Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →An AI agent can return a convincing answer even when its search settings, document filters, or usage counters are wrong. In an October 1, 2026 DEV Community article, Zakaria Khchiche reported three such defects in OGX, describing it as the continuation of Meta’s Llama Stack project. The findings below are the author’s account, not independently reproduced behavior or confirmation of the current project state.
What the reported bugs had in common
Each defect could leave a request apparently successful while changing what the system did or recorded. The relevant checks are different: compare requested and applied search configuration, expected and returned documents, and aggregate usage against the details from every model call.
As an Amazon Associate I earn from qualifying purchases.
Elasticsearch hybrid search: a misspelled setting was dropped
Khchiche reported that the allowed parameter name for an Elasticsearch reciprocal rank fusion (RRF) setting was rank_windows_size, with an extra “s.” A caller using the intended spelling, rank_window_size, could have the setting treated as unknown and silently dropped. The article illustrates a request for a window size of 50 while the previous behavior used 10. These are the author’s example values, not a general default guarantee.
The author said the correction was submitted as pull request #6697. The source does not establish whether it was merged or which release, if any, contains it. For a similar integration, check the effective configuration passed to the search backend rather than treating a successful API response as proof that the requested value took effect.
#1 Best Overall
pgvector filters: Python values did not match JSON metadata
The second reported defect affected IN and NOT IN list filters. According to the article, values were converted to Python text before comparison. That could make Python True become "True", while a PostgreSQL JSON metadata value was extracted as "true". It also describes a numeric mismatch between 2023.0 and 2023.
The consequential case is exclusion: if a filter meant to exclude a document fails to match its metadata, that document can remain eligible for retrieval. In a RAG pipeline, this could allow unpublished or outdated material into the context. The author reported testing on PostgreSQL 16 with two documents; that is the article’s test description, not an independent reproduction.
| Author’s example filter | Before reported fix | After reported fix |
|---|---|---|
published in [true] |
No result | Document 0 |
published not in [true] |
Documents 0 and 1 | Document 1 |
year in [2023] |
No result | Document 1 |
These outcomes are the article’s reported results for its two-document example. Khchiche said the fix was submitted as pull request #6698; current merge and release status are not established by the source. Filter tests should include booleans and numeric representations, and assert the exact returned document IDs for both inclusion and exclusion cases.
Recommended Free Tools
Usage details: token counters reflected only the final call
The third defect concerned multiple model calls in one operation. The article says total input and output tokens were summed correctly, but cached-token and reasoning-token details were overwritten by the final call instead of accumulated.
| Usage detail in the article’s example | Across two calls | Reported before | Reported after |
|---|---|---|---|
| Cached tokens | 80 + 120 | 120 | 200 |
| Reasoning tokens | 10 + 15 | 15 | 25 |
Under-counted detail can distort cost tracking: the cached-token figure understates caching, while the reasoning-token figure understates reasoning usage. The author identified issue #6699 and proposed commit e8177d0. The reviewed article does not establish whether a fix shipped. To catch this class of error, test chained calls and compare every detailed usage field with the sum of the corresponding values from all calls.
Practical checks for silent failures
- Configuration: inspect or assert the value that reaches the downstream search system, not just the request sent to the framework.
- Typed filters: test boolean values and equivalent numeric forms, then check exact result sets for both
INandNOT IN. - Multi-call accounting: verify detailed counters across two or more calls, including cached and reasoning tokens, as well as totals.
- Fix status: check the relevant upstream pull request or issue and the release notes for the version you deploy. The article alone does not show that any reported correction is present in a current release.
Khchiche summarized the risk this way: “A bug that crashes gets noticed right away. A bug that returns a plausible but wrong result can run in production for months without anyone noticing.”
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →
