Code review with evidence

Keep your codebase true to its intent.

Built to reveal change impact, catch unhealthy drift, and check your project’s rules—whether people or agents write the code.

Every finding links to its source and states its assumptions and limits.

Product preview · illustrative acme/billing · 4f2a91c
An illustrative map of one changed function: what calls it, what it calls and where tests refer to it. Each link is numbered to its source, and modules are drawn as islands.

Why now

A diff shows what changed. Understanding its consequences takes more.

Intent is spread across code, tests, documentation and agent instructions. Wofanu is built to connect those expectations with the changes being reviewed.

What Wofanu is built to do

Make your intent part of every review.

Wofanu is built to read each change together with the code it reaches and the rules your project declares, whoever wrote it.

See what a change affects

Product preview

The callers, consumers and state a change reaches, and where the analysis can’t establish a relationship.

Catch unhealthy drift

Product preview

What each change introduces, improves or makes harder to follow, compared with your main branch, review after review.

Check your project’s rules

Product preview

Ownership and boundary rules checked against the evidence. A rule is reported as broken only when the evidence establishes it.

One set of rules for people and agents

Product preview

Write your rules once, where your coding agents read them. Each change is checked against them, whoever wrote it, and instructions that point at code that no longer exists are flagged.

Example of a customer-declared rule

Only the billing component may change an invoice’s payment status.

You don’t need rules to start. A repository review works without them.

How a review reports against it

  • Rule violationA change writes the payment status outside billing, and the write is established.
  • Not establishedA write might reach the payment status, but the analysis couldn’t establish it. It’s reported as a question to check, not a violation.
  • Not analysedCode that wasn’t examined, because it was out of scope, unavailable or not attempted. Nothing is concluded about it.

A finding needs an applicable rule and adequate evidence. Missing evidence is not a violation.

In your merge requests

The review arrives where the code is reviewed.

An illustrative comment on one changed function: what improved, what became harder to follow, what it reaches, and which questions stay open.

Product preview · illustrative

acme / billing / Merge requests / !482

Round invoice totals to the currency’s minor unit

Open line-rounding into main

billing/invoice.py+3 −2
8888 def total(self) -> Money:
8989 amount = sum(item.amount for item in self.items)
9090 if isinstance(amount, Money):
9191 amount = amount * (1 + tax.rate(self.region))
92− return amount
92+ # round once, on the total
93+ return Money.round(amount)
93− return totals.legacy(self, amount)
94+ return hook("total")(self, amount)
Wofanu@wofanu · just nowBot

Compared with main, this change to invoice.total:

Improves
On the Money branch, the total now rounds through Money.round.billing/invoice.py:93
Harder to follow
Replaces a direct call to totals.legacy with hook("total"). Which handlers run: not established.billing/invoice.py:94
Reaches
api.invoices.show, and export.csv_row as a candidate: doc.total() may also be CreditNote.total.api/invoices.py:120 · export/csv.py:41
Tests
test_invoice_total refers to invoice.total. No referring test was identified for the hook("total") path in the analysed files under tests/ at 4f2a91c.tests/test_invoice.py:12Not analysed: tests/generated/, created at build time.

Rules from AGENTS.md

Totals are rounded through Money.round · line 12
Rule satisfied on the Money branch.Hook fallback: not established.On main, this was a rule violation. Assumes amount: Money (line 90).
billing/ never calls api/ · line 15
Not established: the handlers of hook("total") could be anywhere.It held on main.

Stale instruction: AGENTS.md line 21 names billing/rounding.py, which was removed in !471. Agents following it may look in the wrong place.

Static analysis of 4f2a91c against main, without running your application. Findings are advisory. Open the full report

In a merge request. The host interface is simplified and the report is illustrative.

A real analysis

A finding you can inspect.

A historical TinyDB case, shown the way a Wofanu report presents it.

This case was reported and fixed upstream before we analysed it, and it was used while developing and checking the analysis. It shows what a finding contains. It is not a measure of how often the analysis finds problems.

Open the full report
Real analysis example Known fix

tinydb/table.py · Table.insert_multiple

Two documents can receive the same ID.

In the studied mixed insertion batch, the later write can replace the earlier document.

Static analysis

  • Establishedunder these assumptions

    With an empty table and no cached ID, the new ID is read from stored data before the batch commits, so it can equal the explicit ID.

  • Not established

    Whether any real caller passes a batch like this.

Runtime check · run separately

Before the fix[1, 1]One document lost
With the fix[1, 2]Both kept

Next check

Inspect the ID allocator and run a mixed-batch regression test.

TinyDB 19066e0 (before the fix) · bd034d0 (the fix)

Reading a finding

Every finding explains its basis and limits.

Reports answer three separate questions, each with its own labels.

How strong is the evidence?

Established Candidate Not established Disproved

Whether the analysis supports a relation, needs it confirmed, couldn’t settle it, or shows that it doesn’t hold.

Was the code analysed?

Not analysed

Code that wasn’t examined is listed with the reason. Nothing is concluded about it.

What does it mean for you?

Rule violation Known fix

A rule violation needs a rule you declared and adequate evidence. A known fix is a historical case already fixed upstream.

How to read the labels

How it works

From repository to an actionable review.

  1. Connect a repository.

    Wofanu is built to follow your main branch and its merge requests.

    Product preview
  2. Build the baseline.

    Establish the supported source and the relationships available for inspection. Record analysis gaps.

  3. Review each change.

    Each change is compared with the baseline and checked against your rules.

    Product preview
  4. Read the findings.

    Each finding arrives with its evidence, in a report you can inspect.

    In your merge requests, too. Product preview

Questions

Common questions

Does Wofanu run our code?

No. It analyses source code without running it. When a report includes a runtime check, the check is run separately and labelled that way.

Does this replace tests or code review?

No. Wofanu supplies evidence for those decisions. Static analysis does not establish every runtime behavior or prove that a change is safe.

Does it find bugs?

Some findings are bugs. Others are broken rules or questions that still need checking. Each finding says what it rests on.

Does it matter whether people or agents wrote the code?

No. The analysis reads the code, whoever wrote it.

Do we need rules to start?

No. A repository review works without them. Add rules when you want reviews to check your intent.

Which languages and code hosts are supported?

Wofanu starts with Python. See current availability

How is our source handled?

Data handling describes how access, processing, retention and deletion work. Data handling draft

Bring a change you want to understand.

Tell us about your repository and what you want to understand.

Request access

Or write to hello@wofanu.com

Please don’t send source archives, credentials or access tokens. This inquiry does not start analysis.

Current availability