Advisory Brief
Advisory Brief asks a model for a plain-language brief where each claim cites a quote (or is marked unknown), then checks in code that each quote really appears in the source. This page shows one real run, claim by claim.
The run
Recorded Real run of the tool's generator and verifier (gemini, gemini-3.1-flash-lite), recorded once · 2026-10-06 · commit 4271452
The source text
Recordedstellar/stellar-core release v29.0.0, fetched from the GitHub Releases API when the run was made. This is exactly what the tool was given.
## What's Changed
* Improve DEX offer crossing accuracy. by @dmkozh
* P29 bump by @sisuresh
* Don't count pool hops against limit by @sisuresh
* Bump env to internal v29 by @sisuresh
* Remove hash-of-hash optimization. by @dmkozh
* Drop over-limit messages early by @bboston7
* Outbound queue load shed fix by @marta-lokhova
* Don't read messages during auth handshake by @SirTyson
* Lower max message size in protocol 29 by @bboston7
* Point release/v29.0.0-internal to public rs-soroban-env by @anupsdf
**Full Changelog**: https://github.com/stellar/stellar-core/compare/v28.0.1...v29.0.0Every claim, and whether its quote was found
GeneratedThe "quote found" column was recomputed when this page was built, using the repository's own normalizeForMatch (src/brief/grounding.ts at 9dba194) against the recorded source and the model's raw output. 7 of 11 claims have a quote found in the source (4 marked unknown by the model). The tool's own recorded count was 11 of 11, with 0 removed. The difference is the 4 claims marked unknown: at the recorded commit the tool counted a claim with no quote as verified. The repository now counts them separately, and the recount above is the accurate one for this recording.
What happened
| Claim | Quote it cites | Quote found? | Flags added by this site |
|---|---|---|---|
| The release notes document updates for Stellar-core version 29.0.0, including protocol-level changes. | v29.0.0-internal | Yes | very short quote |
| Several technical improvements were made, such as optimizing DEX offer crossing and adjusting message size limits. | Improve DEX offer crossing accuracy. | Yes | |
| Specific fixes related to network capacity and message handling were implemented, including outbound queue load shedding. | Outbound queue load shed fix | Yes |
Urgency reason
| Claim | Quote it cites | Quote found? | Flags added by this site |
|---|---|---|---|
| marked unknown by the model | none | n/a |
Are you affected? (explanations)
| Claim | Quote it cites | Quote found? | Flags added by this site |
|---|---|---|---|
| wallet: UNCLEAR: marked unknown by the model | none | n/a | |
| anchor: UNCLEAR: marked unknown by the model | none | n/a | |
| fintech: UNCLEAR: marked unknown by the model | none | n/a | |
| exchange: UNCLEAR: While DEX offer crossing accuracy is mentioned, it is unclear if this mandates action for exchanges. | Improve DEX offer crossing accuracy. | Yes |
What to tell your team
| Claim | Quote it cites | Quote found? | Flags added by this site |
|---|---|---|---|
| A new version of stellar-core (v29.0.0) has been released with various internal technical updates. | v29.0.0-internal | Yes | very short quote |
| Engineering should review the changelog to understand the specific impacts of protocol 29 changes. | Lower max message size in protocol 29 | Yes | |
| Evaluate the need for an infrastructure upgrade by comparing the new version against current deployment versions. | https://github.com/stellar/stellar-core/compare/v28.0.1...v29.0.0 | Yes | very short quote, quote is a link |
The flags are a heuristic added by this site to point at weak evidence (a quote of three words or fewer, or a bare link). They are not part of the tool.
What the quote check does not tell you
- A found quote proves the words exist in the source, not that they support the claim next to them. Read the claims above against their quotes. In this recording, 3 of 11 claims cite a quote flagged as short or a bare link, which is weak evidence for the sentence it sits beside.
- These are not checked against the source at all, and are the model's judgment: the urgency level (MONITOR here), the per-audience yes/no/unclear flags, and the list of things the advisory doesn't say.
- One recording is one run. Another run on the same text can word things differently.
The model's list of what it does not know, as recorded:
- Whether this release contains mandatory security fixes.
- If this release introduces breaking changes for existing API integrations.
- The official timeline for network-wide adoption of version 29.0.0.