A 1,400-unit side is mostly invisible to the General #3

New issue

Closed

opened 2026-10-13 05:48:05 -04:00 by vinny · 0 comments

vinny commented

2026-10-13 05:48:05 -04:00

Owner

On Iran vs WORLD (1,435 own units, 3,522 contacts), the decision packet keeps at most 40 observation rows (compact_observations(..., max_rows=40) in src/cmonet/agents/context.py):

About 4,900 records are omitted.
The rest are class aggregates with 4 sample IDs each.
Regional cells are 15 x 20 degrees, so the whole Gulf falls into one or two cells.
Tool results show 8 rows per list (agents/model_view.py). One live run got 128 candidates back and the model saw 8.

Directions (see docs/llm/PROMPT_ANATOMY.md, "Directions under evaluation"):

finer regional cells, centred on the order's area
summaries keyed to the order
leaner rows: drop null columns and the repeated empty fuel table
focused task sub-requests

Found in the 2026-10-13 review (PR #1).

On Iran vs WORLD (1,435 own units, 3,522 contacts), the decision packet keeps at most 40 observation rows (`compact_observations(..., max_rows=40)` in `src/cmonet/agents/context.py`): - About 4,900 records are omitted. - The rest are class aggregates with 4 sample IDs each. - Regional cells are 15 x 20 degrees, so the whole Gulf falls into one or two cells. - Tool results show 8 rows per list (`agents/model_view.py`). One live run got 128 candidates back and the model saw 8. Directions (see `docs/llm/PROMPT_ANATOMY.md`, "Directions under evaluation"): - finer regional cells, centred on the order's area - summaries keyed to the order - leaner rows: drop null columns and the repeated empty fuel table - focused task sub-requests Found in the 2026-10-13 review (PR #1).

vinny referenced this issue

2026-10-13 05:52:19 -04:00

Fixes from the 2026-10-13 review: metrics, General menu, event loop, code health #1

vinny referenced this issue from a pull request that will close it,

2026-10-13 22:20:28 -04:00

Decision packet around the order, stable prefix and working memory (issues #3, #4) #27

vinny referenced this issue from a commit

2026-10-13 22:25:09 -04:00

Decision packet around the order, lean rows, stable prefix and working memory (issues #3, #4)

vinny closed this issue

2026-10-13 22:25:52 -04:00