[nightshift] fix documentation drift - #9132
marin-ops-agent[bot] wants to merge 3 commits into
Conversation
The `IndexSpec` enum listed four variants; it now has six. Document `AdaptiveValueCounts` and `AdaptiveGroupExtrema`, their query families, and the `Schema.grouped_extrema` declaration that drives the second one. Adaptive sections are best-effort: a segment that declines one above its cardinality or byte budget stays complete, while a missing declared `ValueCounts` section still makes the bundle stale. Value-count substitution no longer requires complete coverage across visible segments. Contained segments read summaries, boundary and L0 segments keep an ordinary aggregate, and a final aggregate merges both exactly; grouped extrema use the same shape under `FinelogGroupExtrema`. Correct the `training-status` projection column list, which was missing `run_id`, `job_id`, and `attributes_json`, and record that the telemetry schema now declares fourteen projections rather than one. Mark the benchmark section as the acceptance run for the original four-method family so its numbers are not read as current.
|
🤖 Note on the I ran the |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 75727dd8b8
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
docs/design/finelog-exact-analytics.mddescribes Finelog's segment indexfamily in the present tense, and the implementation moved past it.
The
IndexSpecenum listed four variants; it now has six. DocumentAdaptiveValueCountsandAdaptiveGroupExtrema, their query families, and theSchema.grouped_extremadeclaration that drives the second one. Adaptivesections are best-effort: a segment that declines one above its cardinality or
byte budget stays complete, while a missing declared
ValueCountssection stillmakes the bundle stale.
Value-count substitution no longer requires complete coverage across visible
segments. Contained segments read summaries, boundary and L0 segments keep an
ordinary aggregate, and a final aggregate merges both exactly; grouped extrema
use the same shape under
FinelogGroupExtrema.Correct the
training-statusprojection column list, which was missingrun_id,job_id, andattributes_json, and record that the telemetry schemanow declares fourteen projections rather than one. Mark the benchmark section as
the acceptance run for the original four-method family so its numbers are not
read as current.
The other five docs under
docs/design/were checked against the code in thesame pass. Their file paths, symbols, and config values still resolve, and the
markdown links are intact.