Do you know what your systems actually do?
Which paths through your systems carry the numbers that matter?
Which of your data is critical, and where does each field come from?
Where does the transformation logic live, and who still understands it?
Before you approve a release, can you show what it reaches?
Cosmogenesis reads the source code your organisation runs.
A system of proof for the people who answer for those systems.
Moments you may recognise
The auditors have asked how the month-end total is produced. The report has run for years. Nobody here can show the workings.
The release touches the pricing tables. The team says the reach is small. I am the one approving it.
The engineer who built our settlement system leaves in four weeks. Nobody else knows how it works.
Why the code
The workings are in the code.
A data catalogue will tell you that a report reads a particular table, and that is useful as far as it goes. What it cannot tell you is what the code does to the numbers on the way through, whether a value is divided on one line, joined to last month's rate on another, or sent down a branch that only runs at year end. The figure on the report is decided in that code, and it is rare for anyone to have read it in full.
When a number is challenged, someone has to read that code. Cosmogenesis does that reading from the source, statement by statement, and draws the map from what the code actually does, with a grade on each relationship for how much the source could show. Where the source runs out, the reading says so.
What Cosmogenesis does
Cosmogenesis reads the source code your organisation runs.
That reading draws the map, which is the lineage, so you can see which systems feed which and what each figure depends on.
What you get
The answers you can put in front of management.
A map of what depends on what
Which systems feed a result, and which results a system feeds.
A plain-English explanation of any number
Where a figure came from, step by step, in words management can read.
A list of where you are exposed
Where one procedure, one rule with no name or one divisor holds the rest up.
A way to test a change before it is made
Which reports and figures a proposed change can reach, before anyone approves it.
A record of what is really there
It is read from the code your systems actually run, so what it shows is established from the source. A model can read a map once it is given one, although it cannot establish the map itself.
Documentation comes from the same reading and is regenerated when the code changes. The reading works only from the code you give it.
Watch it, then check it
The story comes first and then the evidence.
The story follows one change through a company at night, and the evidence takes one reported result and shows the code behind it. The reading, written down with that code, is the record.
The record behind one reading
Inspectable evidence record
Source-derived change evidence
Synthetic Theory fixture. No customer data.
- Identity
- field-edge:sha256:1825858b710bc509192def233bfcae2fff9a934df33944230e8015e6d207b47a
- Scope
- 8 source artefacts
- Source snapshots
- frame-a site/lab/theory/fixtures/frame-a/02_normalise.sql
sha256:2c66deeb8b98f630e2be7622f7001d2e542ab1864b196c350fa1a4d933ded586bundle:77e28a8882dd399ee1973d2f3ea6baca03e12aa13bccf91a51c5fa270abb0e64 - frame-b site/lab/theory/fixtures/frame-b/02_normalise.sql
sha256:ec781e77cf8463b3f5f4fce74c535dca2e8f62bdbc69fd9b1d468d511947b5efbundle:85efddd930af18c955cdedabd3ea92f5a577ead1ed1d1f9c774b3d9efb5d6426
- frame-a site/lab/theory/fixtures/frame-a/02_normalise.sql
- Selected source span
site/lab/theory/fixtures/frame-b/02_normalise.sql106-122 · inclusive character offsetsraw_value / 100.0- Derivation
- changed edge -> product resolver bridge -> typed operator -> checked source bytes
- Changed relationship
01_read.raw_value -> 02_normalise.normalised_valueraw_value / 10.0 → raw_value / 100.0- Boundary
- External path-grade lift
dscp.lineage.provenance_semiring.lift["external"] → rejected · ValueError - Expiry
- Regenerate when a named source snapshot, interpreter version, generator version, or reading condition changes.
One system to begin.
Choose one system that matters, and its code decides the next step.