Linda Ledger turns the conditions that would prove you wrong into checks that run every evening against SEC filings, prices and the macro backdrop. When one trips, you get the evidence before you think to look. You also hear what is going for it, quoted word for word from the source and never dressed up as a verdict. Every step goes into a record nobody can edit.
“A member stops being able to fund itself” — runway under 12 months
Holds“Dilution overwhelms holders” — shares up more than 15% a year
WatchingYou already wrote down why you own it. Linda Ledger keeps checking whether that is still true.
The investment memo or IC note you already have. We turn it into a draft thesis: what would prove it wrong, what would count as evidence for it, and the partner companies worth reading. Each condition links to the sentence it came from; standard suggestions are labelled as ours.
Each kill criterion, supporting condition and milestone shows the sentence it came from. You edit, then approve.
Against new SEC filings, company reports, prices and the macro backdrop. Missing data says “unknown”, never a guess.
When a kill criterion trips or clears, a milestone is missed or met, or a supporting condition is met, with the number, its source and its date. Every Friday, one digest: the week's evidence for and against each thesis, quoted from its source.
Hold, trim, exit or retire the criterion, with a reason. Everything is chained and timestamped.
On 13 October 2026 the check on a robotaxi thesis tripped: one of its four companies had under two months of cash at its recent burn. No one had to remember to look.
The same day, the check also showed its working. Two other companies' runways looked too short because short-term investments weren't counted. We corrected the measurement in public, next to the alert, and the record kept both.
| Company | 13 Oct | After the fix |
|---|---|---|
| Lucid (LCID) | 2 months | 1.6 months |
| Rivian (RIVN) | 12 months | 16.6 months |
| Pony AI (PONY) | 19 months | 92 months |
| WeRide (WRD) | 40 months | 40 months |
Criterion: runway under 12 months. It stayed tripped after the fix, because of Lucid alone.
Every night, new releases from your companies and the partners you name are read against the statements your thesis says must be true. Each one is labelled as supporting, cutting against, or mixed, and comes with what it does not show.
A quote reaches you only if it appears in the source word for word. That matters: a summary we saw of the release shown here listed “Germany and France” as expansion markets. The release names only Germany.
Evidence for a thesis is not proof of it. Your kill criteria still decide when something you named has broken, and the weekly digest shows both side by side.
“Germany and France are identified as future expansion markets” — not in the release, so not in the digest
DroppedDeal filings for WeRide, Pony AI, Rivian — merger proxy, tender offer, form 425, 13D
None this weekEach check has a precise definition, a public data source and a test, including the case that usually fools a quick script. Anything that can't be measured from public data stays with you, as a review item.
Survival. Known trap covered: cash held in short-term investments counts (Rivian, Oct 2026); a newer 6-K beats older annual XBRL.
Survival. Known trap covered: investments are cash; operating leases are not debt here.
Capital. Known trap covered: a stock split is not dilution (changes beyond -30%/+140% are ignored).
Capital. Known trap covered: a 424B5 can be a debt deal: the alert says 'filed', not 'diluted'.
Operations. Known trap covered: foreign filers with only annual figures use the annual basis, labelled.
Operations. Known trap covered: negative gross margins (early manufacturers) are valid values, not errors.
Operations. Known trap covered: a missing report breaks the run rather than counting as a fall.
People. Known trap covered: appointments also file 5.02: the alert says 'change'.
People. Known trap covered: 13D/A amendments count; 13G (passive) does not.
Governance. Known trap covered: a planned rotation also files 4.01.
Governance. Known trap covered: a voluntary transfer of listing also files 3.01.
Market. Known trap covered: adjusted vs unadjusted prices after a split.
Market. Known trap covered: dividends are ignored (price return), stated plainly.
Governance.
Survival.
Macro. Oil in a shock, credit under stress, a slowing consumer, a stagflation regime: read each morning from market prices, Treasury cash, ship transits and bank and retailer reports. Known trap covered: a stale or missing reading is unknown, never all clear.
Macro. Brent, the 10-year and 2-year yields, the real yield, the high-yield spread, mortgage rates, Fed and recession odds, Hormuz traffic. Known trap covered: a series that stops updating is unknown, not the last value repeated.
Supporting. A merger proxy, a tender offer or the board’s response, a form 425, a 13D, or an 8-K material agreement or completed acquisition. Every alert says what the filing does not show: a 425 can be filed by the buyer, and a tender offer can fail.
Supporting. Months of cash, revenue growth or gross margin reaching your level, or a KPI rising report after report. Known trap covered: more runway often comes from a raise, so the alert says to check dilution.
Across your book. When several theses lean on the same macro condition, such as falling rates or calm credit, you hear it once, with what the macro backdrop reads today. They can all be wrong together.
Your judgement. It becomes a review item with a date, so it is never forgotten.
Every thesis version, check, alert, correction and decision is an entry linked to the one before by a SHA-256 hash. Change any entry and every hash after it stops matching.
That gives you something to show investors, an auditor or a compliance officer: what you believed, when you were warned, and what you did. Export it as a file at any time.
{"seq": 4, "kind": "alert.raised",
"data": {"thesis": "autonomy-consolidation",
"item": "c1", "from": null, "to": "tripped",
"detail": "WRD runway 40 months; PONY runway 19 m…",
"on": "2026-10-13"},
"prev": "47b9…a6c1",
"hash": "d599…9853"}In the best-known experiment on judgment, Kahneman and Tversky described a woman named Linda and asked people which was more likely: a plain fact about her, or that fact plus a detail that fit her story. Most chose the story, even though it can never be more likely than the fact alone.
Investment theses fail the same way. A story that once fit keeps feeling true after the facts have moved. Linda Ledger checks the story against the filings every evening, and keeps the record.
Source: Tversky & Kahneman (1983), “Extensional versus intuitive reasoning: the conjunction fallacy in probability judgment”. Background.
Prices are for the first cohort and may change before general launch.
We're onboarding a small group of analysts and firms first. Tell us where to reach you.