The gate is closed by default
A claim that fails verification does not get softened, footnoted or waved through. The module does not publish until the claim is fixed or removed. Closed is the resting state.
Provenance
You will ask how you know it's true. Here is the answer.
Anyone can generate content that reads as authoritative. The question a board or a regulator asks is different: how do you know? SynTifika is built to answer it. Four things travel with every statement it produces, and none of them depends on trusting the AI.
The four assurances
What travels with every claim.
01
A source you can open
Every claim is linked to the exact document, page or clause it came from, not a reading list at the end. When a learner or an auditor wants to know why the course says what it says, the answer is one click away: here is the provision, in the actual instrument. A pinpoint, not a bibliography.
Traceable to primary source02
Checked by a machine, not an AI
A deterministic, non-AI checker reads each citation and confirms the cited text actually exists in a stored copy of the source. It is a text comparison, so it cannot hallucinate, cannot be persuaded, and does not have an opinion. If a citation does not resolve to real source text, it is flagged. This is the floor beneath everything else.
Deterministic verification03
A named person signed it
Before a module publishes, a named reviewer approves it, and the record shows who and when. Automation does the production; a person remains accountable for what goes live. On regulated or high-stakes content, factual and legal points are routed to a subject-matter expert before that approval.
Human accountability04
The date it was true
Every statement carries the date it was current. Rules change; a course should be honest about when it was last confirmed. An auditor can see at a glance whether a claim still holds, and the pipeline can re-check content against a changed source and flag what needs refreshing.
Dated and refreshableChain of custody
One claim, traced from the source to the certificate.
Provenance is not a folder you keep to one side. It is the spine of the content. A single claim keeps its link to the source at every step, through the script, the quiz, the video, and the record of who completed it.
Follow one statement, end to end
Instrument
Source policy
Provision
Clause 5(1)
Script
Lesson line
Quiz
Question 3
Video
02:14
Certificate
Issued
Change the clause, and the pipeline knows which lesson line, which quiz question and which video second depend on it. Nothing goes stale silently.
Fail-closed
What happens when a claim can't be proven.
The honest test of a validation system is what it does when something fails. Most do nothing, because they only look. Ours stops.
Corrected, then re-checked
The failing claim is re-grounded or rewritten and run through the same checks again. Fixing one thing cannot quietly break another, because the whole module re-validates.
The refusal is visible
When the gate stops something, that is a feature you can see, not a silent pass. The difference between "checked and clean" and "never checked" is never invisible.
Research ≠ grounding
This is the distinction the whole product turns on. Research explores a subject and can sound completely authoritative on a guess. Grounding proves it: every claim tied back to a real source you can open. Most AI content tools do the first and skip the second, which is why their output reads well and cannot be defended. SynTifika does both, in that order, and will not publish anything that fails the second.
Trust is not a tone of voice. It is a chain you can follow.
See the proof
Open a claim, and follow it back to the source.
In a walkthrough we take one statement from a finished course and trace it all the way to the clause it came from, then show the gate refusing a claim that cannot be traced.