5.3.11 is about refusing to lie quietly. If a shared-brain runtime silently rewrites malformed structured input into some other valid state, it may keep going, but it leaves behind poisoned traces: a task that looks failed when it was really malformed, a decision weighted as high urgency when the caller did not ask for that, or a protocol record that says “answer” because the original task type was invalid and the runtime guessed.
What surfaced after 5.3.10
Once packaged runtime truth, deep doctor bootstrap honesty, Evolution telemetry, and synthesis ingestion were repaired, the next real gap was smaller but more insidious. Some protocol and Cortex paths still treated malformed structured values as a cue to coerce them into another valid value instead of rejecting the payload.
nexo_task_close could take an invalid outcome and silently convert it to failed. Several public tool paths accepted invalid task_type values and quietly downgraded them to another valid type. And nexo_cortex_decide could treat a malformed impact_level as high. None of that crashes the runtime, but it does damage trust because persisted semantics stop matching the original request.
Why malformed close outcomes were the worst case
A malformed outcome being rewritten to failed is not a harmless default. It changes protocol history, shapes debt, appears in recent context, and can make it look like the runtime or operator genuinely failed a task when what actually happened was a bad caller payload.
5.3.11 closes that path directly. Invalid close outcomes now return an explicit error and the task stays untouched. No status flip, no fake close timestamp, no contaminated failure trace.
Why task-type coercion mattered across protocol and Cortex
task_type is not cosmetic metadata. It influences how NEXO interprets work, when it should demand evidence, how it scores response confidence, and how Cortex frames risk and alternatives. Quietly changing an invalid task type into a valid one means the runtime is no longer processing the task the caller actually submitted.
5.3.11 hardens both sides of that boundary. nexo_task_open and nexo_confidence_check now reject malformed task types, and the same is true for nexo_cortex_check and nexo_cortex_decide. The runtime now says the contract is wrong instead of inventing a corrected meaning.
Why impact level needed the same treatment
Cortex decisions use impact_level to weight tradeoffs. If a malformed value is silently promoted to high, the persisted evaluation starts carrying a stronger urgency class than the caller actually asked for. That undermines the point of having explicit, inspectable decision support in the first place.
5.3.11 now rejects malformed impact levels explicitly. If the caller wants a high-stakes evaluation, it has to say so with a canonical value.
Why the internal DB helpers also had to change
Public-tool validation alone is not enough if internal code paths can still write malformed semantics into the database. That is why this patch also hardens the DB helpers behind protocol tasks and Cortex evaluations.
The same canonical checks now exist before those rows are persisted. That keeps internal callers from bypassing the public hardening and pushing misleading semantics into task history, debt, context, or evaluation records anyway.
What changes for the operator
The visible effect is simple: when a caller is wrong, NEXO tells you the contract is wrong. It no longer keeps the system moving by silently picking a different meaning. That makes protocol traces, decision telemetry, and trust signals more faithful to reality, which is exactly what a shared-brain runtime should optimize for over time.
Upgrade
Open the 5.3.11 changelog section for the exact release note. For installed users, the intended path looks like this:
npm install -g nexo-brain@5.3.11nexo updatenexo doctor --tier runtimenexo doctor --tier deep
The headline is not a new feature. It is a more trustworthy shared brain: malformed structured payloads fail loudly, valid ones keep their original meaning, and the rows NEXO stores about your work better match what really happened.