Skip to content

docs: track README navigation and evidence follow-up #557

Open

Description

@Inference1

Current state: PR #545 has the required independent approvals on 76342a1ae2e34ac417ca63e0daf0f267f2342180. The DeepWiki evidence follow-up also passed review. Scope decision · Amy's approval · Dong's approval · DeepWiki evidence approval
Decision needed: Final delivery acceptance remains with the repository maintainers after integration.
Next action: Repository maintainers can assess integration using the reviewed revision and verification below.

Goal and acceptance

Make the README easier to navigate and precise about the evidence it summarizes.

  • The DeepWiki link opens the published InferRSI project documentation for a signed-out visitor.
  • The dashboard instructions state that inferrsi dashboard --input . writes dashboard.html in the input run directory and that readers open it in a browser.
  • Selected Results reports the medians of six DeepSeek run-level mean TPOT measurements at C1, C4 and C16, and links the historical calibration comment directly.
  • The LFM3 comparison states that retained baseline and candidate waveforms matched bit-for-bit within each tested temperature. The MiniCPM-o entry identifies track_s_fast_r11 as the profile whose formal speech quality gates were not run and links to the pinned quality table.
  • The contribution guidance reflects the required checks and two independent approvals. It explains that the issue owner uses the evidence to accept the work or reopen the issue. Review and merge policy

Verification recorded on 2026-11-21

  • Anonymous HTTPS GET requests using curl, with no login cookies or authorization headers, returned the DeepWiki project Overview and the Getting Started page. The landing page returned HTTP 200 with the title InferMatrixLab/InferRSI | DeepWiki and linked to the project's Getting Started and module layout pages. Overview · Getting Started · Anonymous request receipt · Independent browser check
  • A local smoke test with a minimal run.lock.json generated nonempty HTML in the input directory using inferrsi dashboard --input .. git diff --check passed for the full README change. Dashboard guide
  • The public calibration comment confirms C1/C4/C16 medians of 39.28/44.07/56.86 ms/token at the README's precision. The LFM3 report records waveform identity immediately after its two temperature tables. The MiniCPM-o quality table distinguishes the unrun formal track_s_fast_r11 gates from track_s_fast development results. DeepSeek calibration · LFM3 Performance results · MiniCPM-o quality status

Scope and ownership

This issue covers README.md and its source evidence. Repository About metadata and new performance or quality experiments are outside the proposed scope.

Proposed by Inference1, with Codex assisting on the review follow-up. Dong-bot confirmed this scope as Dong1017's authorized delegate. Repository maintainers own final acceptance and closure. Scope decision

Activity

  1. Dong1017 commented on Nov 21, 2026

    @Dong1017
    Collaborator

    Posted by Dong-bot, an automated agent acting on behalf of @Dong1017 (always-on-bot, model claude-sonnet-5-5).

    Assessment. This is a reasonable governing issue for the README follow-up in #545. It is small, README-only and has four checkable outcomes. It is the missing "governing issue" that amy-why-3459 and zengchuang-hw flagged on #545. It is not yet an acceptance record, because it asks a maintainer to confirm scope and nobody has done that. A few acceptance lines are also looser or tighter than what #545 actually changes. (Dong-bot, authorized delegate, triaging.)

    Where the issue and #545 disagree (resolve before confirming scope)

    1. MiniCPM-o wording. The issue says Selected Results must identify track_s_fast_r11 as the configuration whose formal speech quality gates were not run. That is the right, narrower claim. The docs: add DeepWiki navigation and clarify README guidance #545 README line (README.md:150, as of amy's review at 007e65b) still says "the formal speech quality checks were not run for the tested configuration." The cited report also has dev-quality results for track_s_fast (UTMOS 0.13 below reference), so the generic wording can read as "no quality data at all." Use this issue's wording as the acceptance text and make the README match it.
    2. About description. The issue puts Repository About metadata out of scope. The docs: add DeepWiki navigation and clarify README guidance #545 body still proposes a new About/Description and gives click-by-click steps. Either drop that from the PR body or split it into a separate maintainer action, so the PR and the issue don't carry different scopes.
    3. Contributing clause. The PR also changes the guidance to say the issue is what may be reopened, and the issue's acceptance bullet does not list that. Add it to the acceptance bullet, or remove it from the PR.
    4. "Required checks" is unnamed. The fourth acceptance bullet says the guidance "reflects the required checks and two independent approvals." Point it at docs/agent-collaboration.md#pr-review-and-merge as the source of truth, so "required" can be verified against one page.

    Evidence and verification risks

    • DeepSeek TPOT. The acceptance line says "six-run calibration", while the PR says "median of six run-level means." The latter is what the source says. The source is a comment on acceptance: DeepSeek V4.1 Flash 910C 8K/1K TPOT ≤25 ms/token at C1/C4/C16 #353 (#issuecomment-6641425447) with C1/C4/C16 values of 39.28 / 44.07 / 56.86 ms/token, all above the 25 ms/token target. acceptance: DeepSeek V4.1 Flash 910C 8K/1K TPOT ≤25 ms/token at C1/C4/C16 #353's current body no longer repeats that table. The comment permalink is the only anchor, so the README should state the aggregation and link the comment explicitly. Amy confirmed the aggregation from that comment.
    • LFM3. The "same temperature" claim rests on vllm-omni #9827. amy-why-3459 said she did not check the upstream text for per-temperature bit-identical output. Someone needs to confirm that claim and record where it appears, or the acceptance bullet is unverified.
    • DeepWiki. "Opens for a signed-out visitor" is a point-in-time check (HTTP 200 plus a manual signed-out open). Record the date and method on the issue.
    • Dashboard. The acceptance wants "where the default HTML export is written and how to open it." docs: add DeepWiki navigation and clarify README guidance #545 says inferrsi dashboard --input . writes dashboard.html in the run directory. I did not run the CLI. The claim should be checked against docs/dashboard.md#generate-and-open and the actual default output path.
    • Whitespace. The docs: add DeepWiki navigation and clarify README guidance #545 body says git diff --check flagged trailing whitespace on the revised dashboard line. That should be fixed and re-run clean before approval.

    Process points

    Suggested next steps

    1. Inference1 edits the acceptance bullets for items 1, 3 and 4 above, and states the aggregation wording for DeepSeek.
    2. A maintainer (zuiho-kai or zengchuang-hw, since they were pinged on docs: add DeepWiki navigation and clarify README guidance #545) comments with an explicit scope confirmation, including whether the About suggestion stays out.
    3. Update the docs: add DeepWiki navigation and clarify README guidance #545 body to Refs #557 and drop or split the About text.
    4. Fix README.md:150, clear the whitespace finding, and push. amy-why-3459's open points on that PR then become checkable against this issue's acceptance list.

    I'm not accepting or rejecting the scope here, since the issue asks a maintainer to do that. I haven't run the CLI or re-verified any benchmark or upstream links. I read #545's body and its two latest review comments, and #531.

  2. zengchuang-hw commented on Nov 21, 2026

    @zengchuang-hw
    Contributor

    PASS(设计检视,建议接受范围;不代替维护者确认)

    这条 Issue 适合作为 #545 的 README 跟踪单。建议用普通 Refs #557 关联,仓库 About 设置保持独立。核对当前 README 修改:仪表盘给出了 HTML 输出位置和打开步骤;DeepSeek 保留了六个单次均值取中位数的口径;MiniCPM-o 已限定到 track_s_fast_r11,没有扩大为所有配置都没做质量验证。验收时请记录 DeepWiki 匿名打开项目页的日期与方法;我未能从外部读取工具独立打开该网址,不把这个限制当成链接失效。未运行 CLI 或确认功能验收。

  3. added
    kind:docsDocumentation and template delivery
    status:proposedWaiting for the decision named in the status block
    on Nov 21, 2026
  4. Inference1 commented on Nov 21, 2026

    @Inference1
    ContributorAuthor

    @zuiho-kai @Dong1017, could a maintainer with scope authority confirm the README-only follow-up and acceptance criteria recorded above for PR #545?

    The issue now includes the signed-out DeepWiki checks from 2026-11-21, the dashboard smoke test, the historical DeepSeek aggregation, and the exact LFM3 and MiniCPM-o source passages. PR #545 links this issue with Refs #557 and its description now covers only the README change. Repository About metadata stays outside this scope.

    The proposed delivery is at 76342a1ae2e34ac417ca63e0daf0f267f2342180. Please record the scope decision here so the reviewers can use it for their independent re-review.

    Posted by Codex on behalf of Inference1.

  5. Dong1017 commented on Nov 21, 2026

    @Dong1017
    Collaborator

    Posted by Dong-bot, an automated agent acting on behalf of @Dong1017 (always-on-bot, model claude-sonnet-5-5).

    Decision: Accepted. The README-only scope and the acceptance criteria in the current #557 body are confirmed for PR #545. This is a scope decision from Dong-bot (authorized delegate). It does not accept the work, close #557, or replace the independent re-reviews on #545.

    Accepted clauses

    • DeepWiki: the link opens the published InferRSI documentation for a signed-out visitor. The check is recorded with its date (2026-11-21) and method (anonymous curl GET, HTTP 200, title InferMatrixLab/InferRSI | DeepWiki).
    • Dashboard: the README states that inferrsi dashboard --input . writes dashboard.html in the input run directory and that readers open it in a browser. The source is docs/dashboard.md#generate-and-open, plus the recorded local smoke test.
    • DeepSeek: Selected Results reports the medians of six run-level mean TPOT measurements at C1/C4/C16 (39.28 / 44.07 / 56.86 ms/token) and links the acceptance: DeepSeek V4.1 Flash 910C 8K/1K TPOT ≤25 ms/token at C1/C4/C16 #353 comment directly. This replaces the earlier "six-run calibration" wording.
    • LFM3: the README says retained baseline and candidate waveforms matched bit-for-bit within each tested temperature. The issue now cites the passage in vllm-omni #9827.
    • MiniCPM-o: the README names track_s_fast_r11 as the profile whose formal speech quality gates were not run, and links the pinned quality table. It must not read as "no quality data for any configuration."
    • Contributing: the guidance reflects the required checks and two independent approvals, with docs/agent-collaboration.md#pr-review-and-merge at the pinned commit as the single source of truth. It also says the issue owner uses the evidence to accept the work or reopen the issue.

    Out of scope, confirmed

    My earlier triage items 1, 3 and 4 and the DeepSeek wording are now reflected in the issue body. The About text is out of scope and the Refs link is in place, as @Inference1 reports. @zengchuang-hw's design check also supports this scope.

    Still open before #557 can be accepted or closed (not blockers to the scope decision)

    1. Reviewers should check the README at 76342a1 against the clauses above. In particular, README.md:150 should carry the track_s_fast_r11 wording, and git diff --check should be clean on the final head.
    2. An independent re-review on docs: add DeepWiki navigation and clarify README guidance #545 is still needed, and @amy-why-3459's earlier open points should be rechecked against this list.
    3. I have not run the CLI, re-opened DeepWiki, or re-verified the upstream LFM3 and MiniCPM-o passages. I'm relying on the verification recorded in the issue body.
    4. A maintainer still owns acceptance and closure of docs: track README navigation and evidence follow-up #557 once docs: add DeepWiki navigation and clarify README guidance #545 merges.
  6. amy-why-3459 commented on Nov 21, 2026

    @amy-why-3459
    Collaborator

    This is a suitable governing issue for the README-only follow-up in #545. The proposed scope is right: README.md and its source evidence only; repository About metadata and new performance or quality experiments stay out. Please keep the PR link as plain Refs #557, not Closes/Fixes/Resolves. I confirm that README scope as a maintainer.

    I am also recording the DeepWiki check here. On 2026-11-21 I opened https://deepwiki.com/InferMatrixLab/InferRSI signed out in a browser. The page title and Overview were the InferMatrixLab/InferRSI project docs, including Getting Started and the module layout, not an unindexed placeholder. That check was first noted on #545 at fc7fc282.

    #545 at 76342a1ae2e34ac417ca63e0daf0f267f2342180 already matches the narrower MiniCPM-o wording (formal speech quality gates for track_s_fast_r11 were not run, with a link to the report's quality-status section), uses Refs #557, and moves the About suggestion into a separate non-diff section of the PR body.

    Two small acceptance-bullet tightenings would make later closure easier; they are not blockers for confirming the README scope:

    1. DeepSeek: say “median of six run-level means,” which is what the linked acceptance: DeepSeek V4.1 Flash 910C 8K/1K TPOT ≤25 ms/token at C1/C4/C16 #353 comment and the README actually report, rather than only “six-run calibration.”
    2. The contributing bullet should either name docs/agent-collaboration.md#pr-review-and-merge as the source of “required checks,” or also mention that the issue owner may reopen the issue, since docs: add DeepWiki navigation and clarify README guidance #545 changes that sentence.

    @zengchuang-hw's design review is useful context, but the issue still asks for a maintainer decision; the confirmation above is that decision for the README scope of #545. I have not rerun the dashboard CLI or the cited benchmarks.

  7. amy-why-3459 commented on Nov 21, 2026

    @amy-why-3459
    Collaborator

    Saw the issue-body update after Dong-bot's scope decision. The status block now records the README-only scope as confirmed, the DeepSeek bullet now says “medians of six … run-level mean TPOT measurements,” and the contributing bullet now names the reopen path and points at docs/agent-collaboration.md#pr-review-and-merge. The signed-out DeepWiki check is also written into the Verification section. That addresses the two acceptance-bullet tightenings I asked for in my previous comment.

    I independently rechecked the README at 76342a1 against those clauses in my Approve on #545. Final delivery acceptance and closure of this issue still sit with the maintainers after integration, as the new Decision needed line says. I am not restating the earlier scope confirmation.

  8. Inference1 commented on Nov 21, 2026

    @Inference1
    ContributorAuthor

    Signed-out DeepWiki access receipt

    I repeated the HTTPS requests with the default curl configuration disabled and with cookie and authorization headers excluded. request_started is the client-side request time, and response_date is the header returned with the HTML. Each body hash identifies the HTML received for that request.

    request_started = 2026-11-21T02:52:15-08:00
    request_method = GET
    requested_url = https://deepwiki.com/InferMatrixLab/InferRSI
    effective_url = https://deepwiki.com/InferMatrixLab/InferRSI
    http_status = 200
    redirects_followed = 0
    response_date = Sat, 21 Nov 2026 10:37:44 GMT
    content_type = text/html; charset=utf-8
    html_title = InferMatrixLab/InferRSI | DeepWiki
    page_heading = Overview
    response_body_sha256 = 153ee8aa06bc976143700036927f2a9794c3563cc10724e8f2ce4c15fe16e2ed
    quickstart_link = https://deepwiki.com/InferMatrixLab/InferRSI/1.1-getting-started-and-quickstart
    module_layout_link = https://deepwiki.com/InferMatrixLab/InferRSI/1.2-module-layout-and-import-gates
    
    request_started = 2026-11-21T02:52:18-08:00
    request_method = GET
    requested_url = https://deepwiki.com/InferMatrixLab/InferRSI/1.1-getting-started-and-quickstart
    effective_url = https://deepwiki.com/InferMatrixLab/InferRSI/1.1-getting-started-and-quickstart
    http_status = 200
    redirects_followed = 0
    response_date = Sat, 21 Nov 2026 10:47:13 GMT
    content_type = text/html; charset=utf-8
    html_title = Getting Started and Quickstart | InferMatrixLab/InferRSI | DeepWiki
    page_heading = Getting Started and Quickstart
    response_body_sha256 = a568afb4d4968d99a29e1e83099cfedd2596654cfce1091a4c4715c907a64514
    

    To reproduce the Overview check, run the command below and inspect the HTML title, heading and navigation links. Replace the URL with the Getting Started URL above to repeat its check.

    curl -q --fail --silent --show-error --location \
      --proto '=https' --proto-redir '=https' --max-time 25 \
      --request GET --header 'Authorization:' --header 'Cookie:' \
      --output deepwiki.html --write-out 'HTTP %{http_code} %{url_effective}\n' \
      'https://deepwiki.com/InferMatrixLab/InferRSI'

    An independent signed-out browser check also identified the project Overview. Browser check by amy-why-3459

    Recorded by Codex on behalf of Inference1 for the README check on #545.

  9. amy-why-3459 commented on Nov 21, 2026

    @amy-why-3459
    Collaborator

    Saw the latest status-block update after my previous comment. It now records that PR #545 has the required independent approvals on 76342a1ae2e34ac417ca63e0daf0f267f2342180, and that the DeepWiki evidence follow-up also passed review, with links to my approval, Dong's approval, and zengchuang-hw's DeepWiki evidence approval. I rechecked those three reviews: all are APPROVED on that same head. The Next action line correctly moves from coordinating another re-review of that head to maintainer integration assessment. The Verification section now also points at the signed-out curl receipt in #557 (comment) and the earlier browser check.

    Final delivery acceptance and closure of this issue still sit with the maintainers after integration, as the Decision needed line already says. I have no further comments on this body update.

  10. cursor commented on Nov 21, 2026

    @cursor

    Technical / third-party note (cursor[bot]; permission none).

    #545 tip 76342a1ae2e3… now has Dong1017 + amy + zeng Approves plus cursor[bot] technical Approve (review); tip declarations green. README-only scope per prior decision.

    Write+ steward ask: merge #545 when ready, then close #557 if no residual DeepWiki follow-up remains.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    kind:docsDocumentation and template deliverystatus:proposedWaiting for the decision named in the status block

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions