this is my AI's analysis:

Here is how I would classify them for the kind of testing you and I care about—e.g. “on each historical rebalance date, rank the then-current Nasdaq-100 or S&P 500 members using EPS growth, sales growth, ROE, valuation, etc., without feeding the system our own historical fundamentals.”

Service Historical fundamental data supplied by them? Can their backtester use it? Would you need to supply your own historical fundamental database?

1. Portfolio123 — this is the strongest fit
Portfolio123 explicitly maintains point-in-time historical fundamental data and is designed to avoid survivorship bias. Its own methodology explains that past simulations see companies that actually existed at that historical date—including companies that subsequently disappeared—and states that its backtests are not affected by survivorship bias. Portfolio123
So, for example, we could tell Portfolio123 to test something like:
On each month-end, take the historical S&P 500 membership, require positive 6-month momentum, rank by trailing EPS growth, ROIC and free-cash-flow growth, hold the top 5 for one month.

You do not have to supply 1999–2026 EPS, sales, ROE, etc. yourself. That is exactly the advantage.
For our purposes, Portfolio123 is the one I would investigate first.

2. TrendSpider — much more interesting than I previously thought
TrendSpider now says its Strategy Tester can use 50+ years of historical data and explicitly advertises the ability to layer “fundamentals, events, news and more” into strategies. trendspider.com
It also has its own fundamental database inside the platform; Sidekick can access fundamentals, SEC filings, earnings transcripts, analyst data, etc. trendspider.com
And the Strategy Tester can explicitly backtest certain historical events including earnings and analyst estimates. TrendSpider Support
So you apparently would not have to upload your own historical price/fundamental database for the data TrendSpider supports.
However, there is one thing I would verify before paying for it: whether a condition such as
EPS growth > 25% as known on that historical date

is truly evaluated using point-in-time fundamentals, rather than today's/latest fundamental value painted over historical price bars.
That distinction is crucial. Their marketing strongly implies historical fundamental backtesting, but I haven't yet found documentation as explicit about point-in-time accounting data and survivorship bias as Portfolio123's.
So I rate TrendSpider:
Promising enough to test, but verify point-in-time handling before trusting a 20-year fundamental backtest.

3. V7 Go — the wording you found is real, but there is a catch
V7 really does say:
the agent can run your screen against historical filings and data to determine which companies would have matched in the past. V7 Labs

So your quotation is accurate.
But farther down that same V7 page, it advertises importing data/files from:
- S&P Capital IQ
- FactSet
- Bloomberg Terminal
- 10-K filings
- earnings transcripts
- investor presentations
- market data feeds
and explicitly says “Import your files.” V7 Labs
That makes me much less certain that a normal V7 subscription includes a ready-to-use historical fundamental database analogous to Portfolio123.
My interpretation is:
V7 provides the AI machinery for doing historical document/fundamental screening, but may expect the customer to connect or supply the underlying data source.
For a hedge fund already licensing FactSet or Capital IQ, that's excellent.
For you, if V7 said, “Connect your Capital IQ account,” it doesn't solve our problem.
Before considering V7, I would ask them this exact question:
“Does a standard V7 Go subscription itself include point-in-time historical U.S. stock fundamental data and historical index membership, or must I provide/connect a data source such as FactSet, Bloomberg, Capital IQ, or my own files?”

That answer determines whether V7 is useful to us.

4. YCharts — it has the data, but this is the frustrating one
YCharts actually has a very substantial database:
- up to 30 years of fundamental data
- 20,000+ companies
- thousands of calculated fundamental metrics
- historical calculations extending as far back as the underlying fundamental data permits. YCharts
And Y Agent can build screens from natural language such as:
S&P 500, P/E under 20, revenue growth over 10%. YCharts

So the historical data are already inside YCharts. You would not need to upload them.
But YCharts' own 2026 documentation says something very important:
They cannot screen for historical data; the screener uses the most recent “as of” data. YCharts

That prevents us from doing what we actually want:
“Go back to March 2004, screen the stocks using the fundamentals available in March 2004, select five, advance a month, repeat.”

YCharts can graph historical fundamentals and analyze historical series, but its ordinary custom Screener/Y Agent workflow is not a point-in-time historical stock-screen backtester.
So I would cross YCharts off for this particular project, despite its excellent database.

5. Horizon.Trade — has the backtester; fundamentals are the unresolved part
Horizon clearly supplies its own historical market data. You describe the strategy in English and it runs the historical backtest. It advertises up to 20 years of historical data. Horizon
It even warns about survivorship bias, which is encouraging: its documentation specifically notes that testing only today's S&P 500 over the past ten years would be incorrect. Horizon
But I have not found authoritative documentation showing that Horizon supplies a point-in-time historical fundamental database suitable for rules such as:
- EPS growth
- sales growth
- ROE/ROIC
- earnings surprises
- P/E
- debt/equity
- margins
as they existed on each historical selection date.
So I would not pay $1,000/year on the assumption that it does.
The question to Horizon would be:
“Can I backtest a monthly cross-sectional stock-selection strategy using point-in-time historical fundamentals such as EPS growth, revenue growth, ROE, P/E and earnings surprise, with values restricted to information that was publicly available on each historical rebalance date?”

If the answer is no, Horizon doesn't add much to what we've already done ourselves with price/volume data.

6. Deepvue
Your characterization is essentially right. It has an impressive current fundamental screener—over 1,100 fields, including hundreds of earnings and sales fields. Deepvue
But I find no evidence that it can replay a screen historically and produce a portfolio backtest.
So:
Excellent screening data; not our historical fundamental backtesting solution.
7. Trade Ideas / Holly
This isn't what we need.
Trade Ideas' OddsMaker backtester currently uses only about 40–64 trading days of history and is primarily an intraday/event/technical backtester based on one-minute OHLC data. Trade Ideas
Holly running “millions of backtests every night” sounds impressive, but it is solving a very different problem from:
Nasdaq-100 historical membership + quarterly fundamentals + monthly ranking back to 1999.

So I would eliminate Trade Ideas for our project.
The two I would investigate further
For our specific missing-data problem, the shortlist is now:

Portfolio123 — clearly yes.

TrendSpider — potentially yes, and worth testing.
Portfolio123 interests me more because it explicitly documents point-in-time/survivorship-bias-safe data. That's exactly what prevented us from testing fundamental variations on your Nasdaq-100 strategies ourselves.

TrendSpider interests me because it may now let us combine the kind of technical rules we've already developed with historical fundamental factors in a much easier natural-language/no-code environment. But I would first make it prove that its historical fundamental fields are genuinely point-in-time.

And V7 Go is potentially fascinating, but I suspect its historical-analysis capability may rely on the customer bringing a FactSet/Capital IQ/Bloomberg/data feed rather than V7 giving you that database itself. Its own page advertises those imports, despite the “historical filings and data” backtesting claim. V7 Labs
If you want, the next useful thing I can do is investigate Portfolio123 versus TrendSpider in detail for one of our actual Nasdaq-100 strategies—including exactly which historical fundamental factors each could add, whether each can handle historical Nasdaq-100 membership, monthly rebalancing, SPY325 timing, and approximate cost. That would tell us whether either one is worth paying for.