docs: close runtime quality review with consolidated lesson
Intrepid/Prometeu/Runtime/pipeline/pr-master There was a failure building this commit
Intrepid/Prometeu/Runtime/pipeline/pr-master There was a failure building this commit
This commit is contained in:
+6
-5
@@ -127,13 +127,14 @@ This evidence does not claim Linux or Intel Mac coverage execution.
|
||||
The module-tree ownership correction derives from accepted DEC-0059. Its
|
||||
[current measured evidence](../../../docs/vm-arch/COVERAGE_DOMAINS_CURRENT.json)
|
||||
preserves the earlier evidence records and restores system/core to 87.71%
|
||||
without changing the other 14 floors. The fresh local map has 257 source files
|
||||
and 15 exclusive domains. All domain gates pass: system/fs measures 78.73%,
|
||||
without changing the other 14 floors. The initial repaired map had 257 source files
|
||||
and 15 exclusive domains. Its domain gates passed: system/fs measured 78.73%,
|
||||
system/core 93.69%, asset/bank 90.52% and drivers/core 89.17%.
|
||||
Ten policy tests cover both generic validity and intended specialized owners.
|
||||
The existing repository target passes 1,021 tests with 11 ignored; strict
|
||||
workspace Clippy still reports existing VM diagnostics. Coverage qualification
|
||||
must not be presented as completed strict workspace qualification.
|
||||
That initial repository target passed 1,021 tests with 11 ignored while strict
|
||||
workspace Clippy still reported VM diagnostics. The final strict gate subsequently
|
||||
passed on the completed source, as recorded in the [runtime quality lesson](../DSC-0064-runtime-branch-quality-review/LSN-0070-runtime-quality-needs-independent-oracles-and-owned-evidence.md).
|
||||
The current evidence retains the initial ownership comparison separately.
|
||||
|
||||
## Takeaways
|
||||
|
||||
|
||||
+1
-1
@@ -110,7 +110,7 @@ data after remount. This establishes priority and retained progress through
|
||||
observable ordering, rather than a host-specific latency threshold. Separate
|
||||
regressions cover cancellation, superseding and preload publication.
|
||||
|
||||
This refinement follows [DEC-0064](../../workflow/decisions/DEC-0064-q7-share-asset-decoding-validation-and-conformance.md).
|
||||
This refinement follows DEC-0064, summarized in the [runtime quality lesson](../DSC-0064-runtime-branch-quality-review/LSN-0070-runtime-quality-needs-independent-oracles-and-owned-evidence.md).
|
||||
The published contract is [Asset Management](../../../docs/specs/runtime/15-asset-management.md#43-decoding-validation-and-physical-payload-completeness).
|
||||
The ordering fixture uses the persistent transaction protocol with a deterministic
|
||||
storage backend; file-host latency measurements remain separate evidence.
|
||||
|
||||
+1
-1
@@ -54,7 +54,7 @@ Interpolation follows the same selected boundary. A one-shot interpolates into t
|
||||
|
||||
The correction derives from DEC-0060 in DSC-0064. Synthetic regressions cover exact PCM, fractional seams and tails, incomplete and invalid markers, empty samples, rate conversion, multiple wraps and extreme finite pitch. These tests exercise the existing delivery and control paths without establishing new device timing or reclamation guarantees. The [sample contract](../../../docs/specs/runtime/05-audio-peripheral.md#6-samples-interpolation-and-ownership) remains normative.
|
||||
|
||||
Local macOS/aarch64 qualification on 2026-10-14 passed 19 host audio tests (one physical-device test ignored), nine driver audio tests and ten runtime audio tests. The existing `make ci-domains` target passed with 1,027 tests passed, zero failed and 11 ignored, including all 15 domain gates. Strict all-target/all-feature Clippy passed for the desktop host. Strict workspace qualification remains pending because of existing VM lint errors; this correction does not claim to resolve the repository-wide quality gate.
|
||||
Local macOS/aarch64 qualification on 2026-10-14 passed 19 host audio tests (one physical-device test ignored), nine driver audio tests and ten runtime audio tests. The existing `make ci-domains` target passed with 1,027 tests passed, zero failed and 11 ignored, including all 15 domain gates. Strict all-target/all-feature Clippy passed for the desktop host. That initial run preceded repair of existing VM lint errors. The later strict shared gate passed on the completed source; see the [runtime quality lesson](../DSC-0064-runtime-branch-quality-review/LSN-0070-runtime-quality-needs-independent-oracles-and-owned-evidence.md) for final acceptance and complementary production decode/save pressure evidence.
|
||||
|
||||
### Original Voice-Control Qualification
|
||||
|
||||
|
||||
+1
-1
@@ -134,7 +134,7 @@ quotas. This is boundary-work evidence, not a claim of faster game frames.
|
||||
Synthetic collision keys and work-limit tests complement real-address timings;
|
||||
allocator-selected address order is not a stable test oracle across loads.
|
||||
|
||||
This implementation follows [DEC-0065](../../workflow/decisions/DEC-0065-q8-index-string-provenance-within-existing-quotas.md).
|
||||
This implementation follows DEC-0065, summarized in the [runtime quality lesson](../DSC-0064-runtime-branch-quality-review/LSN-0070-runtime-quality-needs-independent-oracles-and-owned-evidence.md).
|
||||
The [published ownership contract](../../../docs/specs/runtime/02c-program-registry-and-pbx-v1.md#memory-ownership-and-quotas)
|
||||
remains normative; native import, snapshot retention and typed storage tests
|
||||
verify the same owner lifetimes through the new indices.
|
||||
|
||||
+122
@@ -0,0 +1,122 @@
|
||||
---
|
||||
id: LSN-0070
|
||||
ticket: runtime-branch-quality-review
|
||||
title: Runtime Quality Needs Independent Oracles and Owned Evidence
|
||||
created: 2026-10-14
|
||||
tags: [quality, runtime, ownership, validation]
|
||||
---
|
||||
|
||||
## Context
|
||||
|
||||
A branch review found eight gaps spanning deadline reasons, coverage ownership,
|
||||
one-shot audio, recurring VM allocations, callback reclamation, CI enforcement,
|
||||
asset decoder parity and string provenance lookup. Accepted decisions DEC-0058
|
||||
through DEC-0065 were executed by PLN-0189 through PLN-0196. This lesson preserves
|
||||
the resulting reasoning after removal of those workflow artifacts; the linked
|
||||
published contracts remain normative.
|
||||
|
||||
The common failure mode was evidence that proved a nearby property while leaving
|
||||
the intended property unchecked. A callback can avoid locks while still freeing
|
||||
memory. Two decoders can agree on output while accepting the same malformed file.
|
||||
A coverage percentage can rise because ownership moved, without any new test.
|
||||
A fast lookup can hide expensive construction and uncharged metadata.
|
||||
|
||||
## Key Decisions
|
||||
|
||||
### Observe the boundary where the guarantee becomes true
|
||||
|
||||
The [memcard contract](../../../docs/specs/runtime/08-save-memory-and-memcard.md)
|
||||
keeps the first deadline escalation reason stable while distinguishing durable
|
||||
completion from pending or failed work. Fake-clock boundary tests and explicit
|
||||
publication ordering make those distinctions observable without sleeping.
|
||||
|
||||
The [audio contract](../../../docs/specs/runtime/05-audio-peripheral.md) distinguishes
|
||||
one-shot completion from looping and logical inactivity from physical ownership.
|
||||
A finished voice may become silent before its sample can be reclaimed. Two bounded
|
||||
return packets move sample and vector owners to host service; pool exhaustion
|
||||
retains bounded ownership and defers delivery. Count both allocations and frees
|
||||
inside the callback, then independently prove eventual owner release outside it.
|
||||
|
||||
### Shared mechanics still need independent expected results
|
||||
|
||||
The [asset contract](../../../docs/specs/runtime/15-asset-management.md) uses common
|
||||
metadata validation and decoding mechanics for synchronous and incremental paths.
|
||||
Conformance checks include independently specified pixels, palette bytes and PCM,
|
||||
plus malformed metadata and physically truncated payloads. Agreement between two
|
||||
entry points is useful, but cannot by itself establish correctness.
|
||||
|
||||
The real production decoder/save ordering fixture pauses at a safe decode step,
|
||||
queues persistent work and checks writes and synchronization before decode resumes.
|
||||
That deterministic ordering oracle is stronger evidence of priority than a latency
|
||||
threshold on one host. Host latency remains a separate observation.
|
||||
|
||||
### Move cost deliberately and account for the destination
|
||||
|
||||
Recurring VM stack/return work reuses capacity after warmup; release allocator
|
||||
checks measure the execution region separately from setup. String provenance uses
|
||||
image and session indices under existing ownership quotas. Index metadata and
|
||||
construction cost are charged and measured, including collision and refusal paths.
|
||||
See the [Registry contract](../../../docs/specs/runtime/02c-program-registry-and-pbx-v1.md)
|
||||
and [provenance qualification](../../../docs/vm-arch/string-provenance-qualification.md).
|
||||
Lookup improvements do not establish faster game frames or cheaper construction.
|
||||
|
||||
### The repository owns acceptance and measurement identity
|
||||
|
||||
The [coverage policy](../../../docs/vm-arch/COVERAGE_DOMAINS.md) assigns each source
|
||||
to exactly one domain, includes module descendants, and keeps acceptance floors
|
||||
independent of observed percentages. Ownership repair and behavioral improvement
|
||||
must be evaluated separately against the same report before comparing fresh runs.
|
||||
|
||||
The shared strict gate records source/config identity, preserves command failures,
|
||||
removes stale reports and publishes available evidence even when a stage fails.
|
||||
Local execution and CI call the same target. A successful focused test does not
|
||||
replace the full gate; a successful local gate does not certify an unexecuted OS.
|
||||
|
||||
## Published Evidence and Its Limits
|
||||
|
||||
The final source commit is `3f9d2355462d3b5675f5edc6e488ab325eb024d8`, with Rust
|
||||
fingerprint `309a069f8b0560d1e9d1f9f4e011b709c89729383aeacd100c8c2ff5c3cff28d`.
|
||||
On macOS/aarch64 with rustc 1.98.1, all twelve `make ci-domains` stages passed:
|
||||
1,060 ordinary tests, five socket tests, three runner tests and all fifteen fixed
|
||||
coverage floors. Thirteen ordinary tests were ignored. The fresh coverage replay
|
||||
also passed 1,060 tests; those are repeated tests, not additional unique cases.
|
||||
[Current coverage evidence](../../../docs/vm-arch/COVERAGE_DOMAINS_CURRENT.json)
|
||||
retains the earlier ownership-repair comparison and identifies the final corpus.
|
||||
The ten focused coverage tests and two release/all-feature VM allocation tests
|
||||
passed again during closure without source changes.
|
||||
|
||||
[Audio qualification](../../../docs/vm-arch/audio-callback-qualification.md) retains
|
||||
the named physical-device smoke and the initial device failures. Complementary
|
||||
release measurement ran 1,000 sixteen-voice callbacks while a separate process
|
||||
repeated the real production decoder/FileMemcard fixture fourteen times. Callback
|
||||
p50/p95/p99/max were 14.542/97.667/399.041/1,070.708 microseconds against a nominal
|
||||
5,333.333-microsecond period. Callback allocation/free counts were both zero;
|
||||
sample owners went from sixteen to zero after teardown. The
|
||||
[measurement record](../../../docs/vm-arch/audio-production-pressure-qualification.json)
|
||||
records overlapping process intervals, environment and individual pressure rounds.
|
||||
|
||||
This is host-concurrent pressure evidence, not same-process console integration
|
||||
or a worst-case latency guarantee. Each pressure round alternates twenty production
|
||||
loads and durable 64-KiB commits; the separate deterministic fixture establishes
|
||||
save preemption during decoding. Physical audio, synthetic timing, allocator
|
||||
counts and save ordering answer different questions and retain separate scopes.
|
||||
No Linux/remote CI, human listening assessment, long-duration soak or gameplay
|
||||
speedup is claimed by this closure.
|
||||
|
||||
## Related Domain Lessons
|
||||
|
||||
- [Save publication and resource ownership](../DSC-0061-memcard-fixed-slots-v1/LSN-0067-save-publication-and-resource-ownership-are-separate.md).
|
||||
- [Repository-owned domain gates](../DSC-0002-runtime-edge-test-plan/LSN-0037-domain-gates-must-be-owned-by-the-repository.md).
|
||||
- [Audio delivery and lifecycle](../DSC-0055-guest-audio-voice-control/LSN-0069-audio-admission-needs-a-delivery-boundary-and-independent-lifecycle-control.md).
|
||||
- [Materialization and recurring VM cost](../DSC-0014-perf-vm-allocation-and-copy-pressure/LSN-0035-first-materialization-is-not-the-same-as-hot-path-copy-pressure.md).
|
||||
- [Serial async lanes](../DSC-0009-perf-async-background-work-lanes-for-assets-and-fs/LSN-0050-serial-async-lanes-bound-backlog-complexity.md).
|
||||
- [Registry boundaries and owned values](../DSC-0058-program-registry-type-and-function-metadata/LSN-0059-registry-boundaries-and-owned-values.md).
|
||||
|
||||
## Pitfalls and Takeaways
|
||||
|
||||
- Test the promised boundary with an independent oracle, including failure paths.
|
||||
- Separate admission, execution, publication and reclamation when reasoning about cost.
|
||||
- Preserve owner counts and quota accounting when replacing scans or moving work.
|
||||
- Bind acceptance evidence to source/config fingerprints, not merely a nearby commit.
|
||||
- Close deferred gates on the final source, while keeping environmental limits explicit.
|
||||
- Preserve reusable reasoning and durable measurements; discard completed workflow scaffolding.
|
||||
Reference in New Issue
Block a user