feat: idle-time self-test, context probe, resources roll-up and usage measurement - #52
Merged
Merged
Conversation
… measurement Four intents from one lane, landed together because their surfaces are the same files: the panel, the control plane, the app's composition root, the configuration and the pool. The self-test (itd-2609100457007827) measures every downloaded model while nobody is using the Mac — llama-bench's pp512 and tg128, a concurrent tg128 and the load time — through the pool's ordinary Acquire, never evicting, yielding the moment a client appears, into a bounded 0600 file with a view on the Statistics tab. The context-window probe (itd-2609091301112705) is a second job of that one idle loop: a sweep and a bisection through this Mac's own OpenAI endpoint, the window recorded on the model's registry entry with the provenance that makes it stale, published on the models list as measured_context and measured_bound, adopted as the served window only by the operator's hand. A reading one of the gateway's own bounds stopped is a floor and says so. The resources roll-up (itd-2609091903463596) adds up what the Models tab already knows, with free disk on the snapshot from the one reader. Usage measurement (itd-2609091712141073) gives the request record the facts that decide a served window and a concurrency, the load event the launch sampling, and the store a fifth record kind — a footprint reading every thirty seconds per server, from top's physical figure with a deadline — with three views on the Statistics tab and every field on the reference page. Each intent had two adversarial reviews applied; the decision lines in the records say what changed and why. Assisted-by: Claude:claude-fable-5-1 Claude-Session: https://claude.ai/code/session_01NU6NHFLGFYFuA1z4eqFz8j
The intent, spec, capture, decision and calibration records for the self-test, the context probe, the resources view and usage measurement: four intents to shipped/, four specs closed, three follow-up captures for this lane, one resolved, the decision record superseding the statistics store's record kinds, and the dated lines behind every choice. Assisted-by: Claude:claude-fable-5-1 Claude-Session: https://claude.ai/code/session_01NU6NHFLGFYFuA1z4eqFz8j
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Four intents from one lane, built and reviewed on one branch because their surfaces are the same files.
What lands
Residency.Refusalscounter), into a bounded 0600 file with a view on the Statistics tab. Off by default;self_testin config, a Settings checkbox, a posture-page line.measured_contextandmeasured_boundon the models list; Measure now and Use this window on the card; a floor when a gateway bound stopped it.context_probeandidle_threshold_secin config.footprint(adr-2609121450000000 supersedes adr-2609090716413337), a sampler reading top's physical footprint with a deadline, three views on the Statistics tab.ApproxRecordBytes230 → 380, and the docs' arithmetic with it.Review
Each intent had a ruthless review and a security review; every blocking finding is fixed and recorded in
.abcd/work/DECISIONS.md. Three integration tests drive the probe through the real gateway and a real pool over a fake launcher.Gates
go test -race ./...,go vet ./...,gofmt -l .,abcd docs lintgreen on 6b1f600.abcd lintadds nothing to main's count.Owed after merge
https://claude.ai/code/session_01NU6NHFLGFYFuA1z4eqFz8j