Introducing Second, our SA precheck app
A second pair of eyes before SA review
Second is an agentic review workflow for equity research. Analysts run it on their own drafts before review, whether that is a FINRA Series 16 supervisory analyst or your 審査担当者 under JSDA rules, and research managers can run it across the back catalogue to get ahead of audit risk.
サンプル精機（0000）
バリュエーション：PERで評価
目標株価4,300円は、29/3期予想EPS 206円にPER 21倍を適用して算出。
29/3期の営業利益は2兆円を見込む。当社では、新製品の寄与により中期的な成長が続くと考える。
- Fix Number check FY3/29 operating profit in the text (2兆円) exceeds the earnings model (1兆8,100億円).
- Fix Arithmetic EPS ¥206 × P/E 21x = ¥4,326, not the ¥4,300 target price stated.
- Minor House style Ambiguous 当社: the issuer or the firm? Use 弊社 for the firm’s own view.
Two ways to use Second
-
For analysts
Before SA review
Analysts run Second on their draft and earnings model before they submit. Every figure is reconciled against the model and the company’s TDnet and EDINET filings, house style and SA rules are applied, and findings appear in place on the draft in about a minute. The SA gets a cleaner draft; the analyst gets fewer review comments.
How it fits the workflow -
For research managers
Risk control across your back catalogue
Run Second over the research your team has already published. See where it falls short of JSDA or FINRA rules, by team, sector or analyst, and fix the pattern before an inspector finds it.
Risk control for research managers
Why we built Second
We started with translation. Our AI translation editor helps expert translators publish research in English more than 50% faster.
Once translation kept pace, the queue moved to the next step: SA review. Every report still has to pass a supervisory analyst, and there are fewer of them each year.
Second is how we are tackling that bottleneck: an agentic review workflow built on the same idea as our translation editor. Your own data and rules, applied the same way every time, with people making the final call.
About the AI translation editorWhere SA review time goes
What analysts and supervisory analysts tell us about checking reports today.
-
Most of the time goes on numbers
Supervisory analysts tell us 60–70% of review time is numerical checking: P&L, balance sheet and cash flow agreeing, quarters summing to the full year, actuals against estimates.
-
Three sources drift apart
The text, the earnings model and the report’s tables don’t always agree. Reused files carry stale target prices and forecasts.
-
Feedback varies by reviewer
Wording comments (予想 or 弊社予想, 今期 or an explicit fiscal year, tense, tone) differ from one reviewer to the next. Analysts want rules they can predict.
And there are fewer supervisory analysts each year, so the load falls on fewer people.
What it checks
-
Numbers
Every figure in the text and tables reconciled against the earnings model, with the arithmetic re-derived.
-
House style
Fiscal-year labels, 前年比, numerals, units, attribution and tense, from your style guide and terminology.
-
Restricted terms
Banned and promotional vocabulary, wording that could suggest MNPI, and first-person usage.
-
SA rules
Rating and target-price consistency, valuation rationale and sourcing, from your SA manual. Set it to whichever standard applies to your research: JSDA, FINRA or your own.
Cross-checks against company filings
Figures in the draft are also checked against the company’s own disclosures on TDnet and EDINET, such as earnings releases (決算短信) and securities reports (有価証券報告書), so the actuals and company guidance quoted in the report match the source.
How it fits your process
-
Analyst runs the check
On the draft (Word or BlueMatrix XML) and the Excel earnings model, before submitting.
-
Findings appear in place
Graded by severity, so minor items collapse and what matters is seen first.
-
Analyst responds
Accept, fix, or override with a reason. Every decision is recorded.
-
SA receives a cleaner draft
Along with the check results, so review starts from what is left.
Risk control for research managers
Every report your team has published is still on the record. Second lets research managers check the whole back catalogue against JSDA rules on analyst reports and FINRA Rule 2241, so you know where the risk sits instead of finding out in an inspection.
That matters more as inspectors adopt AI. A gap such as a missing risk disclosure can then be checked across every report at once, so one oversight repeated through a year of research becomes hundreds of findings. Second lets you run that sweep yourself, first.
The FSA, for one, is developing an “AI chief inspector”: an assistant trained on supervisory guidelines, past training materials and veteran inspectors’ know-how, with a prototype planned by July 2027. Asahi Shimbun, 30 September 2026
- See where the risk sits, by team, sector or analyst
- Know which rule each finding relates to
- Track whether the pattern improves over time
- Feed what you find back into the pre-submission check, so it doesn’t recur
Explainable, not a black box
-
Every finding traces back
Each finding links to the rule that raised it and the step that produced it, so analysts and SAs can see why.
-
Your rules, applied the same way
Your SA manual, style guide and banned-terms list are the rulebook. Every report is checked against the same standard.
Also translating your research into English? See the AI translation editor
See a check on a real report
We’ll walk you through a check on a report and earnings model like yours.