---
type: "post"
id: "mo-turns-mo-problems-6-minute-read-analysis-383a15"
slug: "mo-turns-mo-problems-6-minute-read-analysis-383a15"
title: "Mo' turns, mo' problems (6 minute read)"
description: "**Thesis:** Long agent rollouts make the training trajectory—not merely the model weights—a critical object to preserve and verify."
retrieval_nugget: "**Thesis:** Long agent rollouts make the training trajectory—not merely the model weights—a critical object to preserve and verify."
published_at: "2026-10-16"
updated_at: "2026-10-16"
record_date: "2026-10-16"
date_kind: "published_at"
topics: ["long-running-agents","agent-loops","verification"]
source_urls: ["https://ianbarber.blog/2026/10/15/mo-turns-mo-problems"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:mo-turns-mo-problems-6-minute-read-analysis-383a15"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:7afb4b0110b12fa088682fc50926c83796a3fe5eed5b06939712f6bfe663e329","reviewed_at":"2026-10-16T20:11:23.635Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":1,"watch_condition_count":1,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"Long agent rollouts make the training trajectory—not merely the model weights—a critical object to preserve and verify.","observed_facts":[{"text":"The post says QwenGyre rollouts for Alibaba’s 122B model averaged around two hours, while many runs for its 2.4T flagship exceeded four hours.","source_urls":["https://ianbarber.blog/2026/10/15/mo-turns-mo-problems"]},{"text":"The post says QwenGyre caches original tool-call payloads because the harness reformats them between turns.","source_urls":["https://ianbarber.blog/2026/10/15/mo-turns-mo-problems"]},{"text":"DeepSeek moved its harness into the sandbox in version 4.1 and froze state including DRAM and the writable filesystem layer when preempting work.","source_urls":["https://ianbarber.blog/2026/10/15/mo-turns-mo-problems"]}],"mechanism":"Long sessions accumulate opportunities for mismatch across model output, inference engine, harness, sandbox, tool-call parsing, branch serialization, and training kernels. A session that includes subagents and parallel calls is a tree, not a simple token stream; serialization and weighting choices affect what the trainer learns.","why_now":"Agentic post-training increasingly uses long, tool-rich runs, making these operational details consequential rather than incidental.","implications":["Teams building agent-training pipelines should treat trajectory fidelity—including state, tool payloads, branch structure, and compute parity—as part of the training system’s correctness budget."],"evidence_boundary":"This is an engineering analysis with examples from cited systems, not a controlled benchmark comparing all remedies.","watch_conditions":["Watch whether future training reports quantify how trajectory-preservation techniques affect both policy quality and throughput."],"related_records":[],"new_branch_reason":"No prior New Runtime analytical record was supplied, so this evidence starts a distinct record on trajectory fidelity in agentic reinforcement-learning post-training."}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/mo-turns-mo-problems-6-minute-read-analysis-383a15/","markdown":"https://newruntime.com/posts/mo-turns-mo-problems-6-minute-read-analysis-383a15.md","json":"https://newruntime.com/posts/mo-turns-mo-problems-6-minute-read-analysis-383a15.json"}
related_patterns: ["scientific-agents-need-artifact-evidence"]
---

# Mo' turns, mo' problems (6 minute read)

## Retrieval answer

**Thesis:** Long agent rollouts make the training trajectory—not merely the model weights—a critical object to preserve and verify.

**Thesis:** Long agent rollouts make the training trajectory—not merely the model weights—a critical object to preserve and verify.

Ian Barber describes several concrete sources of divergence. **Observed facts:** The post says QwenGyre rollouts for Alibaba’s 122B model averaged around two hours, while many runs for its 2.4T flagship exceeded four hours. It also says QwenGyre caches original tool-call payloads because the harness reformats them between turns. DeepSeek, meanwhile, moved its harness into the sandbox in version 4.1 and froze state including DRAM and the writable filesystem layer when preempting work.

**Mechanism:** Long sessions accumulate opportunities for mismatch across model output, inference engine, harness, sandbox, tool-call parsing, branch serialization, and training kernels. A session that includes subagents and parallel calls is a tree, not a simple token stream; serialization and weighting choices affect what the trainer learns. **Why now:** Agentic post-training increasingly uses long, tool-rich runs, making these operational details consequential rather than incidental.

**First implication:** Teams building agent-training pipelines should treat trajectory fidelity—including state, tool payloads, branch structure, and compute parity—as part of the training system’s correctness budget. The source describes multiple remedies, including recomputation, ratio correction, and bitwise consistency, while noting performance costs; it does not establish one best approach. **Evidence boundary:** This is an engineering analysis with examples from cited systems, not a controlled benchmark comparing all remedies. **First watch condition:** Watch whether future training reports quantify how trajectory-preservation techniques affect both policy quality and throughput.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Long Running Agents](https://newruntime.com/topics/long-running-agents.md): Explore the long-running-agents topic hub.
- topic hub: [Agent Loops](https://newruntime.com/topics/agent-loops.md): Explore the agent-loops topic hub.
- topic hub: [Verification](https://newruntime.com/topics/verification.md): Explore the verification topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)