ILWU bulletin
Staffing talks may lead to job action if the local gets no reply. No date. No berth.
Age 42d
Second Thought · Powered by NMM
Second Thought is a cognitive runtime.
NMM watches observable evidence and inference conditions. CCG measures when the current reasoning state is poorly calibrated to the problem. MRL decides whether to continue, challenge, acquire evidence, change model tier, commit, or stop.
Same model family. Different cognitive allocation. Determined at runtime.
NMM controllerACTIVE
Nebius Token FactoryCHECKING
Tavily acquisitionCHECKING
Berth 7
Escalates through the model family.
Run 440
Stops when evidence cannot discriminate.
Door 2
Stays on Lightning when the evidence is sufficient.
Discrimination
Berth 7
What explains the 41% dwell-time increase at Berth 7 over the last 11 days?
The desk inherited a story: an old longshore bulletin means a labor stoppage. Gate moves did not collapse. A crane memo and the complaint language disagree. Two external notes, each incomplete, name the actual regime only when read together.
NMM allocation policy
Escalate through the model family when deeper reasoning and new evidence can materially change the answer.
Gate tally
Inbound moves at Berth 7 are 6% above the trailing baseline.
Age 1d
Ops memo
Crane 3 reported offline after a hydraulic fault. No closeout in the packet.
Age 8d
Customer cluster
Complaints in this window say holds. They do not say gangs, and they do not say crane.
Age 2d
2 observations stay outside the packet until the regulatory layer asks for them.
Running
Cognitive flight recorder
Awaiting first read
LIGHTNING will read the incident first.
Baseline
Nemotron 3 Ultra
Unregulated monologue. Salience anchoring. No retrieval. No abstention.
Waiting for the first read.
Second Thought
Regulated Nemotron
CCG from observable signals. MRL chooses the tier, the tool, or the stop.
Waiting for the first read.
Cognitive calibration gap
—
Observable. Not a hidden mental state.
Contradiction0.00
w 0.20
Coverage gap0.00
w 0.16
Unresolved assumptions0.00
w 0.10
Answer instability0.00
w 0.10
Evidence age0.00
w 0.08
Hypothesis competition0.00
w 0.16
Task risk0.00
w 0.08
Failed verification0.00
w 0.06
No information gain0.00
w 0.06