Thesis alerts
Not a filing feed. Every row below is a state that moved — it names where it was, where it is, and the document that moved it — and it is here only because it passed a numeric test that a reader can read and disagree with. “A filing arrived” is on /filings; the overnight edition is the brief.
Thesis changes
in the last 30 days, dated by the evidence — not by when we read it
Rules firing
the rest fired nothing and are listed with their test
Rules rejected
designed, measured, and not shipped — each with the number that killed it
SPACs touched
of the live universe
Estimated NAV is now above the market price · 2
The stock has crossed below the cash behind it. Where a redemption right still exists, that gap is claimable; the crossing is the moment the arithmetic of holding changes sign.The last close against the estimated NAV (trust accreted to that session)
- 2026-09-14 close
- $12.35 vs ~$12.10 (−2.04%)
- 2026-09-21 close
- $12.00 vs ~$12.11 (+0.93%)
The price is now 0.93% below the estimated cash behind it, having been at or above it 5 sessions ago. The estimate is the last filed trust accreted at the 3-month T-bill — our arithmetic, not a figure any filing states.
buyredeemtwo closing prices (14 Sept 2026 → 21 Sept 2026) against the trust stamp of 30 Jun 2026, accreted — our arithmetic, no filing states itThe last close against the estimated NAV (trust accreted to that session)
- 2026-09-14 close
- $11.10 vs ~$10.97 (−1.15%)
- 2026-09-21 close
- $10.89 vs ~$10.98 (+0.80%)
The price is now 0.80% below the estimated cash behind it, having been at or above it 5 sessions ago. The estimate is the last filed trust accreted at the 3-month T-bill — our arithmetic, not a figure any filing states.
buyredeemtwo closing prices (14 Sept 2026 → 21 Sept 2026) against the trust stamp of 30 Jun 2026, accreted — our arithmetic, no filing states it
Ran, found nothing · 8
These rules were evaluated over the same window and cleared no row. A rule that fires nothing is not a rule that is missing.A >5% holder materially increased its stake
The stated combination deadline moved
The minimum-cash closing condition moved
The deal's outside date moved
The PIPE changed size
A redemption tally was filed
The trust account was drawn down
A shareholder vote is now on the calendar
What this page could not judge
The known blind spots, counted. A channel that reports only what it found teaches its reader that what it found is everything there is.- 0 future votes name a scheduling filing we do not hold, so they cannot be dated and no alert is invented for them.
- 44 live SPACs carry no trust snapshot at all, so no NAV crossing can be measured on them.
- 13 live SPACs have fewer than six sessions of prices, which is the minimum a crossing needs.
- 0 deadline changes were read off a proxy and refused: a proxy proposes an extension, a 10-Q reports the one that stands. Each of those SPACs appears above under the vote instead.
- 0 trust figures rose by more than a quarter between two filings and were refused as extraction artefacts. A trust cannot grow by a quarter of itself.
Rules that were designed and rejected · 7
Every one of these is true and detectable. None of them earned an interruption, and the measurement that decided it is printed rather than summarised.a new 13G — an arb fund entered
226 first-appearance 13G/13D rows at ≥5% on live SPACs in 30 days. Volume is the smaller problem: with no prior filing there is no "from" state, so the alert can only say "not previously on file" — and the whole InstHolder table was harvested in two sweeps (2,316 rows on 2026-08-15, 188 on 08-14, filings dating back to 2018). An alert cannot distinguish "the fund bought this month" from "we read the filing this month", and "an arb fund entered" asserts the first. `ARB_POSITION_GREW` keeps the half that has two states. Revisit when 13Gs are harvested on a clock, so a first appearance has a date we own.
we now hold a $90m minimum-cash condition and held nothing before
9 of the 15 minimum-cash rows are `unextracted → value` (and 8 of 14 PIPE rows). That arrow is a claim about the earlier FILING that we are not entitled to make: the clause may be there in wording our regex does not match. It is news about our coverage and belongs on the dossier's diff strip, which already carries it in the four-state form.
substantial-doubt language appeared where the previous filing had none
Built, measured, and withdrawn before it shipped. 187 appearances against 12 resolutions all-time; 34 in the last 30 days, of which **33 landed inside two weeks** of the Q2 10-Q season. A channel that delivers 33 rows in a fortnight and nothing for the five months either side teaches its reader to mute it, twice a year. The seasonality alone might be survivable; what is not is that the alert had to concede its own irrelevance in its consequence sentence — the doubt is about the OPERATING SHELL, and the trust is held separately, so a holder whose downside is the trust does not buy, wait, sell or redeem differently. (The obvious defence, that it is just ASC 205-40 firing once the mandatory liquidation date is inside 12 months, is only 14 of the 34 — so it is not merely mechanical. It is simply not about the trust.) It stays on the dossier diff strip, which is where a standing condition belongs.
read the deadline change off a DEF 14A as well as a 10-Q
ROSS, in the first run: the alert read "previous DEF 14A 16 Sept 2026 → this DEF 14A 16 Mar 2028" and concluded **547 more days before the trust must be returned**. The 2028 date is what the proxy PROPOSES, subject to the vote it is calling for 16 Sept 2026. That is an unproven relief of urgency in a channel whose whole claim is that it never implies urgency it cannot prove, and it is the same distinction `src/lib/deadline-authority.ts` already refuses on (`deadline-unapproved`). A periodic report STATES the deadline as it stands; a proxy ASKS. Restricting the rule to 10-Q/10-K costs 7 of 37 rows in 30 days and loses no information: every one of those proxies also produces a `VOTE_SCHEDULED` alert, which is the true and dated version of the same news.
the sponsor lent the SPAC more money
20 changes in 30 days, 12 of them increases — so the volume is fine. It fails the ONLY test that matters here: name the decision it moves. A holder whose downside is the trust does not buy, wait, sell or redeem differently because the sponsor funded another extension payment out of a promissory note; the loan sits outside the trust and behind it. It is real signal about sponsor commitment and it belongs on the sponsor dossier.
the SPAC changed the sector it says it is hunting
Already measured by `filing-diff.ts`: 10 of 124 `sectorLanguage` signals changed, and most of those cosmetically — LATA gained an article and an abbreviation, MUZE had a grammar fix, DRDB is a tag-strip artefact. `normalizeSnippet` cannot tell those from CUB pivoting to Venezuelan upstream oil and gas. A prose-change signal wants its own treatment, not a seat in a numeric channel, and shipping it at this precision would spend the reader's trust on three real pivots buried in seven typographical ones.
the redeemable share count dropped — its own alert
It is the SAME EVENT as the trust drawdown seen from the other side. Over 365 days, 21 of the 45 material share-count falls land on a filing that ALSO carries a material trust fall. Two rules would interrupt a reader twice for one redemption. It is now a second line on `TRUST_DRAWN_DOWN`, and the 24 share-falls with no trust fall of their own still fire through it — one event, one interruption, both numbers.