AI-agent launches that fell out of monitored result windows
Through the 2026-08-09 seal, 888 source-keyed signals were absent from the newest sealed result set where they had appeared before. This is source-window evidence, not a judgment about any venture.
The useful fact is not that a launch disappeared, but that a dated source result stopped carrying it. A capped ranked query changes as newer or higher-ranked items enter it. We preserve the dates when a signal was present and the first later seal where that same monitored source key no longer carried it.
If you track AI-agent startups from public sources, you are reading a moving surface. Hacker News search windows are ranked and capped. Product Hunt and dev.to feeds roll forward. Each answers the question you ask at that moment with the fields and ranking available then. Ask again later and the answer changes, without the source supplying the earlier result set beside it.
That gap is what this ledger measures. Since 2026-07-01, the clock has written 41 sealed collection runs. When a signal that appeared in one dated result set is absent from the newest result set for the same source key, the archive can state both observations. It cannot turn that membership change into a claim about the underlying project.
What “fell out” means
“Fell out” has exactly one meaning here: a source-keyed signal is absent from the newest sealed result set for the same monitored source key where it was previously observed. It does not mean the item was deleted from the internet. It does not mean a company shut down, failed, became inaccessible, or changed strategy. The page names no founder and makes no assessment of a named business.
6 monitored source keys are Hacker News Algolia searches, each capped at 50 results. A newer or higher-ranked item can push an older one outside that window. Ranking, recency, and query behavior move independently of the venture being discussed. A dropout is therefore a prompt for research: a signal is no longer where the analyst had previously observed it, and the archive preserves when that changed.
How long signals have been outside the newest window
The absence age is measured from a signal's last sealed appearance to the newest seal—not to the current clock on this page. The split keeps one headline count from hiding the difference between ordinary short-window churn and a signal that last appeared many sealed days earlier.
Source coverage
The monitored set contains 6 Hacker News Algolia result windows, 2 Product Hunt feeds, and 2 dev.to feeds. In the newest seal, 10 of 10 source requests returned HTTP 200. The reader refuses to calculate absence at all if any newest source request lacks a successful HTTP result, so a collection error cannot become a confident dropout.
| Monitored source key | Distinct observed | In newest set | Absent from newest set |
|---|---|---|---|
| dev.to: AI tag | 107 | 2 | 105 |
| dev.to: AI agents tag | 458 | 12 | 446 |
| Hacker News: AI agent + MRR | 76 | 36 | 40 |
| Hacker News: agent + paying customers | 30 | 30 | 0 |
| Hacker News: AI agent + business | 59 | 50 | 9 |
| Hacker News: autonomous agent | 57 | 50 | 7 |
| Hacker News: built with AI agents | 51 | 50 | 1 |
| Show HN: AI agent | 166 | 47 | 119 |
| Product Hunt: artificial intelligence | 73 | 6 | 67 |
| Product Hunt: main feed | 103 | 9 | 94 |
Coverage boundary. This corpus covers publicly narrated AI-agent venture signals that entered these monitored query windows and feeds. It is not a census of companies or launches. Anything never returned by a monitored source is outside the archive, and its absence here says nothing.
What the archive can show
Counted from the read-only store, the 2026-08-09 view contains 10,710 dated signal observations across 41 runs. It can show prior result membership, the last sealed appearance for a source key, and dated points values when Hacker News returned them. The stored series include 54 source-keyed signals with at least two different points values.
Hacker News Algolia serves current item fields and current query results. The archive adds our dated observations of earlier values and earlier query membership. It is not an independent timestamping authority and not irrefutable history. It is a sealed record of requests this collector made, read through a projection that checks row hashes and run commitments before returning an aggregate.
Weekly AI Agent Venture Change Ledger
$249/month — proposed
For a sourcing or scouting analyst at an early-stage AI venture fund, or a competitive-intelligence lead at an AI developer-tools company: a private Monday ledger covering the prior seven days, delivered by noon Arizona time as matching PDF and CSV plus a SHA-256 manifest.
- First-seen public launch signals entering a sealed source result set.
- Dated Hacker News score movers where stored points values changed.
- Source-result-window dropouts, their last observed date, and absence age.
- Row-level source links so an analyst can inspect the current public surface.
- The same rows in the PDF and CSV, with a manifest over both delivered files.
Cancel before the next Monday delivery. This page is a demand test. No payment is accepted, no checkout is opened, nothing is scheduled automatically, and no particular company, source outcome, or weekly row count is promised.
What it does not prove
- Not population completeness: the archive holds only what the named monitored sources returned.
- Not venture success, funding, revenue, team quality, or a ranking of any named company.
- Not a compliance, diligence, investment, or legal determination.
- Not evidence that anything was removed from the internet or that a project is defunct, failed, or inaccessible.
- No post text, author names, handles, private analyst rankings, or model assessments cross the public reader.
A dropout is a fact about one dated query result. Every inference past that point belongs to the analyst, which is why the proposed private artifact carries the source link, source key, and observation dates rather than a verdict.
Method and seal
Each sealed run issues the fixed source set, stores the returned result membership with a snapshot date, and records every source request outcome. Records are keyed by source key and signal identifier, so membership is compared only inside the same monitored window. A dropout is a set difference between the newest successful sealed result and prior appearances for that key.
The public reader opens the database in read-only mode. Before producing a count, it reproduces the stored signal hashes, fetch-row hashes, per-run row counts, and batch commitment. If it cannot verify the store or newest source outcomes, the page returns unavailable rather than printing zero or an unsupported count.
Newest batch SHA-256: 45c35a90468187bf721ce86fbc4720833885ff1af841c1a0d3ed917d0aaa9ce1
Request a sample week
A useful sample should match the work product: the same PDF, CSV, manifest, dated source links, and explicit gaps. Tell us which source windows or research decisions matter to your team. The request opens no checkout. It tests whether dated membership changes and score trajectories improve a real sourcing workflow before a recurring product is activated.