Repository navigation
Using Codex to check what changed in published evidence #62494
Replies: 3 comments
|
Hii @ghostlyd, This looks like a useful approach for maintaining reliable research workflows with Codex. I particularly like the idea of tracking published evidence through version checkpoints instead of repeatedly comparing everything manually. A few things I found interesting:
Overall, this is an interesting use case for Codex and MCP. Looking forward to seeing how the two-check pilot performs and whether it leads to automated watchlists and alerts. Thanks for sharing! |
|
I'd test an unavailable source and an unrecognized checkpoint separately. In the first case, the comparison should leave that source's saved version alone and mark its evidence as unchecked. In the second, I'd want a fresh baseline that is clearly labelled so a first-seen item doesn't look like a new correction. One useful pilot fixture might be a sample decision note that cites one public edition. If the next check reports a correction, can Codex quote what changed, link both editions where available, and identify the sentence in the note that needs review? I work on Zortex and can help try the note side with one sample Google Doc and a read-only Drive connection. I haven't tested the two connectors together. |
|
Thanks, @TsangKwunHei. Let’s try it with one synthetic Google Doc and
read-only access to that doc only. No real documents or broader Drive
access needed.
I’d start with a two-sentence decision note that cites a test limit of
10, then correct the source to 12. The useful result is to quote the
correction, link both editions where available, and flag only the
sentence that relied on 10. We can test the two failure cases
separately: keep the saved version and mark the evidence unchecked if
the source is unavailable; label a fresh baseline if the checkpoint
isn’t recognized.
Can Zortex scope read-only access to a single sample doc for this test?
…On Fri, 20 Nov 2026 13:17:33 -0800, TsangKwunHei ***@***.***> wrote:
I'd test an unavailable source and an unrecognized checkpoint separately. In the first case, the comparison should leave that source's saved version alone and mark its evidence as unchecked. In the second, I'd want a fresh baseline that is clearly labelled so a first-seen item doesn't look like a new correction.
One useful pilot fixture might be a sample decision note that cites one public edition. If the next check reports a correction, can Codex quote what changed, link both editions where available, and identify the sentence in the note that needs review? I work on Zortex and can help try the note side with one sample Google Doc and a read-only Drive connection. I haven't tested the two connectors together.
--
Reply to this email directly or view it on GitHub:
#62494 (comment)
You are receiving this because you were mentioned.
Message ID: ***@***.***>
|
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
We built LrrK Agent Desk with Codex for research that needs to track what changed in published evidence. Here’s a small workflow to try with the public MCP server:
lrrk_check_updateswith an empty argument object and retain the published version keys.knownVersions. A recognized checkpoint returns additions, corrections, and withdrawals. Keep the earlier key for a source if it is unavailable.We’re running a two-check pilot with 12 seats in each track. Both checks open at 9 AM Pacific on November 24 and December 1.
After each check, tell us whether you used a cited source in a note, review, or decision. An unchanged week offers a current reference, and a “no use” report is helpful too. I’d also welcome feedback here on how your Codex workflow handles an unavailable source or an unrecognized checkpoint.
All reactions