Quality
How we guard quality.
This platform is AI-generated and makes no secret of it. Trust here does not come from a name on the door but from a process you can check: every claim is tested against its source, every correction is counted, and every lecture carries the date of its last check.
Two layers: cast wide, verify hard
A review panel of specialised AI agents reads every lecture through five lenses: rigour and source accuracy (does every reference match what the work actually says), the evidence label (is a heuristic ever presented as a law), internal consistency (do summary and text agree, do cross-references to other lectures hold up), practice (does this survive contact with a real campaign) and clarity (does a reader get lost anywhere).
What that wide layer finds is not yet truth. Every finding then passes a second, adversarial verification: a heavier model that puts the report next to the source file with instructions to refute it. In our most recent round, more than half of the candidate findings died on exactly that test. What survives can be made hard.
How decisions get made
Verified findings are split in two. Indisputable corrections — a wrong year, a source cited but missing from the source list, a summary promising something the text does not deliver — get applied. Anything requiring an editorial judgement is weighed within the review methodology itself, along the five lenses; the human editorial team directs the topics, the structure and the method — not every individual sentence.
Corrections are always applied in both languages at once, and every change passes the same automated check that guards the whole curriculum: every lecture carries an evidence label, and the build refuses content with broken structure.
Your report is input — anonymous by design
Under every lecture sits a report button: factual error, missing source, unclear, or wrong evidence level. No account needed. We store no name, no e-mail address and no IP address — only the lecture, the category and an optional short note. That note is erased the moment the report is processed in a review round.
We knowingly pay the price of that anonymity: we cannot recognise or reward good reporters. The panel is the engine; reports are the extra pair of eyes that aim it.
The audit trail: checkable, without names
At the bottom of every lecture you can see when it was last checked against its sources, and how many corrections review rounds have applied. No names, no signatures — a date and a count that are true. That is the deal: we claim no human proofreading that did not happen; we show the process that does.
The numbers so far
In the most recent full round (July 2026) all 300 lectures were reviewed. That produced 331 candidate findings, of which 183 died under adversarial verification as unsustainable. Over 160 verified corrections and sharpenings were applied, in both languages. Every one of the 300 lectures carries an evidence label; the automated check stands at zero errors and zero warnings.
These figures are updated per round. They are aggregated and cannot be traced to individuals — because there are no individuals to trace to.
What we deliberately do not do
We put no reviewer names on the platform and store no personal data around this process. We do not pay for reports or reviews, so there is no incentive to score rather than improve. And we never claim more than the process delivers: where the evidence is thin, the label says so — and where we are wrong, the report button is the shortest route to telling us.