Hold every AI run to your analysts’ standard.

Your analysts write the bar for an earnings review or an IC memo. Stellifai tests every agent run against it, keeps the history of every rule, and finds the cheapest model that still clears it.

Book an eval sprint
  1. Rule 1Revenue ties to the 10-Q within 0.5%
  2. Rule 2Guidance change in first paragraph
  3. Rule 3Every figure cites its source filing
  4. Rule 4Peers on the same fiscal calendar

Example data

Q3 earnings review · rubric v14

Rules:

1. Revenue ties to the 10-Q within 0.5%

2. Guidance change flagged in the first paragraph

3. Every figure cites its source filing

4. Peers on the same fiscal calendar

Agent log:

→ reading 10-Q

→ drafting earnings summary

→ checking rules 1–4

Models:

Model A$1.48/run

Model B$0.62/run

Model C$1.96/run

Rule 185% of example runs clear rule 1
Rule 278% of example runs clear rule 2
Rule 392% of example runs clear rule 3
Rule 488% of example runs clear rule 4

Example: run 0412 clears the bar on Model B at $0.62 per run.

The right rules. The right model.

A navigator needs a fix on every star before trusting a position. You need three things before trusting an agent’s work.

  1. Write the bar

    An analyst writes what a good earnings review looks like, rule by rule, in plain language. No engineer in the loop.

    rule 2 · written by the coverage lead
Flag any change to guidance in the
first paragraph, with the prior figure.

  2. Test every run

    Each agent run is checked against every rule before it reaches a reviewer. You see which rule it missed and why.

    run 0413   4 of 4 rules   clears
run 0414   4 of 4 rules   clears
run 0415   3 of 4 rules   held: rule 4

  3. Pick the model

    When a new model ships, re-run the suite. Each workflow moves to the cheapest model that still clears the bar.

    Model A   $1.48/run   clears
Model B   $0.62/run   clears   selected
Model C   $1.96/run   clears

Example data

Every rule has a history.

Standards change. A PM tightens a tolerance, a new analyst adds a check. Every rule is versioned: who changed it, when, why, and whether last quarter’s runs would still clear it.

One team’s proven rubric becomes a starting point the next team can fork, adapt and re-test, so good standards spread without drifting.

TRIP LOGQ3 earnings review

  1. v14Oct 2coverage lead

    Rule 1 tolerance tightened

    - ties to the 10-Q within 1.0%
+ ties to the 10-Q within 0.5%

    212 past runs re-tested · 9 no longer clear

  2. v13Sep 18analyst, consumer team

    Rule 4 added

    + peers on the same fiscal calendar

    212 past runs re-tested · 14 held on rule 4

  3. v12Aug 30forked from the TMT team's rubric

    Adapted for consumer coverage

    ~ 3 rules kept, 1 rewritten

    baseline set

Example data

Built for investment research.

Workflows Stellifai is built for
DestinationWhat it has to get rightRe-run
Earnings and coverageReviews that have to run the same way every quarter.every quarter
DiligenceWhere the numbers have to tie across every team's inputs.every deal
Portfolio monitoringOne workflow, re-run against every asset you hold.every asset

Piloted with three institutional research and investing teams.

Start with one workflow.

In two weeks your team has a written bar, a test suite, and a cost per run for every model that clears it.

Duration
2 weeks
Scope
1 workflow
You get
written bar · test suite · cost per run by model
Book an eval sprint