docs(pitch): honest F10 narration — recalibration, not a fake flip#90
Conversation
The live /judge cohort footage shows 'no rank change in this cohort', AUDIT EFFECT +/-0 pts, hit@13 delta=0 — so the old F10 line 'the audit changes who wins' would contradict the on-screen result (the exact overclaim the review flagged). Rewrote F10 to match the real footage: the audit recalibrates every score against the evidence, the ranking holds, delta=0, 'the audit doesn't fake a flip — it earns the score'. Turns the honest Delta=0 into a credibility point. Per-beat + continuous take updated + a code comment guarding against re-introducing the overclaim. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
Wondering what really moved? Review this PR in Change Stack to inspect semantic changes, definitions, and references. 개요
변경사항F10 내레이션 일관성 업데이트
관련 PR
예상 검토 난이도🎯 1 (간단함) | ⏱️ ~3분 축하 시
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![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 narration script in pitch/narration-script.md for the F10 segment to reflect an honest recalibration rather than claiming the audit changes who wins, ensuring the narration matches the actual video footage. The review feedback correctly identifies that these updates need to be synchronized with the corresponding slide script in pitch/deck.html to avoid discrepancies, and suggests adding missing pause markers to the full continuous take to maintain consistent pacing.
Important
The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.
| `F10` · the payoff · honest recalibration · speed 1.0 | ||
| ``` | ||
| Open Arize. Every span is there. <#0.35#> And on a whole cohort, the audit changes who wins. | ||
| Open Arize. <#0.3#> Every span is there — six hats, forty-eight evidence spans under one judgment. <#0.4#> Nothing hidden. <#0.6#> Now run the whole cohort. <#0.3#> Every score, recalibrated against the evidence. <#0.5#> Here, the ranking holds — and we show it: delta, zero. <#0.4#> The audit doesn't fake a flip. <#0.3#> It earns the score. |
There was a problem hiding this comment.
The updated narration script for F10 is not synchronized with the corresponding slide script in pitch/deck.html (Slide 11 / F10). Specifically, pitch/deck.html lines 1134–1142 still contain the old heading (<h2>The audit changes who wins.</h2>), the old voiceover text ("Open Arize. Every span is there. And on a whole cohort — <b>the audit changes who wins.</b>"), and outdated stage delivery notes. Please update pitch/deck.html to match this new honest recalibration narration to prevent discrepancies during the presentation.
| ## Full continuous take — paste as ONE (≈1,050 chars, under the 5,000 limit) · speed 0.97 | ||
| ``` | ||
| AI writes the submissions now. By the thousands. Every hackathon, every grant, drowning in them. <#0.35#> But the judging didn't change. Still a score nobody audits. <#0.6#> An AI judge is worse. Confident. Fast. Wrong. <#0.45#> With no way to check it. <#0.6#> So we flipped it. Don't only judge the work. <#0.3#> Audit the judgment itself. <#0.8#> Trace it. <#0.6#> Trust it. <#1.0#> How? <#0.3#> Every hat. Every retrieval. Every score. <#0.35#> A span in Arize AX. A hundred and four of them. One evaluation. <#0.6#> One ADK 2.0 Workflow. Six hats in parallel. Deployed as a real agent on Agent Engine. <#0.4#> Not a notebook. <#0.6#> It reads the evaluator's own rules. Synthesizes the rubric. <#0.3#> Then every hat grounds its score in retrieved evidence. <#0.6#> Six perspectives, in parallel. <#0.3#> Watch Yellow. Optimism. <#0.3#> Run hot. <#0.6#> Now watch. Yellow over-scored. <#0.3#> Against a held-out calibration prior, strongest where the evidence is thin, <#0.4#> the audit catches its own over-confidence. And pulls the score back. <#0.55#> Live. With the math shown. <#0.6#> Open Arize. Every span is there. <#0.35#> And on a whole cohort, the audit changes who wins. <#0.8#> Open source. Apache 2.0. <#0.5#> Trace it. <#0.45#> Trust it. | ||
| AI writes the submissions now. By the thousands. Every hackathon, every grant, drowning in them. <#0.35#> But the judging didn't change. Still a score nobody audits. <#0.6#> An AI judge is worse. Confident. Fast. Wrong. <#0.45#> With no way to check it. <#0.6#> So we flipped it. Don't only judge the work. <#0.3#> Audit the judgment itself. <#0.8#> Trace it. <#0.6#> Trust it. <#1.0#> How? <#0.3#> Every hat. Every retrieval. Every score. <#0.35#> A span in Arize AX. A hundred and four of them. One evaluation. <#0.6#> One ADK 2.0 Workflow. Six hats in parallel. Deployed as a real agent on Agent Engine. <#0.4#> Not a notebook. <#0.6#> It reads the evaluator's own rules. Synthesizes the rubric. <#0.3#> Then every hat grounds its score in retrieved evidence. <#0.6#> Six perspectives, in parallel. <#0.3#> Watch Yellow. Optimism. <#0.3#> Run hot. <#0.6#> Now watch. Yellow over-scored. <#0.3#> Against a held-out calibration prior, strongest where the evidence is thin, <#0.4#> the audit catches its own over-confidence. And pulls the score back. <#0.55#> Live. With the math shown. <#0.6#> Open Arize. Every span is there — six hats, forty-eight evidence spans under one judgment. Nothing hidden. <#0.5#> Now run the whole cohort. Every score, recalibrated against the evidence. <#0.4#> Here, the ranking holds — and we show it: delta, zero. The audit doesn't fake a flip. It earns the score. <#0.8#> Open source. Apache 2.0. <#0.5#> Trace it. <#0.45#> Trust it. |
There was a problem hiding this comment.
The F10 segment in the full continuous take is missing several pause markers (<#0.3#>, <#0.4#>, etc.) that are present in the per-beat F10 script (line 58). Aligning these pause markers ensures that the continuous voiceover generation maintains the same pacing and rhythm as the individual per-beat clips.
| AI writes the submissions now. By the thousands. Every hackathon, every grant, drowning in them. <#0.35#> But the judging didn't change. Still a score nobody audits. <#0.6#> An AI judge is worse. Confident. Fast. Wrong. <#0.45#> With no way to check it. <#0.6#> So we flipped it. Don't only judge the work. <#0.3#> Audit the judgment itself. <#0.8#> Trace it. <#0.6#> Trust it. <#1.0#> How? <#0.3#> Every hat. Every retrieval. Every score. <#0.35#> A span in Arize AX. A hundred and four of them. One evaluation. <#0.6#> One ADK 2.0 Workflow. Six hats in parallel. Deployed as a real agent on Agent Engine. <#0.4#> Not a notebook. <#0.6#> It reads the evaluator's own rules. Synthesizes the rubric. <#0.3#> Then every hat grounds its score in retrieved evidence. <#0.6#> Six perspectives, in parallel. <#0.3#> Watch Yellow. Optimism. <#0.3#> Run hot. <#0.6#> Now watch. Yellow over-scored. <#0.3#> Against a held-out calibration prior, strongest where the evidence is thin, <#0.4#> the audit catches its own over-confidence. And pulls the score back. <#0.55#> Live. With the math shown. <#0.6#> Open Arize. Every span is there — six hats, forty-eight evidence spans under one judgment. Nothing hidden. <#0.5#> Now run the whole cohort. Every score, recalibrated against the evidence. <#0.4#> Here, the ranking holds — and we show it: delta, zero. The audit doesn't fake a flip. It earns the score. <#0.8#> Open source. Apache 2.0. <#0.5#> Trace it. <#0.45#> Trust it. | |
| AI writes the submissions now. By the thousands. Every hackathon, every grant, drowning in them. <#0.35#> But the judging didn't change. Still a score nobody audits. <#0.6#> An AI judge is worse. Confident. Fast. Wrong. <#0.45#> With no way to check it. <#0.6#> So we flipped it. Don't only judge the work. <#0.3#> Audit the judgment itself. <#0.8#> Trace it. <#0.6#> Trust it. <#1.0#> How? <#0.3#> Every hat. Every retrieval. Every score. <#0.35#> A span in Arize AX. A hundred and four of them. One evaluation. <#0.6#> One ADK 2.0 Workflow. Six hats in parallel. Deployed as a real agent on Agent Engine. <#0.4#> Not a notebook. <#0.6#> It reads the evaluator's own rules. Synthesizes the rubric. <#0.3#> Then every hat grounds its score in retrieved evidence. <#0.6#> Six perspectives, in parallel. <#0.3#> Watch Yellow. Optimism. <#0.3#> Run hot. <#0.6#> Now watch. Yellow over-scored. <#0.3#> Against a held-out calibration prior, strongest where the evidence is thin, <#0.4#> the audit catches its own over-confidence. And pulls the score back. <#0.55#> Live. With the math shown. <#0.6#> Open Arize. <#0.3#> Every span is there — six hats, forty-eight evidence spans under one judgment. <#0.4#> Nothing hidden. <#0.6#> Now run the whole cohort. <#0.3#> Every score, recalibrated against the evidence. <#0.5#> Here, the ranking holds — and we show it: delta, zero. <#0.4#> The audit doesn't fake a flip. <#0.3#> It earns the score. <#0.8#> Open source. Apache 2.0. <#0.5#> Trace it. <#0.45#> Trust it. |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (1)
pitch/narration-script.md (1)
58-58: ⚡ Quick winUI 배지 문구와 내레이션 핵심 표현을 더 직접 정렬하는 것을 권장합니다.
현재 문구도 의미상 맞지만, 화면의 고정 카피(
no rank change in this cohort)와 내레이션의 핵심 문장을 더 가깝게 맞추면 이후 카피 변경/재녹음 시 불일치 리스크를 줄일 수 있습니다.Also applies to: 73-73
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@pitch/narration-script.md` at line 58, Adjust the narration copy to match the on-screen badge text verbatim so there's no drift between UI and voice: replace or append the current phrasing around "the ranking holds — and we show it: delta, zero." and/or "The audit doesn't fake a flip. It earns the score." with the exact badge copy "no rank change in this cohort" (or include that exact phrase immediately after the existing line) and apply the same change to the parallel line later in the script ("no rank change in this cohort" also used at the other noted instance).
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@pitch/narration-script.md`:
- Around line 57-59: The fenced code block containing "Open Arize. Every span is
there — six hats, forty-eight evidence spans under one judgment..." is missing a
language tag and triggers MD040; update the opening fence from ``` to include a
language identifier (e.g., ```text or ```plain) so the block is recognized as
plain text and the markdownlint warning is resolved.
---
Nitpick comments:
In `@pitch/narration-script.md`:
- Line 58: Adjust the narration copy to match the on-screen badge text verbatim
so there's no drift between UI and voice: replace or append the current phrasing
around "the ranking holds — and we show it: delta, zero." and/or "The audit
doesn't fake a flip. It earns the score." with the exact badge copy "no rank
change in this cohort" (or include that exact phrase immediately after the
existing line) and apply the same change to the parallel line later in the
script ("no rank change in this cohort" also used at the other noted instance).
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 45463a18-ad3b-41e5-86a6-8fdf2683d2ef
📒 Files selected for processing (1)
pitch/narration-script.md
| ``` | ||
| Open Arize. Every span is there. <#0.35#> And on a whole cohort, the audit changes who wins. | ||
| Open Arize. <#0.3#> Every span is there — six hats, forty-eight evidence spans under one judgment. <#0.4#> Nothing hidden. <#0.6#> Now run the whole cohort. <#0.3#> Every score, recalibrated against the evidence. <#0.5#> Here, the ranking holds — and we show it: delta, zero. <#0.4#> The audit doesn't fake a flip. <#0.3#> It earns the score. | ||
| ``` |
There was a problem hiding this comment.
코드 펜스 언어를 명시해 markdownlint 경고를 해소하세요.
Line 57 코드 블록에 언어 식별자가 없어 MD040 경고가 발생합니다. ```text 같은 언어 태그를 붙여 문서 린트 일관성을 맞추는 게 좋습니다.
🧰 Tools
🪛 markdownlint-cli2 (0.22.1)
[warning] 57-57: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@pitch/narration-script.md` around lines 57 - 59, The fenced code block
containing "Open Arize. Every span is there — six hats, forty-eight evidence
spans under one judgment..." is missing a language tag and triggers MD040;
update the opening fence from ``` to include a language identifier (e.g.,
```text or ```plain) so the block is recognized as plain text and the
markdownlint warning is resolved.
Source: Linters/SAST tools
1 participant
The live
/judgecohort recording shows 'no rank change in this cohort', AUDIT EFFECT ±0 pts, hit@13 Δ=0 — so the old F10 line 'the audit changes who wins' would directly contradict the on-screen footage (the exact overclaim the 3-agent review flagged as highest-risk).Rewrote F10 to match the real footage: the audit recalibrates every score against the evidence, the ranking holds, Δ=0, 'the audit doesn't fake a flip — it earns the score.' This turns the honest Δ=0 into a credibility point (stronger than a flashy fake flip).
Pairs with the cleaned
/judgeclip (loading 15s→1s) for the video's F10 segment.Summary by CodeRabbit
릴리스 노트