Evidence
The failure this product exists to avoid is a confident, unauditable picture of an organisation. Everything below is a structural decision taken to make that failure hard rather than a discipline anyone is asked to maintain by hand.
Nothing is edited in place. A correction is a new entry that supersedes an old one, and the earlier state remains reconstructable. You can always ask what was believed before a correction arrived, and get a real answer rather than a reconstruction.
Every file is fingerprinted by a hash of its contents. Two files with the same name and different contents are two different sources, and a file that changes after ingestion does not silently become a different thing underneath a finding that already cited it.
Not a document, not a page — a byte range in a text file, or a specific row and column in a register. When the finding is rendered, the anchor is re-checked against the stored bytes rather than trusted, and a broken anchor is shown as broken rather than quietly dropped.
Findings, and claims created from mapped source regions, are content-addressed, so re-running an assessment over unchanged material reproduces the same identifiers rather than a fresh set that merely looks similar. A finding that persists across runs is demonstrably the same finding. Claims an analyst enters by hand are identified when they are recorded, not derived from their content.
Grading is computed from the evidence beneath a finding, never written by hand. One person’s recollection does not become fact because it agrees with another claim, and a finding resting on an unevidenced assumption is labelled as resting on an assumption.
There is no model anywhere in the chain from your documents to your findings. Registers are read by written mappings; prose is anchored by a person. The result is that every step can be checked by re-reading the source, and nothing has been invented in between.
This list is here for the same reason the rest of the product exists. A tool that hides its own limits is asking to be trusted further than its evidence goes — and that is the failure mode, not an edge case.
Systemantic reports the patterns it has rules for. Silence from a rule is not evidence that the corresponding problem is absent — it means that rule did not fire on the material it was given.
There is no severity, no score and no ranking, because nothing in the evidence supports one. Prioritisation is a judgement about your organisation and its context, and it belongs to you.
It can tell you a control is recorded, and whether anything measured shows it operating. Whether it actually works is a judgement the vocabulary deliberately cannot carry.
Only that the reading is reproducible and written down where you can check it. If a mapping misreads a column, the misreading is visible and correctable rather than buried.
If two rows in a register share an identifier, Systemantic currently treats them as one thing, and nothing records that a merge happened. Whether a supplied identifier is an identity is an open design decision, and until it is made a finding can rest on a merge nobody chose.
Everything proven so far was proven against Kelmarsh Mutual, a synthetic corpus written by the same people who wrote the rules. Real material is untidy in ways we did not choose, and that is precisely what the private beta exists to discover.
Registers and logs are read automatically. Policies, procedures and interview notes are anchored by an analyst, sentence by sentence. For a policy-heavy pack that remains most of the work.
Organisational documents are confidential, and an assessment needs the real ones. During the private beta, handling will be agreed in writing with each organisation that takes part, before anything is shared — what is provided, where it is held, how long it is kept, and how it is destroyed.
This website accepts no documents and holds no organisational material. It is a public description of the product. There is no upload here, and nothing you can submit through it.
Entering private beta
Systemantic is entering private beta and is not open for signup. We will begin with a small number of organisations.