Recurring, defensible documents get redrafted every cycle. Fynto captures every reviewer decision and brings it back next cycle — as a question when the draft disagrees, an offer when it doesn't. Never as an answer.
Transfer pricing reports, audit workpapers, regulatory filings, credit reviews — all written every year about mostly the same subjects. The report records the conclusion, not the alternatives considered or the evidence weighed. When the person moves on, so does the file.
Every AI drafting tool picks one of these. Both fail defensibility.
Always draft from this cycle's evidence. Also remember, durably and with context, every judgement a human made last cycle.
Software forgets.
Fynto remembers.
The compounding advantage
Analysis flows through a single connected pipeline — so every number in the final report traces back to the evidence that produced it, and every remembered judgement knows exactly where it belongs next year.
Drafting agents extract this cycle's positions from this cycle's evidence. They do not know the ledger exists. The review workflow, where a human confirms or corrects the agent's output, is the only writer and the only reader. That independence is what makes cross-cycle comparison meaningful.
Every gate answer, dropdown pick and text save becomes one append-only row. Same shape regardless of what it's about — so a new kind of judgement is a key function and a gate, not a schema migration.
"business_key": "far|acme services gmbh|it_support",
"topic": "far_characterization",
"fiscal_year": 2025,
"prior_value": "limited_risk_service_provider",
"chosen_value": "entrepreneur",
"disposition": "confirmed_change",
"rationale": "Acme took on DEMPE functions
for the platform IP in FY2025.",
"supersedes_decision_id": null,
"context": { /* agent proposal, evidence, source */ }No auto-carry, no inferred identity, no bulk confirmation, no feedback loop into extraction. The recalled value is an offer, never a default.
The ledger is generic. It currently captures four kinds of judgement inside the transfer pricing workflow — and the storage layer didn't change when we added the fourth.
The layer runs when a reviewer opens a workflow step, and again on every decision they make.
A business key from stable external references — ERP refs, agreement codes, registrant ids. If it doesn't resolve to exactly one subject on both sides, we skip. A recall gap, never a false flag.
A genuine position change becomes a blocking gate: confirm, revert, or leave open. Revert restores the prior value server-side. No model between the reviewer's click and the row that gets written.
Where there's no contradiction, last cycle's value and rationale sit under the current one with a Reuse button. Every reviewer action — gate, edit, reuse — is one append-only row.
Several dozen controlled transactions. A handful of entities characterised across categories. A benchmarking study that screens hundreds of candidate comparables down to twenty. Almost every item is a human call that has to be defended to a tax authority, sometimes years later.
Fynto works with the comparables database you already license — no new subscription, no switching. Export the search output, or hand over the accept/reject matrix your advisory firm already worked, and Fynto screens it, records every accept and reject as ledger rows, computes the arm's-length range, and carries the result through method selection into the report.
Uploaded comparables live in your workspace only. Every accept and reject is one row keyed by the normalised company name — so next cycle a rejected comparable comes back with its reason.
A stable subject, a closed set of positions, a named human, and a reason. The ledger already has the columns.
Talk to us about a design-partner engagement|
|
POPULAR
|
|
|
|---|---|---|---|
| Who manages infrastructure | Fynto | Your team | Your team |
| Where your data lives | Our cloud | Your cloud account | Your data center |
| Data stays in your boundary | |||
| AI analysis | Full | Full | Optional |
| Enterprise controls | Standard | Full | Maximum |
| Best for | Mid-market | Enterprise | Regulated / Tax-sensitive |
A 30-minute walkthrough of the Judgement Layer, on your data.