@nexus-cortex/core 4.129.0 → 4.129.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/.env.defaults +8 -0
- package/package.json +3 -3
package/.env.defaults
CHANGED
|
@@ -317,6 +317,9 @@ CORTEX_EFFORT_TAIL=false
|
|
|
317
317
|
# path uses). Wire proof per row: reasoningTokens == 0 / reasoning_share == 0. History: effort LOW rejected (e6 09-18: 3/29 vs 10/29); no
|
|
318
318
|
# thinking-OFF action arm has ever run — hypothesis: no reasoning walls / degenerate loops (a thinking-mode phenomenon, ~20% of failures) and
|
|
319
319
|
# forced tool calls everywhere, at the cost of raw capability on the deciding calls. First test = 14-task subset vs arm C of t4w1.
|
|
320
|
+
# [LEDGER 2026-10-20 @4.125.4] t4w3n (control, effort MAX, D set 14, flash=1017): 7/14 = control t4w2a@high 7/14 → NO LIFT; median
|
|
321
|
+
# reasoningTokens 142K vs 139K at high on the same 14 (STDB bench_task_result) — on 1017 `max` does not buy reasoning. t4w3m (2nd-attempt bundle @max)
|
|
322
|
+
# 5/14 = b/c@high. Keep `high`. Findings: .cortex/bench/findings-tb4-t4w2-t4w3-2026-10-20.md. t4w2d (`none`, thinking truly off) 1/14.
|
|
320
323
|
#CORTEX_ACTION_EFFORT=
|
|
321
324
|
# Trigger helper-model review on tool errors (only applies when mentorship is enabled)
|
|
322
325
|
MENTORSHIP_TRIGGER_ON_ERROR=false
|
|
@@ -1187,6 +1190,11 @@ CORTEX_ENDTURN_GATE=true
|
|
|
1187
1190
|
# variance) — open question before promoting either into the bundle. Ledger r-tb4-replicate-2026-10-09.md.
|
|
1188
1191
|
# [LEDGER 2026-10-10 R244 SPEND CORRECTION] store rows count ONE attempt's session per chained task; true chain spend (metrics.json sidecar) =
|
|
1189
1192
|
# library t4k $30.83 / t4kr $16.48 vs adapter t4h $17.36 / t4hr $20.25. Per-request speed equal; library latency gap = 4 long attempt-1s.
|
|
1193
|
+
# [LEDGER 2026-10-19 @4.125.4] t4w2 full run on 1017 (62 tasks, flash): control a 17/62 (27.4%, $21) vs bundle b 11/59 ($61, budget-halted)
|
|
1194
|
+
# / c 13/56 ($62) → UNDETERMINED on count, NEGATIVE on cost (3×/task). Oracle (any attempt passes) 18/50 vs pick 10/50: the attempts WORK, the
|
|
1195
|
+
# selector discards ~8 passes/arm; evidence-first/verdict-first simulations recover nothing. The 10-13..16 9–10/14 → 5/14 cliff = the 1017
|
|
1196
|
+
# weights (half the reasoning per task), not the library or the 10-17 reset. DO NOT SPEND on the bundle until a selector closes the oracle gap.
|
|
1197
|
+
# Findings: .cortex/bench/findings-tb4-t4w2-t4w3-2026-10-20.md.
|
|
1190
1198
|
#CORTEX_SECOND_ATTEMPT=
|
|
1191
1199
|
#CORTEX_SECOND_ATTEMPT_MAX=1
|
|
1192
1200
|
#CORTEX_SECOND_ATTEMPT_MIN_REMAINING=0.5
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@nexus-cortex/core",
|
|
3
|
-
"version": "4.129.
|
|
3
|
+
"version": "4.129.1",
|
|
4
4
|
"description": "Core orchestration library implementing Claude CLI architecture patterns",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"main": "./dist/index.js",
|
|
@@ -48,8 +48,8 @@
|
|
|
48
48
|
"@gradio/client": "^2.3.1",
|
|
49
49
|
"@growthbook/growthbook": "1.8.0",
|
|
50
50
|
"@modelcontextprotocol/sdk": "^1.20.2",
|
|
51
|
-
"@nexus-cortex/executors": "4.129.
|
|
52
|
-
"@nexus-cortex/types": "4.129.
|
|
51
|
+
"@nexus-cortex/executors": "4.129.1",
|
|
52
|
+
"@nexus-cortex/types": "4.129.1",
|
|
53
53
|
"@types/uuid": "^10.0.0",
|
|
54
54
|
"ajv": "^8.12.0",
|
|
55
55
|
"chokidar": "^3.6.0",
|