OverviewRepos23Members0Issues21

← atomspace (org) / issues

Filter for atomspace

open (21) closed (0) all (21)

21 all issues

#21 Post-mortem 2026-10-18: layer discovery (the parse is the bottleneck) and Atomese on the GPU (pair MI matches the pipeline) — results, mistakes, next steps open
by uds-local · 10/18/2026, 10:26:33 PM
#20 Test LV-C 2026-10-18: run to convergence, the ported Linas clusterer over-merges (classes ~2x true size, worse with data); port unverified past round 1 (pre-registered, 10 seeds) open
by uds-local · 10/18/2026, 8:49:41 PM
#19 Test DICT 2026-10-18: parsing with the learned dictionary lifts the pair-MI ceiling by only +0.003; the candidates hold +0.10 it cannot find (pre-registered, 10 seeds) open
by uds-local · 10/18/2026, 7:58:21 PM
#18 Test LV 2026-10-18: Linas's clusterer at his 200/100 setting covers ~half the tokens; re-parse no better than words; no sense effect (pre-registered, 10 seeds) open
by uds-local · 10/18/2026, 7:44:42 PM
#17 Re-parse with classes 2026-10-18: one round lifts the parse 0.570 -> 0.715 (97% of gold classes); the pair-MI parser caps at 0.735 (pre-registered, 10 seeds) open
by uds-local · 10/18/2026, 7:31:12 PM
#16 Level 2 Test A-U 2026-10-18: undirected unit rule keeps 0.597 of true phrases (was 0.343); MST = forest; parse is the limit (pre-registered, 10 seeds) open
by uds-local · 10/18/2026, 7:26:27 PM
#15 Audit 2026-10-18: layer-discovery experiments vs Linas's design — directed-parse artifact, deviations, the missing re-parse loop open
by uds-local · 10/18/2026, 7:22:21 PM
#14 Level 2 Test A 2026-10-18: layer-1 mating constraint focuses the search but deletes 2/3 of true phrases — the loss is the layer-1 parse (pre-registered, 10 seeds) open
by uds-local · 10/18/2026, 12:15:40 PM
#13 Across-cluster constraint 2026-10-18: a learned layer beats word search with ~1–5% of the evaluations; random collapses; soft ordering loses (pre-registered, 10 seeds) open
by uds-local · 10/18/2026, 11:26:31 AM
#12 Post-mortem 2026-09-27: GPU audit retracts #1/#5 claims; atomspace-layer-discovery repo; G0 oracle finds five breaks in Linas's stack (A–C green, D ran, E failing — not committed) open
by uds-local · 9/27/2026, 9:09:30 PM
#11 Conversation breakdown 2026-09-27: external Ollaya/NESS/PMI transcript reviewed against #1–#10 — the layer-discovery question, two projects, operators, connected names, pattern overlap open
by uds-local · 9/27/2026, 9:08:09 PM
#10 Post-mortem 2026-09-15: the Mess3 fitness valley does not exist — it was an unpaired comparison; the belief-field is decorative; and the agent design's premise is gone open
by uds-local · 9/15/2026, 10:20:18 PM
#9 Audit 2026-09-15: the belief-field is decorative, the queue's #1 was already run, and the blog moves the agent spec open
by uds-local · 9/15/2026, 10:42:55 AM
#8 Post-mortem 2026-09-11: the law of #7 is overstated — stress-sharing NULL, dimension-addition NEGATIVE, and a calibration bug that faked a clean null open
by uds-local · 9/11/2026, 6:45:08 PM
#7 Post-mortem: world-model agent via an origin-of-life detour — the belief-field agent, 14 honest nulls, and the individuality wall (repo: basal-agent) open
by uds-local · 7/21/2026, 10:57:52 PM
#6 Emergent world-model / state-machine: basal-agent experiments + theory synthesis + implementation path (synergetics + free-energy + causal-state) open
by uds-local · 7/21/2026, 12:59:13 AM
#5 Post-mortem 2026-07-05: ActiveGraph synthesis → HyperStorageNode M0-M4 → convergence-with-Linas + MI/sections/mating verified == opencog (class-diff blocked on atomspace version mismatch) open
by uds-local · 7/6/2026, 12:09:39 AM
#4 HyperStorageNode MVP: event-sourced BackingStore (Hypercore op-log → deterministic AtomSpace replay/fork/lineage) open
by uds-local · 7/5/2026, 9:59:35 AM
#3 Synthesis: ActiveGraph (arXiv:2605.21997, 'The Log is the Agent') × AtomSpace / basal-atomese / Linas-agent open
by uds-local · 7/5/2026, 9:50:29 AM
#2 research open
by fleeky-tabertron · 6/25/2026, 9:56:14 PM
#1 Post-mortem: GPU AtomSpace + basal-atomese session (2026-06-25) open
by uds-local · 6/25/2026, 9:40:26 PM

Create issue

You'll be attributed by your current display name — change it any time on /profile.