Your system does a thousand things nobody can explain any more.
Traceway connects to the places your work actually lives, finds the decisions already written down in them, and gives anyone in the company one way to ask the question no tool has ever answered: why does this exist, who decided it, and does it still hold.
Reads what you connect. Never scores a team, never blocks a release, and never files anything a linter could have filed.
Why do we hold dead letters for exactly 14 days?
A retention window of 14 days, agreed at the operations forum on 12 March, signed by Priya N.
“Dead-lettered deliveries are retained for 14 days and then discarded.”
30 days. Storage cost was the stated reason, and the reason is on the record.
Nothing. This is still current, so the code is right.
You can already show the decision was made. Nobody can show it actually happened.
A closed ticket is evidence that somebody pressed done. It is not evidence that the thing exists, that it does what was agreed, or that the reason for it still holds. Most organisations discover the difference during an audit, a migration, or an incident, which are the three worst moments to find out.
Decided, and nobody built it
The decision is signed, the tickets are closed, and nothing in the running system does what it said. This is invisible to every project tool, because they all read the tracker.
Built, and nobody knows why
A rule, a limit, a cut-off time. It is load bearing, it is old, and the person who set it left in 2019. Every change near it is a gamble.
Decided twice, and only one took
The rule changed and the system kept the old one. Nothing failed, nothing alerted, and the record and the reality quietly disagree.
Finding out why something works the way it does is currently archaeology.
Blame the line. Find a commit message that says fix retry logic. Find the ticket, which describes what to do and never why. Search four hundred chat messages. Ask the one person who was there, if they still work here. Give up, keep the behaviour, and add a comment saying do not change this.
One question, and an answer with its sources attached
Ask about a file, a service, a setting or a table. Traceway answers from the record: the decision, the date, who signed it, the constraint in the words that were agreed, the outcome it serves, and whether anything has replaced it since.
The alternative that was considered and rejected. It answers the most expensive question in software, which is whether the obvious idea has already been tried, and nothing else in anybody's toolchain holds it.
No decision in the record explains this. Somebody decided it and it was never written down. Here is what was searched, and here is what was out of scope.
That is the most useful moment in the whole product. The person asking wants to know, right now, and either has the answer or knows exactly who does. One tap turns it into a record.
Nothing decided and something you are not cleared to read look completely different. A restricted decision shows that it exists and offers you a way to ask, because knowing a decision exists is how somebody knows to ask for it.
Your decisions are written down. Just not anywhere anyone would call a decision log.
Architecture notes, standards, policy documents, control statements, committee minutes, change records, readme files, code comments citing a memo from 1998. Traceway reads what you connect, finds the material that is genuinely decision shaped, and proposes each one as a record with the passage it came from and a link that opens exactly where it was found.
Nothing is asserted
Every find arrives as a proposal with its source. A person confirms it, edits it, or rejects it. Nothing enters your record unconfirmed, ever.
A recovered record says it is recovered
It carries a distinct status, it is counted separately, and it never satisfies an approval on its own. A record assembled from a configuration value and a fragment of an old standard has not been signed by anybody, and it does not pretend to have been.
Somebody who knows can vouch for it
Attesting says this matches my understanding of what the system does today. That is a smaller, honest act, and it is one a person who arrived in 2018 can make truthfully about a rule set in 2004.
You had no recorded decisions for this system. You now have 340 recovered records, 61 of them vouched for, each carrying its sources. 214 behaviours remain unexplained, ranked, and 38 of those are load bearing with nobody left to ask.
Most of these decisions are not coming back, and we will not tell you otherwise. What changes is how much does, and knowing precisely which part still has no explanation. For anyone replacing an old system that list is the single most valuable thing in the room, because it is exactly what gets silently dropped when the old system is switched off.
How much of your own record could ever be checked against reality?
Retries are capped at six attempts is checkable. Improve reliability is a direction. The difference decides whether a decision can ever be verified against the thing that got built, and no organisation has ever measured it about themselves.
It needs no connection to anything, because it reads your decisions and nothing else. It is usually the first finding that changes how people write them, and it is the reason everything further down this page is possible at all.
Decisions with a bound anyone could test
A number, a limit, a must or must not, a named scope.
Decisions stating a direction only
Real decisions, properly made, and unverifiable by anything.
Decisions with no date that took effect
They cannot be ordered against anything, so nothing downstream can reason about them.
Every number Traceway shows you carries what it was computed over, next to it. We do not publish a figure whose denominator we cannot state.
Five answers, and two of them cannot come from anywhere else.
Once decisions carry testable constraints, Traceway can compare them with what your systems actually do, and report one answer per pair. Sorted by the outcome that is losing, not by the age of the finding, and every count carries the scope it was computed over.
Every one of these can be closed honestly. Create the work, or record why it is not a problem: descoped, deferred with an owner, deliberately left on the old policy. The reason goes on the record, dated and attributed, and the finding closes without anybody pretending something was built.
A decision signed at four in the afternoon can invalidate work that nobody touched.
Every tool that watches your code waits for something to change. Most of what goes wrong here involves no change at all: a policy is replaced, a limit is revised, an obligation is updated, and three systems carry on implementing yesterday's rule perfectly. Traceway watches both sides, so the moment the reason changes, the work built on the old one is answerable in one list.
It says when it last looked
Every figure carries its date and its scope. A stale reading presented as a current one is the one failure a system of record cannot have, so we show it as stale and offer to run again.
Anything it could not see is counted separately
Not observed is its own answer. It is never folded into the good number to flatter a report, or into the bad one to manufacture urgency.
And it never says work was wrong
Something built correctly under the old rule was correct when it was built. The finding says so in its first line, because a queue that blames people for compliant work stops being opened.
Three teams did something else. The likeliest explanation is that the decision lost.
Teams do not accidentally converge on the same deviation. When several parts of a system independently do something other than what the record says, Traceway proposes a new decision to supersede the old one rather than filing a complaint about the code, and it shows you the evidence it reasoned from so you can disagree with it.
Whatever you decide is what gets recorded, and the challenge is logged either way. That is the difference between a record you can trust and a record everybody works around.
Names a person, a team or a squad. A finding is a property of a rule and an implementation, never of who wrote either.
Rewrites the record to match the code, or the code to match the record. A human decides which side was right.
Blocks a release, gates a merge, or stops anybody shipping anything.
Files anything a linter, a scanner or a code review tool could have filed. We compare your work with your reasons, and nothing else.
Four thousand repositories cost the same as forty.
Because Traceway does not read your estate looking for things. Your decisions name what matters, and only those places are ever resolved. The work is set by the size of your decision record, which is small and human sized, not by how much code you own.
Names, not contents
By default Traceway reads paths, names, settings keys and the text of documents you nominate as decision sources. Not the contents of your code.
Scope is a rule you write
All repositories in this group with this label, excluding archived. New matches are proposed to you rather than absorbed silently, and every addition and removal is in the audit log with its date.
It looks where decisions live
Listing a repository is nearly free; reading one is not. So it narrows on cheap signals first and reads almost nothing. Of 4,120 repositories, 63 held written decisions, and it tells you that number too.
Wherever your work actually is
Not only modern repositories. Document libraries, exports, configuration, records archives and change management systems. Old estates are the ones with the most to recover.
Traceway keeps the finding and a link. Your source is not stored, and it is never used to train a model.
Named by you or matched by your rule, revocable in place. Turn it off and the findings freeze with their date rather than quietly becoming clean.
One connection at the top, delegated scope underneath with a name against it. Adding a system can never widen who is allowed to read a decision.
It has no opinion about your code.
Everything else in this space compares your code with itself: style, bugs, complexity, conventions. There are good tools for that and we are not one of them. Traceway compares your work with the reasons you agreed, which is a different question and the one nobody is answering.
That distinction is also a promise about how it behaves. If a linter could have found it, we do not file it.
Not a review tool
No quality findings, no bot in a review thread, no editor plugin.
Not a scoreboard
Nothing here rolls up to a person, a team or a repository, and there is no setting that turns one on.
Not a gate
It never blocks a merge or a release. If you want a check in your own pipeline, the state is available and it is your choice.
Not surveillance
It reports on rules and implementations. Who wrote what is not a field in any finding.
Not a project tracker
Your work stays in the tracker you already run. Traceway holds the reasons, not the tasks.
Not a compliance certificate
It produces evidence about your systems. It does not discharge anybody's obligation, and we will not imply that it does.
The questions people actually ask, answered without the hedging.
Does our source code leave our systems?
By default nothing about the contents of your code is read at all: Traceway reads names, paths, settings keys and documents you nominate. Checking a specific constraint against a specific place is a separate permission you grant per system, and even then we keep the finding and a link, never the code. Institutions that cannot let source leave at all can run that part inside their own boundary.
Is this going to be used to measure engineers?
No, and it is built so that it cannot be. A finding is about a rule and an implementation. Who wrote either is not recorded on it, no view rolls anything up to a person or a team, and there is no setting that turns that on. If your leadership wants a productivity dashboard, this is the wrong product.
Will it block our releases?
Never. It reports state after the fact. If you want to wire that state into your own pipeline you can, and that is your decision to make, not a default we ship.
We have thousands of repositories and no convention. Does this work?
Yes, and the size of your estate barely affects it. Your decisions name the places that matter, so only those are resolved. Scope is a rule you write rather than a list you maintain, new matches are proposed rather than absorbed, and a monorepo is scoped by path.
Most of our decisions were made twenty years ago by people who have left. Is there any point?
This is the case where it is worth the most and where we are most careful about what we promise. Most of those decisions are gone and we will not pretend otherwise. What you get is the share that is recoverable, each with its source, plus a ranked list of the behaviour that remains unexplained. If you are replacing that system, the second list is the one that saves you.
How is this different from an AI code review tool?
They compare your code with itself. We compare your work with the reasons you agreed. If a linter could have found it, we do not file it, and that is a rule rather than a positioning statement.
Can it tell us something is wrong when it is not?
It can, which is why every finding has to quote the words it is testing and point at the exact place it looked. A finding that cannot produce both is not filed. Anything it could not read is reported as not read, rather than assumed to be missing.
What happens when we replace a decision?
Work built under the version you replaced is listed immediately, with the date the rule changed, and with the fact that it was correct when it was built stated first. Nothing in your code has to change for that to appear, which is why we watch both sides.
Find out why your systems do what they do.
Join the waitlist, book a demo, or just get the details, and get a direct line to the team building Traceway. No drip sequence.
Launching first in the United States and Canada. The United Kingdom and the European Union are planned for 2027–28, with the rest of the world to follow. Join from anywhere and we’ll tell you when your region opens.