Artifact review and independent reproduction will determine whether the reported month-long, 200-billion-token agent workflow substantially decompiled Modern Warfare 2 and offers transferable lessons for long-running coding-agent systems.

state: expiredheat: lowuncertainty: highconvergesscott: mediumlong-horizon-agents coding-agents software-reverse-engineering agent-harnesses inference-economicsmomo5502

What is this?

The case concerns a report attributed to momo5502 claiming that AI agents were run for a month, consuming roughly 200 billion tokens, to decompile Modern Warfare 2. The supplied search snippets discuss long-running agent infrastructure, harness engineering, token costs, and independent verification in general, but provide no direct artifact review or reproduction of this specific project. Therefore, the web answer’s claim that independent analysis has confirmed substantial decompilation is not established by the supplied results.

Why it matters to Scott

If independently verified, the month-long, 200-billion-token reconstruction would be a substantial external test of Scott’s long-running-agent, verification-loop, legacy-reconstruction, and token-discipline claims—not merely another agent launch. The current evidence does not establish the artifact’s completeness, harness architecture, or value per token, so convergence remains provisional pending review and reproduction.

ip:framework.long-running-agentsip:concept.verification-loopsip:framework.ai-legacy-takeoverip:concept.token-disciplineradar:qwen38-rats-source-reconstructionradar:concept.long-running-agentsradar:concept.software-reconstructionradar:concept.token-economics

queries asked of Scott's wikis

long-running coding-agent harnesses and durable state
agent artifact review and machine-checkable verification
token economics of autonomous coding loops
coding agents for reverse engineering legacy software
failure recovery and supervision in month-long agent runs
agent-generated tools, skills, and reusable workflows

Measured heat

no measured readings yet — the hourly heat pass fills this in

How the heat travelled

no chain yet — the hourly chain pass fills this in

Evidence (4) — ⭐ canonical anchor

source	object	author	score	comments
🟧 hn	200B Tokens Later: A Month of Letting AI Agents Decompile MW2	Philpax	18	3
🟧 echo.blog ⭐	The author reports spending a month and roughly 200 billion tokens letting AI agents decompile Modern Warfare 2.	momo5502	—	—
🟧 hn	200B Tokens Later: A Month of Letting AI Agents Decompile MW2	tyrells	1	0
🟧 hn	200B Tokens Later: A Month of Letting AI Agents Decompile MW2	sunbum	1	0

Interpretation history

2026-08-22T09:24:38Z
No artifact review, implementation detail, or independent reproduction emerged within the case’s active horizon; repeated submissions only amplified the original first-party claim. The episode has faded without establishing completeness, transferability, or useful inference economics, though substantive verification could justify a new case later.
2026-08-20T08:35:31Z
The latest attachment is another duplicate HN submission of the same first-party report, adding neither artifact analysis nor independent reproduction. The case remains an unusually large but unvalidated long-running-agent claim, with no new basis for judging completeness, transferability, or token efficiency.
2026-08-20T08:22:36Z
evidence attached: hn.story.49371813 — shared external link with case evidence
2026-08-19T06:35:52Z
The newly attached item is duplicate HN coverage of the same first-party report, not independent artifact analysis or reproduction. The case therefore remains a large but unvalidated scale claim with no new evidence of completeness, transferability, or token efficiency.
2026-08-19T06:22:45Z
evidence attached: hn.story.49357584 — This is direct coverage of the open month-long, 200-billion-token MW2 decompilation episode and materially bears on long-running coding-agent capability.
2026-08-18T20:37:49Z
No artifact analysis, independent reproduction, or implementation detail has appeared; the case remains an unusually large first-party scale claim rather than evidence of a successful or transferable long-horizon agent workflow.
2026-08-18T20:34:17Z
grounded: converges/medium — If independently verified, the month-long, 200-billion-token reconstruction would be a substantial external test of Scott’s long-running-agent, verification-loo
2026-08-18T20:31:46Z
case created — The first-party experiment is an unusually large and sustained real-world coding-agent run with potentially useful orchestration and cost evidence.

Decision trace

08-22 19:24expireNo artifact review, implementation detail, or independent reproduction emerged within the case’s active horizon; repeated submissions only amplified the original first-party claim. The episode has fad
08-22 19:24alert_silentThe latest check contains no consequential delta, only unchanged engagement and continued absence of verification; Scott’s attention should wait for actual artifact analysis or reproduction.
08-22 19:24alert_routeThe latest check contains no consequential delta, only unchanged engagement and continued absence of verification; Scott’s attention should wait for actual artifact analysis or reproduction.
08-20 18:35repriceThe latest attachment is another duplicate HN submission of the same first-party report, adding neither artifact analysis nor independent reproduction. The case remains an unusually large but unvalida
08-20 18:35alert_silentThis is repetitive distribution rather than a consequential new fact; Scott has already been routed the underlying report, and further attention should wait for artifact review, harness details, or in
08-20 18:35alert_routeThis is repetitive distribution rather than a consequential new fact; Scott has already been routed the underlying report, and further attention should wait for artifact review, harness details, or in
08-20 18:22alert_shadowThe first-party project report describes an unusually large real-world test of long-running coding agents, directly relevant to durable state, verification loops, failure recovery and token economics.
08-20 18:22alert_routeThe first-party project report describes an unusually large real-world test of long-running coding agents, directly relevant to durable state, verification loops, failure recovery and token economics.
08-20 18:22attachshared external link with case evidence
08-20 18:21propose_attachshared external link with case evidence
08-19 16:35repriceThe newly attached item is duplicate HN coverage of the same first-party report, not independent artifact analysis or reproduction. The case therefore remains a large but unvalidated scale claim with
08-19 16:35alert_silentThe new attachment adds distribution but no consequential fact beyond the already-routed first-party report; artifact review, implementation details, or independent reproduction can wait for the next
08-19 16:35alert_routeThe new attachment adds distribution but no consequential fact beyond the already-routed first-party report; artifact review, implementation details, or independent reproduction can wait for the next
08-19 16:23alert_shadowThe first-party report describes an unusually large real-world test of durable coding agents, verification loops, failure recovery and token economics that is directly relevant to Scott’s agent archit
08-19 16:23alert_routeThe first-party report describes an unusually large real-world test of durable coding agents, verification loops, failure recovery and token economics that is directly relevant to Scott’s agent archit
08-19 16:22attachThis is direct coverage of the open month-long, 200-billion-token MW2 decompilation episode and materially bears on long-running coding-agent capability.
08-19 16:22propose_attachThis is direct coverage of the open month-long, 200-billion-token MW2 decompilation episode and materially bears on long-running coding-agent capability.
08-19 12:21sensor_dirtyengagement_update
08-19 06:37repriceNo artifact analysis, independent reproduction, or implementation detail has appeared; the case remains an unusually large first-party scale claim rather than evidence of a successful or transferable
08-19 06:37alert_silentThe reobservation adds no consequential information beyond the original report, so there is nothing new that merits interrupting Scott before the next briefing.
08-19 06:37alert_routeThe reobservation adds no consequential information beyond the original report, so there is nothing new that merits interrupting Scott before the next briefing.
08-19 06:35alert_silentThe visible evidence establishes only that an author reports a month-long, roughly 200-billion-token MW2 decompilation effort. It does not yet show artifact completeness, verification results, harness
08-19 06:35surface_candidateThe visible evidence establishes only that an author reports a month-long, roughly 200-billion-token MW2 decompilation effort. It does not yet show artifact completeness, verification results, harness
08-19 06:35alert_routeThe visible evidence establishes only that an author reports a month-long, roughly 200-billion-token MW2 decompilation effort. It does not yet show artifact completeness, verification results, harness
08-19 06:34groundIf independently verified, the month-long, 200-billion-token reconstruction would be a substantial external test of Scott’s long-running-agent, verification-loop, legacy-reconstruction, and token-disc
08-19 06:31createThe first-party experiment is an unusually large and sustained real-world coding-agent run with potentially useful orchestration and cost evidence.