Zero look-ahead bias
Every row carries a snapshot-date stamp, so your backtest engine can verify: on signal day, this number had actually been announced. Look-ahead bias is blocked at the data layer, not by discipline.
Public beta · official-source direct · live backend data
Snapshot-grade Taiwan equity data
Every number is the version you could see that day.
A backtest’s first assumption is that the data was knowable at the time. We turned that assumption into a database: after every announcement day’s close, the database state is frozen into a snapshot — query any past date and you get that day’s as-of version, never the restated one. Statements are gated by statutory disclosure deadlines, monthly revenue is aligned to actual announcement dates, all direct from TWSE / TPEx / MOPS.
Public beta · demo keys work without signup · production keys by application (below)
$ curl "https://8888fortune.com/v1/valuation/vintage?stock_id=2330&as_of=2026-07-01" -H "x-api-key: demo-quant"
↳ returns the database snapshot as of 2026-07-01 — not the later-restated version. The demo key demo-quant works as-is.
Trust marks
Three promises written into the spec
Not marketing adjectives — auditable product guarantees, each mapped to something you can verify.
Official-source direct
Every field is labeled with its official origin — TWSE / TPEx / TAIFEX / MOPS — and snapshot time, fully auditable; every response carries source attribution.
Honest labeling
Every screener and composite metric is labeled “data tool, not a trading signal” — our own quant research validated what works, and what works is not what we sell.
Core features · all built on snapshots
One point-in-time snapshot series,
six modules on top
The database state frozen after every announcement day is the shared foundation of the six modules below — query, backtest or score, you always get the version that was knowable that day.
Snapshot query
Pick any past date and get the database state as of that day — statements gated by statutory disclosure deadlines, monthly revenue aligned to actual announcement dates, every restatement and retroactive revision excluded. Other sources only serve the latest revised values; we serve what the market could actually see.
as-of API · Quant tierBacktest data bundle
A market-wide panel with a snapshot_date stamp on every row, ready to feed into a backtest engine for strict point-in-time alignment — the data-layer fix for look-ahead bias. Parquet format with a full column dictionary.
Quant tier · bulk downloadStructured earnings calls
An LLM extracts guidance, capex plans and margin outlook into structured JSON — Chinese-language earnings calls made machine-readable, for event studies and fundamental tracking.
Chinese text · machine-readableFinancial scorecards
Piotroski F / Altman Z / Beneish M plus a composite health score, computed in-house from official quarterly statements with open formulas and per-line contributions — and snapshot-aware: query any past date for the score computed from statements visible that day.
Open formulas · auditableChip-structure dashboard
Institutional net-flow streaks, TDCC holder-concentration bands and futures OI — three chip-flow stacks on one page, refreshed daily from official sources, snapshot-stamped.
3 sources · daily snapshotsValuation percentile history
Where a stock’s current PER / PBR / dividend yield sits inside its own historical distribution — percentile and z-score against itself. No calls, no price targets.
Own history · not a signalHow snapshots work
Why snapshots can’t be rebuilt after the fact
The same reported number lives on two very different timelines — a conventional source vs. a snapshot database.
Conventional source — revisions overwrite
Aug 10 · announcedQ2 EPS announced at 3.2
Nov 12 · restatedRetroactive restatement — the same cell is overwritten to 3.8 3.2; the old version is gone
Backtest asks for Sep 1You can only get the latest 3.8 — but on Sep 1 the market saw 3.2look-ahead bias leaks into the backtest
Snapshot database — every version frozen
Aug 10 snapshotFrozen after the announcement-day close: EPS 3.2
Nov 12 snapshotThe restatement is saved as a new version 3.8 — the Aug 10 snapshot stays intact
Query Sep 1Returns the Sep 1 as-of snapshot: 3.2 — what was actually knowable that dayzero look-ahead bias · audit trail intact
A vintage series is the full version history of one metric across snapshot dates — every post-announcement revision is traceable. It can only exist if someone was saving it at the time; once data is overwritten, nobody can rebuild the as-of version.
For serious backtesting this is the line between honest and inflated results: backtesting on restated values silently uses information that wasn’t knowable on signal day. Every extra day of freezing adds a day of exclusive history — databases of this grade used to live behind enterprise annual contracts.
Vintage series accruing daily (+1 day every day) · Quant tierCompare
How the snapshot database differs from a conventional source
Conventional sources (resellers / charting apps)
Historical numbers are restated versions: revisions overwrite in place, silently leaking future data into backtests
Chip flows fragmented across separate subscriptions: institutional flows, holder bands, futures OI
Financial scores are a black box: one number, no per-line breakdown
Official data relayed second-hand, about a month behind
The snapshot database
Snapshot query: any past date returns the as-of version; restatements can’t reach frozen snapshots
Chip-structure dashboard: institutional + holder bands + futures on one page, 2 years deep
Financial scorecards with open formulas, per-line contributions, snapshot-aware backtesting
Direct from TWSE / TPEx / MOPS, frozen at T+1 after the close
Pricing
Three tiers — the deeper, the stronger
Free serves the latest data point, for trial; Indie unlocks 90-day history + the multi-factor screener + financial scorecards; Quant unlocks complete history + snapshot query + the backtest data bundle. No over-engineering — at most two paid tiers.
Free
Trials / students / teaching
NT$0free forever
Latest data point · whole market · 500 calls/day
- Valuation PER/PBR/yield, latest data point (whole market)
- Monthly revenue YoY/MoM, latest reported month
- Three major investors’ net flows, latest session
- Screener trial (via the Indie demo key)
- Historical time series (latest only)
- Full screener / financial scorecards
Indie
Solo quants / bloggers
NT$500/ mo
+ 90-day history · full screener · 10K calls/day
- Everything in Free, plus
- +90-day history time series (valuation / revenue / flows)
- Full screener: valuation × fundamentals × chip flow, freely combined
- F-score / Altman Z financial scorecards
- One-page company overview (5 sources stitched)
- Bulk monthly parquet download
Quant
Small prop shops / serious backtesters
NT$2,000/ mo
Complete history · snapshot query · backtest bundle · 100K calls/day
- Everything in Indie, plus
- Complete history panel (no depth limit)
- Snapshot query: the database state as of any past date
- Backtest data bundle: daily bulk download, snapshot_date on every row
- All derived factors + valuation percentile history + Beneish M
- Outage email alerts · priority support
Each tier’s unlock scope is machine-readable — call /v1/me (with your x-api-key) to see exactly what quota / history depth / factors your key unlocks.
Production keys (public beta): email [email protected] with your use case — free during beta. Self-serve signup is in the works.
Coverage · read live from the backend
Fundamentals to chip flows — one database
Valuation, monthly revenue, financial statements, institutional flows, holder dispersion, futures open interest and earnings-call summaries — market-wide coverage, direct from Taiwan's official agencies (TWSE / TPEx / MOPS), refreshed after each close with snapshots accruing daily. Each card below is one ready-to-use dataset; the numbers are read live from frozen parquet — not marketing copy. externally servable = open-licensed official source / our derived work; internal-first = licensing being confirmed item by item.
Loading…
Reading /v1/coverage live…
Reading /v1/coverage…
Live demo · hits the real backend
Not a static screenshot — real data
Pick a tier key, type a ticker, hit run, and it actually calls the backend API and returns a row from a frozen snapshot — real data, not a mockup. Watch the paid gate turn free into a 402.
Hit “Run” to fire the first call…
FAQ
About the database
What is a point-in-time (as-of) snapshot?
After every announcement day’s close the database state is frozen once — query any past date and you get that day’s snapshot, not the later-restated version. Conventional sources overwrite history in place; a snapshot (vintage) series keeps the full version history of every revision, traceable and auditable.
How is this different from Statementdog / Goodinfo / CMoney?
Their historical numbers are the latest revised versions, chip flows are fragmented across products, and scorecards are black boxes. We serve snapshot queries (the as-of version of any past date), transparent scorecards (open formulas, per-line contributions, snapshot-aware), and a chip-structure dashboard (institutional flows + TDCC holder bands + futures OI on one page) — all direct from TWSE / TPEx / MOPS, frozen at T+1.
How far back does the snapshot query go?
Honestly: the vintage series accrues daily from the day we started freezing — valuation snapshots started recently and are still shallow; institutional flows and statement data carry multi-year official history (aligned to statutory disclosure deadlines). Per-source snapshot depth is read live from /v1/coverage — no inflated claims.
Why not just scrape the data myself?
Scraped data is the latest revised version only, so a backtest silently uses information that wasn’t knowable on signal day — that is look-ahead bias, and it inflates results. The backtest data bundle stamps snapshot_date on every row: monthly revenue aligned to actual announcement dates, statements gated by statutory deadlines, ready to feed straight into a backtest engine.
Are the scorecard formulas really transparent?
Four scores: Piotroski F (nine financial-strength checks), Altman Z (bankruptcy risk), Beneish M (earnings-manipulation red flag) and a composite health score — all published academic formulas, computed in-house from official MOPS quarterly statements, each with per-line contributions (which line item added +1 or −1). Snapshot-aware: query any past date for the score computed from statements visible that day.
Is the Free tier really free forever? Do I need a credit card?
Yes — Free is free forever, no credit card. Register an email to get an API key with the market-wide latest data point, 500 calls/day and one demo screener. Paid tiers (Indie NT$500/mo, Quant NT$2,000/mo) unlock historical time series, the multi-factor screener, the complete scorecards, snapshot query and the backtest data bundle.
Is the data licensed? Can I use it commercially?
Yes. Externally we serve only sources listed on Taiwan’s open-government data portals, plus our own derived works (scores / factors / structured summaries, over which the operator holds editorial copyright). We never resell non-official raw values, and every response carries source attribution — see Data sources & licensing.
How fresh is the data?
Frozen at T+1 after each close, following the official publication schedule; snapshots accrue daily. The Quant tier adds outage email alerts and priority support.