Mission · 1849 worldbuild · solo session

Mission room

4 threads running in parallel. Two agents, one provider model, one local model — comparing how each handles the same query about Sierra mortality figures.

Active threads

2 running · 2 paused

Agents in mission

br-native · scoped

Models

2 provider · 1 local

Memory candidates

review pending

Handoffs queued

to RoadBook · RoadCode · RoadWire

Missions / 1849 worldbuild · solo

1849 worldbuild · solo · 01 Jun

Parallel threads · 4same mission · different intelligences

LucidiaAGENT · ORCHESTRATOR RUNNING

Holding the mission state. Tracking 4 threads, 3 handoffs, 7 memory candidates. Suggests Cordelia reply on Bancroft is the strongest for citation.

TASK · orchestrate · score · memory triage

TURNS · 14·SCOPE · this mission

CordeliaAGENT · RESEARCH RUNNING

"Comparing Parkman 1849 against Stewart 1936 — Parkman has the narrower confidence band (±3%) on November mortality. Recommend using Parkman as primary, Stewart as secondary."

TASK · source confidence on 1849 figures

SOURCES · 4 saved·VIA · ROADVIEW

Claude 4.7PROVIDER · ANTHROPIC IDLE

"For a marching-band-style RoadWorld scenario, you'd want to model the 4 Bancroft diaries as event streams rather than static reference. Want me to sketch the data shape?"

TASK · scenario shape · roadworld

TURNS · 6·OUT · 2.1 KB · NO TRAIN

GPT-4oPROVIDER · OPENAI IDLE

"Three diaries give you triangulation. Suggest weighting by source proximity — Parkman 0.5, Reed 0.3, the two Donner-adjacent 0.1 each. Confidence intervals naturally follow."

TASK · weighting scheme · 4 sources

TURNS · 4·OUT · 1.8 KB · NO TRAIN

Qwen 3 · 32BLOCAL · MLX · M2 MAX PAUSED

"My take is narrower — without the secondary literature I'd lean on Parkman alone and treat Stewart's range as the upper bound. I lack the Bancroft holdings beyond what was attached."

TASK · offline cross-check · narrow reading

TURNS · 2·LOCAL · NO NETWORK

RoadieAGENT · WRITER RUNNING

Drafting the Nov–Feb timeline against Cordelia's saved primaries. Needs WRITE_REPO grant on roadworld/1849/timeline.md — awaiting your approval in CarKeys.

TASK · draft timeline.md · cite as you go

GRANT · PENDING·VIA · CARKEYS

Compare · Claude vs GPT-4o vs Qwen Localsame prompt · 3 responses

PROMPT / DIMENSION

CLAUDE 4.7

GPT-4O

QWEN LOCAL

Weights the 4 sources

By proximity + corroboration. Suggests event-stream framing.

Specific: 0.5 / 0.3 / 0.1 / 0.1. Concrete numbers.

Won't weight — uses Parkman alone. Lacks Bancroft access.

Confidence interval

Implied via streams, no number given.

Computes from weights: ±5.4%

Uses Stewart's ±8% as upper bound.

Output length

2.1 KB

1.8 KB

920 B · offline

Cost · this turn

14 USAGE units

9 USAGE units

0 USAGE · LOCAL

Best for

Framing / structure

Numeric weighting

Offline conservative reading

Handoffs queued · 3awaiting your approval before send

RoadBook · 4 sources + citations

CORDELIA → ROADBOOK · BANCROFT PRIMARIES

14:30

RoadCode · timeline.md draft

ROADIE → ROADCODE · NEEDS WRITE_REPO

14:19

RoadWire · mission summary

LUCIDIA → ROADWIRE · END-OF-SESSION

EOD