Research rounds four to six, combined
Three rounds of reading on counterevidence, uncertainty, and the separable mechanisms behind analysis. Here is what the work found, and where it stops.
Round four: seeing evidence is not using it
Information that argues against a held answer can be present in brain activity and still be under-used once the answer is committed to. Confidence in a first judgement predicts reduced processing of later disconfirming evidence. Separately, language models use decisive passages less reliably when those passages sit in the middle of a long input. Both findings treat position and commitment as things to test, not assume.
Round five: analysis is not one operation
Attention depends on circuit coordination rather than a single switch. Holding something in mind is not one continuous signal. Knowledge is organised around relationships, and confidence is computed from several inputs rather than read off accuracy. Treating these as separate mechanisms is why our pipeline keeps reading, locating, reviewing, and release decisions apart instead of collapsing them into one model call.
Round six: planning, consolidation, and where the theories stop
Offline processing has more than one route, and model-based control only pays off under particular conditions. The honest reading is that no single memory or planning process explains good reasoning, so we avoid building the system as though one does.
What we built, and what we did not
These rounds produced an evidence-use audit that runs on request and stays off in normal analysis until it has been evaluated. Other ideas remain proposed, not built. None of the cited work tested Encevra or Noetis, none of the authors have evaluated our systems, and nothing here reproduces a human brain.