Move faster through the first-pass review
Instead of starting every case from a blank document, the tool gathers the key treaty facts, narrows the review lane, and returns a structured package for follow-up work.
For international tax teams
Trusted AI workflow for international tax pre-review
Start with a faster first-pass review before full legal or tax analysis. The tool guides key facts, narrows to the treaty lane in scope, and returns a structured handoff for the next reviewer.
It is designed to save time at the front of the workflow, not to replace a final tax opinion.
3 treaty pairs
Dividends, interest, royalties
Guided review first
Human-reviewed handoff
Guided facts
Structured review output
Workflow handoff
Terminate on missing critical facts instead of guessing
Unsupported or ambiguous cases are surfaced explicitly. The workflow is designed to help teams avoid accidental overreach in high-stakes tax scenarios.
Carry the review forward with traceable handoff detail
Results are packaged with treaty lane references, review signals, and workflow-ready notes so the next reviewer does not have to reconstruct the first pass from scratch.
How it works
Three steps, built for the way tax teams already work
The product is not an open-ended chatbot. It is a bounded pre-screening workflow that helps the first reviewer organize facts quickly and hand the case forward cleanly.
Collect the facts that actually matter
The guided path asks for the facts needed to place the payment in the right treaty branch, such as payment type, treaty pair, ownership status, and dividend-specific thresholds where relevant.
Narrow to the treaty lane in scope
The engine uses structured treaty data and conservative rule logic to determine whether the case stays in scope, needs more facts, or should be rejected as unsupported.
Return a review package, not just an answer
Outputs include the pre-review state, next actions, source-aware references, and handoff notes for the human reviewer who takes the case further.
What the team receives
A pre-screening output that is ready to discuss internally
- Guided fact summary So reviewers can see what was declared before they rely on the narrowed branch.
- Structured review state So the team can distinguish supported, incomplete, and out-of-scope scenarios quickly.
- Workflow handoff package So downstream reviewers do not need to re-interpret the first pass from scratch.
- Review signals So BO, MLI/PPT, short holding period, and conflict prompts remain visible without pretending to be final determinations.
90-second case walkthrough
One real CN-NL dividends case, shown from input to final boundary
This walkthrough does not try to explain the whole system. It first establishes the case quickly, then stops on the treaty branch and the source chain behind it, before showing the handoff artifact and ending on the boundary the product keeps.
CN -> NL dividends
Input
Choose the supported treaty pair and income type.
Guided facts
Click into the actual dividend fields that determine the review path.
Treaty branch
First pause on the Article 10 branch that narrows the case.
Provenance
Stay longest on the source chain behind that branch.
Handoff
Then confirm the workflow-ready package for the next reviewer.
Boundary
Close on the explicit boundary: pre-review only, not a final tax opinion.
How new treaties are onboarded
A human-reviewed compiler replaces manual treaty-to-rule coding from scratch
New treaty lanes are compiled offline from governed source documents. The compiler compares bilateral Articles 10 to 12 against the OECD reference, writes structured candidates, and only promotes a reviewed dataset into runtime after explicit approval.
Baseline-aware delta extraction
The authoring pipeline compares the bilateral treaty against the OECD Model 2017 reference for Articles 10, 11, and 12, then emits delta artifacts together with a full runtime candidate dataset.
Human review remains the gate
Compile output does not go live automatically. Review, approval, and promotion remain explicit workflow steps so tax-domain judgment stays human-controlled.
Single controlled CN-KR pilot has measured timing evidence
The first real onboarding pilot recorded a 26-second reviewer session and a 10m45s repo-internal source-build-to-promote elapsed time on governed official inputs. This is measured pilot evidence, not a guaranteed onboarding SLA.
Public evidence layer
Public proof is separated from product claims
The public story is backed by concrete evidence artifacts: a measured pilot summary, a treaty onboarding proof matrix, and a regression snapshot. The goal is to show what is already implemented without claiming more than the repo has actually proven.
Measured pilot
Single controlled CN-KR timing evidence
Reviewer elapsed
26 seconds
End-to-end elapsed
10m45s
Pilot type
Single controlled pilot
Measured on governed official source inputs. The result is evidence of a real workflow, not a general onboarding SLA.
Proof matrix
Compiler moat, shown across more than one lane
- CN-SG shadow rebuild
- CN-NL shadow rebuild
- CN-SG OECD delta proof
- CN-NL OECD delta proof
- CN-KR initial onboarding
The offline treaty compiler is not proven on one toy case only; it now holds across shadow rebuilds, baseline-aware delta proofs, and one real new-pair onboarding path.
Regression snapshot
Replayable checks sit behind the public story
199
backend passed
20
frontend passed
5
site tests passed
Build
frontend build passed
Public runtime, onboarding workflow, and product-site assertions all sit under replayable checks rather than screenshots alone.
3-minute verifier path
How to verify this repo quickly
If you only have a few minutes, do not start with the full local setup. Verify the moat in this order.
Watch the 90-second walkthrough
See one real CN-NL dividends case go from input to final boundary.
Check the measured pilot card
Confirm that a real CN-KR onboarding pilot has recorded timing evidence.
Scan the proof matrix
Verify that the compiler path holds across shadow rebuilds, delta proofs, and a real new-pair onboarding.
Only then open the local quick proof
Use the README smoke path if you want to validate the public runtime locally after the evidence layer makes sense.
Implemented vs not claimed
Show what is real, and say out loud what is still not claimed
This product is stronger when it is explicit about both its current capability and its current limits.
Implemented today
Source-anchored treaty pre-screening, guided fact collection, workflow-ready handoff, a human-reviewed onboarding compiler, OECD baseline-aware delta extraction, and a measured CN-KR pilot.
Not claimed
No final tax opinion, no automatic MLI/PPT override, no guaranteed onboarding SLA, and no claim that future treaty pairs will always complete inside the same measured window.
Current coverage
Deliberately narrow, because trust matters more than breadth
The product focuses on a bounded set of treaty pairs and income types so the review path remains auditable and conservative.
China – Netherlands
Dividends, interest, royalties
China – Singapore
Dividends, interest, royalties
China – Korea
Dividends, interest, royalties
Boundaries
This tool does not replace a final tax opinion
It helps teams move faster through the front of the workflow, but it does not automate the final legal judgment.
Not a final determination
The output is a first-pass pre-review, not a substitute for legal, tax, or internal approval.
MLI/PPT stays a review signal
The tool surfaces these items for human follow-up instead of pretending to resolve them automatically.
Unsupported scope is rejected
When a case falls outside the supported treaty pairs or lanes, the workflow says so directly.