Issues

• A compiled moment on one CPU core: read, write, choice and update around the solve

  muellerberndt/cadence#282 · Filter by author muellerberndt opened on Nov 9, 2026 · v0.96.0

• Simplification for 0.96.0: the candidates 0.95.0 kept

  muellerberndt/cadence#281 · Filter by author muellerberndt opened on Nov 9, 2026 · v0.96.0
  3

• The default composition's falls: locate and repair what the recall memory gives up

  muellerberndt/cadence#280 · Filter by author muellerberndt opened on Nov 9, 2026 · v0.96.0

• The whole lifecycle on the device: observation, settle, act, reward, learn

  muellerberndt/cadence#274 · Filter by author muellerberndt opened on Nov 7, 2026 · v0.96.0
  3

• Parity with transformer attention heads: tracker

  type:epic
  muellerberndt/cadence#272 · Filter by author muellerberndt opened on Nov 6, 2026 · Attention parity

• Exposure parity: the read's rule under more passes and rehearsal

  muellerberndt/cadence#270 · Filter by author muellerberndt opened on Nov 6, 2026 · v0.96.0
  4

• Composed attention functions: duplicates, suppression and indirect objects

  muellerberndt/cadence#269 · Filter by author muellerberndt opened on Nov 6, 2026 · v0.96.0
  2

• The crossing: a question whose fact depends on the context

  muellerberndt/cadence#267 · Filter by author muellerberndt opened on Nov 6, 2026 · v0.96.0
  3

• Large vocabulary: the read's rule on thousands of words

  muellerberndt/cadence#266 · Filter by author muellerberndt opened on Nov 6, 2026 · v0.96.0
  3

• Deep habits: repetition hardens a skill, disuse fades it, later learning does not erase it

  muellerberndt/cadence#253 · Filter by author muellerberndt opened on Nov 4, 2026 · v0.96.0
  2

• Learned codes of meaning: compute and index memory on compressed representations instead of raw observations

  area:architecturearea:learningenhancementpriority:nextresearchroadmap
  muellerberndt/cadence#211 · Filter by author muellerberndt opened on Oct 20, 2026
  12

• Epic: feasibility comparisons for continuing learning, repair and scaling

  area:qualificationfeasibilitypriority:nextroadmaptype:epic
  muellerberndt/cadence#183 · Filter by author muellerberndt opened on Oct 11, 2026
  4

• Goal: a language model that understands sequences and grammar

  area:learningarea:qualificationpriority:nextroadmaptype:epic
  muellerberndt/cadence#173 · Filter by author muellerberndt opened on Oct 9, 2026
  4

• Acquired behaviour of a continuing brain does not survive continued learning at its own actor step

  area:learningarea:qualificationroadmap
  muellerberndt/cadence#169 · Filter by author muellerberndt opened on Oct 9, 2026
  14

  

• Goal: a working Doom player

  area:qualificationpriority:nextroadmaptype:epic
  muellerberndt/cadence#118 · Filter by author muellerberndt opened on Oct 2, 2026

• Roadmap: a Doom player and a language model in one continuing brain

  roadmaptype:epic
  muellerberndt/cadence#109 · Filter by author muellerberndt opened on Oct 2, 2026
  8

• Runtime: GPU execution, batched lives, memory and latency for the two goals

  area:qualificationarea:runtimeroadmap
  muellerberndt/cadence#98 · Filter by author muellerberndt opened on Oct 2, 2026
  1