docs: tighten rediffuse second-asset control plane#317
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request updates the research documentation to reflect a "Resting State," reclassifying the ReDiffuse DDPM on CIFAR-100 path as unreleased and condensing the task queue to summarize active candidates like H2 output-cloud and STL-10 scouts. Feedback focused on improving documentation consistency and readability, specifically regarding the placement of audit dates, the use of backticks for numeric values, standardized project naming, and improved punctuation.
| 阻断点不是 split，而是缺少公开第三方 checkpoint / score packet。2026-05-25 | ||
| STL-10 bounded scout 已证明同一路径可以执行并产出 `256 / 256` score packet， |
There was a problem hiding this comment.
The date 2026-05-25 is currently appended to the end of the first sentence, which makes the transition to the next sentence describing the scout results on that date feel disjointed. Moving the date to the start of the next sentence would improve the flow and clarity of the audit trail.
| 阻断点不是 split，而是缺少公开第三方 checkpoint / score packet。2026-05-25 | |
| STL-10 bounded scout 已证明同一路径可以执行并产出 `256 / 256` score packet， | |
| 阻断点不是 split，而是缺少公开第三方 checkpoint / score packet。 | |
| 2026-05-25 STL-10 bounded scout 已证明同一路径可以执行并产出 `256 / 256` score packet， |
| | Next GPU candidate | none selected | | ||
| | CPU sidecar | none selected after Lane A metadata triage sync | | ||
| | Gray-box status | PIA remains admitted; FMIA, Rectified Flow, SimA, VAE2Diffusion, DME, and FreMIA remain watch-plus or paper/stub-only; Fashion-MNIST SimA score-norm and score-Jacobian sensitivity remain weak | | ||
| | CPU sidecar | none selected after H2 output-cloud order-control seed-stability and ReDiffuse STL-10 weak metric verdicts | |
There was a problem hiding this comment.
The phrase "H2 output-cloud order-control seed-stability" is missing punctuation or a conjunction between the two types of scouts, making it difficult to parse. Adding a comma and "and" would align it with the phrasing used in line 12 and improve readability.
| | CPU sidecar | none selected after H2 output-cloud order-control seed-stability and ReDiffuse STL-10 weak metric verdicts | | |
| | CPU sidecar | none selected after H2 output-cloud order-control, seed-stability, and ReDiffuse STL-10 weak metric verdicts | |
| ### 2026-05-25 Research Resting State | ||
|
|
||
| - `mode`: post-H2 output-cloud order-control / post-ReDiffuse STL-10 weak scout / post-feature-packet consumer verdict. | ||
| - `status`: H2 output-cloud geometry is the latest strong Research-side candidate, with the existing `512 / 512` response-cache review and two `256 / 256` shared-position order-control scouts retaining strong signal while label-shuffle stays random-level. It is not admitted because it is still a single H2 response-cache geometry candidate, not a second public asset or Platform/Runtime contract. |
There was a problem hiding this comment.
The numeric ranges 512 / 512 and 256 / 256 are used without backticks, which is inconsistent with the updated formatting in docs/evidence/second-asset-contract.md (e.g., lines 266, 271). Standardizing the use of backticks for packet sizes and counts across all documentation files improves clarity and maintainability.
| - `status`: H2 output-cloud geometry is the latest strong Research-side candidate, with the existing `512 / 512` response-cache review and two `256 / 256` shared-position order-control scouts retaining strong signal while label-shuffle stays random-level. It is not admitted because it is still a single H2 response-cache geometry candidate, not a second public asset or Platform/Runtime contract. | |
| - status: H2 output-cloud geometry is the latest strong Research-side candidate, with the existing `512 / 512` response-cache review and two `256 / 256` shared-position order-control scouts retaining strong signal while label-shuffle stays random-level. It is not admitted because it is still a single H2 response-cache geometry candidate, not a second public asset or Platform/Runtime contract. |
|
|
||
| - `mode`: post-H2 output-cloud order-control / post-ReDiffuse STL-10 weak scout / post-feature-packet consumer verdict. | ||
| - `status`: H2 output-cloud geometry is the latest strong Research-side candidate, with the existing `512 / 512` response-cache review and two `256 / 256` shared-position order-control scouts retaining strong signal while label-shuffle stays random-level. It is not admitted because it is still a single H2 response-cache geometry candidate, not a second public asset or Platform/Runtime contract. | ||
| - `status`: ReDiffuse STL-10 is closed by default. The official split and local CUDA path are executable, but the bounded `300`-step target produced random-level fixed-timestep denoising-loss metrics, and the same checkpoint/split also failed a SimA-style denoiser-output score-norm scorer. |
There was a problem hiding this comment.
The value 300-step should use backticks for the numeric part to maintain consistency with the formatting of other counts and ranges in the document and the associated contract files.
| - `status`: ReDiffuse STL-10 is closed by default. The official split and local CUDA path are executable, but the bounded `300`-step target produced random-level fixed-timestep denoising-loss metrics, and the same checkpoint/split also failed a SimA-style denoiser-output score-norm scorer. | |
| - status: ReDiffuse STL-10 is closed by default. The official split and local CUDA path are executable, but the bounded `300`-step target produced random-level fixed-timestep denoising-loss metrics, and the same checkpoint/split also failed a SimA-style denoiser-output score-norm scorer. |
| - `stop rule`: if a candidate cannot pass target identity, exact split, query/response or score coverage, provenance, and non-adjacent mechanism gates, stop rather than writing another scope/audit/reselection chain. | ||
| - `GPU cap`: none selected. | ||
| - `CPU sidecar`: none selected. | ||
| - `integration`: no schema change; admitted five-row consumer set remains intact; H2 output-cloud, H2 simple-distance, ReDiffuse, CLiD, Tracing Roots, SecMI stat/NNS, and MT-MIA remain outside Platform/Runtime admitted rows. |
There was a problem hiding this comment.
The reference to "Tracing Roots" is inconsistent with the full name "Tracing the Roots" used in line 16 and elsewhere in the document. Using the full name ensures consistency across the research task queue.
| - `integration`: no schema change; admitted five-row consumer set remains intact; H2 output-cloud, H2 simple-distance, ReDiffuse, CLiD, Tracing Roots, SecMI stat/NNS, and MT-MIA remain outside Platform/Runtime admitted rows. | |
| - integration: no schema change; admitted five-row consumer set remains intact; H2 output-cloud, H2 simple-distance, ReDiffuse, CLiD, Tracing the Roots, SecMI stat/NNS, and MT-MIA remain outside Platform/Runtime admitted rows. |
1 participant
Summary
Verification