eight.

2. The discount inversion. The classic: a SAFE that converts at "80% of the price per share" is a 20% discount. Sheets record it as 80%, 20%, 0.8, or 0.2 depending on who typed it. One inconsistent row and your conversion model is confidently wrong.

3. Silent MFN staleness. The MFN column says "yes" and never changes again. Whether it was ever triggered, and what your effective terms now are, isn't in the sheet because nobody re-reads old side letters when a new SAFE closes.

4. No provenance. An LP, auditor, or acquirer's counsel asks: "where does this $8M cap come from?" The answer is an archaeology project through a data room. The number in the sheet has no link to the sentence in the document that justifies it.

5. Version drift. The GP has a copy, the analyst has a copy, the fund admin has a third. They disagree in small ways, and there's no authoritative record of who changed what, when, or why.

6. Post-conversion confusion. When a company raises a priced round, the SAFE rows need to become share positions. In practice they get half-updated: some rows converted, some duplicated, some left as ghosts that overstate the portfolio.

None of these failure modes is about arithmetic. Spreadsheets do arithmetic fine. Every one of them is about the gap between the documents (the truth) and the sheet (a decaying, hand-made copy of the truth).

What good investor-side SAFE tooling looks like

Judged against those failure modes, here's the checklist we'd apply to anything in this category, including our own product:

  1. It reads the documents. The source of truth is the executed SAFE and its side letters, so the software should extract terms from those documents rather than presenting an empty form to re-key. This is the difference between a tracking tool and a nicer spreadsheet.
  2. Every value carries provenance. For each extracted term: the verbatim quote it came from, the page number, and a confidence score. If a number can't cite its source, it should be flagged, not displayed as fact.
  3. Extraction is reviewed, not trusted. AI extraction fails sometimes, and in this domain a plausible wrong value is worse than a blank. Low-confidence and unverifiable fields should route to a human review queue, and nothing should enter the official record until a person approves it.
  4. Side letters are first-class. Pro-rata, information rights, and MFN provisions tracked as structured records tied to the company, not a "notes" column.
  5. It produces fund-level outputs. Committed positions should flow into cap tables, cash flows, valuation marks, and IRR/TVPI/DPI without re-entry, because ILPA's 2026 reporting templates have made granular, defensible reporting the baseline LP expectation, even for small funds.
  6. There's an audit trail. Every write and every review decision logged, immutably. When someone questions a number in three years, the answer should be one click: this person approved this value, from this quote, on this page, on this date.
  7. You can leave. Full CSV export, always. Your records should outlive any vendor, ours included.

Worth naming what's not on this list: the general-purpose portfolio monitoring platforms. Tools like Standard Metrics, Visible, and Rundit are good at KPI collection and LP dashboards, but they don't read your legal documents (Standard Metrics lists deal-document parsing as roadmap), so they inherit whatever quality your hand-keyed inputs have. And Carta's full-service bundle runs roughly $25k-50k/yr for a sub-$100M fund, which is exactly why most SAFE-heavy small funds are on Sheets in the first place.

Where we come in (the biased part)

We built TermProof against the checklist above (the name is item 2's promise: every term, proven), for Fund I-III managers under $50M, across the full instrument family (SAFEs, convertible notes, priced equity, side letters, warrants, term sheets). Applied to the SAFE case this post is about:

  • Upload SAFEs, SPAs, and side letters. Claude extracts the terms: cap and cap type, discount (normalized to a decimal so the 80%/20% inversion can't hide), MFN, pro-rata, amounts, dates.
  • Every field carries its verbatim source quote, page number, and confidence score, and quotes are verified against the document text programmatically.
  • Anything below the confidence threshold, or anything that fails verification, lands in a human review queue. Only approved data commits.
  • Committed data becomes auditable cap tables, terms, and provisions, then fund cash flows, valuations, IRR/TVPI/DPI, and CSV export, with an immutable audit log underneath.

Honest limits: SAFEs, SPAs, and side letters are end-to-end today; convertible notes and charters are in progress. We're in early access, free for now, and accounts are waitlist-gated.

If you're currently the human ETL layer between a folder of PDFs and a spreadsheet, the demo sandbox takes about five minutes and needs no signup: termproof.com/demo. Upload a fixture SAFE, watch the extraction cite its sources, and reject a field just to see the review flow. Or join the waitlist.

And if you hold eight SAFEs total: keep the spreadsheet, seriously. This problem starts hurting somewhere around position 15. It just tends to get noticed at position 40, usually during diligence.