meta: add contributing/roadmap/citation/architecture notes

hf: add model card metadata (license/tags)

docs: add preliminary results + reproducible 4k panel config/scripts

bag: bihmoe-bag_20260222-030013 (snapshot)

train: schedule-2 curriculum w/ ramp + braid mix (single-run)

train: schedule-2 curriculum with ramp (single-run, no restarts)

train: add schedule-2 curriculum (phase switch for long/noise mixing)

train: add curriculum mix (perturb + long) to drive ood_long learning

arch: add chiasm_mode=bias to avoid global-branch starvation

ui: add KEYLINE (non-wrapping) summary for logs/clipboard

trainer: pass output_gate/output_tau to structured model

arch-v2.3: confidence-gated output fusion (stabilize ood_long)

arch-v2.2: add QUERY type + soft optic-chiasm split (route query/mid/val to right)

fix: restore lb_loss aux + robust logging; update GradScaler API

arch-v2: asymmetry (local vs global), competitive callosum, optic-chiasm bias; default stem=0

moe: add router warmup (round-robin expert) to reduce seed variance

moe: add load-balance loss + training hook (reduce seed variance)

ui: fix one-line summary (Δ deltas) and repair syntax

tools: add seed sweep runner

ui: add Δ deltas to one-line summary

task: add twohop_bind + multitask trainer + config

task: bind_query braid encoding + stable seed

ui: pretty-print eval metrics (rich table + summary)

head: make models query-aware (last-token readout) + trainer update

task: add braid + ood_long eval splits; batch eval

train: add AMP+GradScaler and grad clipping (fix NaNs)

poc: add bind_query task + english render + side-by-side trainer

probe: unify CLS objective + add compute-matching helper

build: add pyproject for editable install (src layout)

probe: add minimal dense + structured models and VRAM probe script

docs: freeze PoC skeleton (arch/tasks/compute/protocol) + artifact policy

docs: add North Star and early-kill criteria

init: scaffold bilateral hierarchical MoE PoC repo