Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
251 commits
Select commit Hold shift + click to select a range
93265da
test: account for matched task verification metrics
EvanProgramming Oct 21, 2026
9a94b32
test: add V3 matched evaluation controls
EvanProgramming Oct 21, 2026
8952842
feat: add matched V3 plan evaluation runner
EvanProgramming Oct 21, 2026
67d689f
fix: fail closed on incomplete matched workflows
EvanProgramming Oct 21, 2026
0193183
fix: harden V3 matched evaluation evidence
EvanProgramming Oct 21, 2026
5c3ee5b
fix: apply RMB cap to V3 evaluation costs
EvanProgramming Oct 21, 2026
39790c7
fix: stop matched run after provider errors
EvanProgramming Oct 21, 2026
d902a3e
fix: fail-stop V3 relay on transport errors
EvanProgramming Oct 21, 2026
49d6630
fix: ignore local relay client disconnects
EvanProgramming Oct 21, 2026
2fc0afb
test: normalize HTTP header names in provider coverage
EvanProgramming Oct 21, 2026
0abecc9
fix: keep streaming relay calls concurrent
EvanProgramming Oct 21, 2026
47ee773
fix: stop matched run on failed stream events
EvanProgramming Oct 21, 2026
f8268b4
fix: isolate concurrent CLI spinner display
EvanProgramming Oct 22, 2026
01763a6
fix: serialize CLI spinner output process-wide
EvanProgramming Oct 22, 2026
1e82817
fix: gate benchmark on sandbox and emit JSON results
EvanProgramming Oct 22, 2026
cb57fc9
test: allow delegated review to finish in CI
EvanProgramming Oct 22, 2026
a01fb70
fix: enforce matched benchmark evidence
EvanProgramming Oct 22, 2026
7a14d38
fix: keep V2 baseline compatible in evaluation
EvanProgramming Oct 22, 2026
3c9608c
fix: retain stale benchmark reservations across model changes
EvanProgramming Oct 22, 2026
f1a3826
feat: retain archived reservations in cohort reports
EvanProgramming Oct 22, 2026
a292374
test: handle shallow checkouts without V2 baseline
EvanProgramming Oct 22, 2026
6aa9e63
feat: pin matched benchmark to gpt-6-luna pricing
EvanProgramming Oct 22, 2026
fd142b2
fix: account for Panstar benchmark pricing
EvanProgramming Oct 22, 2026
29d7bec
fix: reconcile V3 benchmark reservation accounting
EvanProgramming Oct 22, 2026
552e89e
fix: normalize system messages in matched relay
EvanProgramming Oct 22, 2026
97d4df0
fix: stop relay at terminal response events
EvanProgramming Oct 22, 2026
7a6e3cc
fix: preserve cumulative benchmark spend across retries
EvanProgramming Oct 22, 2026
aa12e40
fix: mark incomplete benchmark workers inconclusive
EvanProgramming Oct 22, 2026
bbed941
fix: serialize parallel conversation journal appends
EvanProgramming Oct 22, 2026
e83a22a
fix: enforce the approved benchmark spend cap
EvanProgramming Oct 22, 2026
b0f6c9e
fix: make unknown benchmark settlement idempotent
EvanProgramming Oct 22, 2026
5ac77f0
fix: retry failed matched benchmark workers
EvanProgramming Oct 22, 2026
cbffab9
fix: carry forward cumulative benchmark exposure
EvanProgramming Oct 22, 2026
400c41e
fix: retain latest benchmark exposure
EvanProgramming Oct 22, 2026
33f029a
fix: update approved V3 evaluation spend cap
EvanProgramming Oct 22, 2026
09c6c93
fix: carry forward failed V3 benchmark exposure
EvanProgramming Oct 22, 2026
41a591a
fix: retain safe provider failure codes in benchmark
EvanProgramming Oct 22, 2026
40f4c81
fix: carry forward diagnostic benchmark exposure
EvanProgramming Oct 22, 2026
20984a1
fix: capture bounded HTTP provider error codes
EvanProgramming Oct 22, 2026
19b3ddf
fix: carry forward latest V3 provider holds
EvanProgramming Oct 22, 2026
e8efde8
docs: carry forward latest V3 benchmark reservation
EvanProgramming Oct 22, 2026
bd5fc23
docs: carry forward latest V3 benchmark hold
EvanProgramming Oct 22, 2026
e1209e9
fix: enforce cumulative benchmark spend cap
EvanProgramming Oct 22, 2026
ede7f5b
fix(benchmark): raise approved evaluation cap to twenty dollars
EvanProgramming Oct 22, 2026
d46b428
fix(benchmark): restore approved ten dollar cap
EvanProgramming Oct 22, 2026
e37e815
fix(benchmark): carry forward interrupted run exposure
EvanProgramming Oct 22, 2026
718cd2c
fix(benchmark): carry forward aborted candidate exposure
EvanProgramming Oct 22, 2026
9974fc9
feat: enforce signed V3 release evidence gates
EvanProgramming Oct 21, 2026
054a74f
fix: keep release evidence outside candidate checkout
EvanProgramming Oct 22, 2026
e2df426
fix: publish release manifest only after signing
EvanProgramming Oct 22, 2026
ad6ad41
fix: prevent release manifest overwrite races
EvanProgramming Oct 22, 2026
31bcc39
feat: collect exact-candidate V3 platform evidence
EvanProgramming Oct 22, 2026
3e01e12
test: run R266 evidence matrix on release PRs
EvanProgramming Oct 22, 2026
70ac495
feat: validate V3 release candidate platforms
EvanProgramming Oct 22, 2026
c3ff7aa
test: make gateway acceptance proxy-safe
EvanProgramming Oct 22, 2026
e1bb0d7
fix: make production soak dependency-free
EvanProgramming Oct 22, 2026
2b6a663
test: isolate startup log readiness from loopback
EvanProgramming Oct 22, 2026
c0fb1e1
test: wait for TUI turn completion
EvanProgramming Oct 22, 2026
cf68555
fix(memory): close vector index on application shutdown
EvanProgramming Oct 22, 2026
93a78ee
fix(memory): release orphaned vector clients
EvanProgramming Oct 22, 2026
b7a376c
fix(tui): emit ready after clearing busy state
EvanProgramming Oct 22, 2026
187d882
fix(tests): close temporary memory bank after bug workflow
EvanProgramming Oct 22, 2026
eb4d2a3
test(tui): dismiss error dialog before update command
EvanProgramming Oct 22, 2026
b9ff1e4
test: disable unused vector index in bug workflow
EvanProgramming Oct 22, 2026
3ca4e2c
ci: gate Windows V3 validation on supported features
EvanProgramming Oct 22, 2026
b7fad07
ci: force UTF-8 for Windows feature acceptance
EvanProgramming Oct 23, 2026
7bcff3b
test(tui): isolate Windows home directory
EvanProgramming Oct 23, 2026
c912686
ci: record Windows PTY acceptance limitation
EvanProgramming Oct 23, 2026
e848ed9
fix(soak): measure Windows handle growth from baseline
EvanProgramming Oct 23, 2026
cf0a694
fix(sqlite): apply intended sync mode to runtime connections
EvanProgramming Oct 23, 2026
a611b07
fix(cli): honor stdout encoding for Windows rendering
EvanProgramming Oct 23, 2026
59f261e
test: stabilize cross-platform soak and TUI checks
EvanProgramming Oct 23, 2026
415b492
fix(ci): invoke release evidence collector as module
EvanProgramming Oct 23, 2026
e3f7c46
test: allow scheduling delay in cancellation checks
EvanProgramming Oct 23, 2026
a740594
fix: enforce cumulative evaluation spend cap
EvanProgramming Oct 23, 2026
166fe4b
fix: refresh benchmark budget exchange rate
EvanProgramming Oct 23, 2026
deee1c6
fix: run matched cohort with candidate worker
EvanProgramming Oct 23, 2026
dbdb83d
fix: enforce approved cumulative benchmark cap
EvanProgramming Oct 23, 2026
030268a
merge: stack release evidence on matched evaluation
EvanProgramming Oct 23, 2026
2a5a20f
fix: extend soak task completion window
EvanProgramming Oct 23, 2026
5f5a612
test: allow macOS delegation review to finish
EvanProgramming Oct 23, 2026
985f725
fix: use approved cumulative benchmark cap
EvanProgramming Oct 23, 2026
ad13af4
merge: update release evidence candidate cap
EvanProgramming Oct 23, 2026
2104cd3
fix: validate Windows TUI evidence status
EvanProgramming Oct 23, 2026
a4cba7c
fix: carry forward benchmark campaign reservations
EvanProgramming Oct 24, 2026
9b7c3bb
merge: update release evidence budget carry-forward
EvanProgramming Oct 24, 2026
1f0be44
fix: account for all benchmark campaign exposure
EvanProgramming Oct 24, 2026
36882e6
merge: refresh release evidence candidate accounting
EvanProgramming Oct 24, 2026
d85ce5e
fix: resume matched benchmark checkpoints
EvanProgramming Oct 24, 2026
9c6ef88
merge: include matched benchmark resume fix
EvanProgramming Oct 24, 2026
46df10f
feat: allow signed limited-evidence V3 previews
EvanProgramming Oct 26, 2026
7955dfb
fix(benchmarks): pin selected model and remove stale hold
EvanProgramming Oct 27, 2026
396a2d0
merge: update matched-evaluation base
EvanProgramming Oct 27, 2026
1656066
test: diagnose delegated provider startup timeouts
EvanProgramming Oct 27, 2026
d4679da
fix(release): require complete preview safety evidence
EvanProgramming Oct 27, 2026
0b8ebab
fix(release): parse multiline safety test results
EvanProgramming Oct 27, 2026
8f4ba7b
fix(benchmarks): verify strategic plan fixtures
EvanProgramming Oct 27, 2026
cb5087b
Merge branch 'Evan/v3-matched-evaluation' into Evan/v3-release-evidence
EvanProgramming Oct 27, 2026
7f1883d
fix: close benchmark ledger connections
EvanProgramming Oct 28, 2026
6eb881a
fix: verify release evidence snapshots
EvanProgramming Oct 28, 2026
5fc2e62
test: reject failed V3 security gate waiver
EvanProgramming Oct 28, 2026
0fa7551
fix(ci): fetch pinned V2 baseline for matched tests
EvanProgramming Oct 28, 2026
668a3e4
docs: document pinned V2 checkout requirement
EvanProgramming Oct 28, 2026
0734dbb
fix: reject observed protocol recovery failures
EvanProgramming Oct 29, 2026
54d1722
fix(benchmarks): align plan fixtures with scoped execution
EvanProgramming Oct 29, 2026
08d2b06
Merge branch 'main' of https://github.com/EvanProgramming/OpenKyrozen…
EvanProgramming Oct 29, 2026
86e22dd
merge: update V3 evaluation dependency
EvanProgramming Oct 30, 2026
8ce7ec9
feat: gate V3 preview publication
EvanProgramming Oct 30, 2026
0f172f9
chore: sign preview candidate with verified identity
EvanProgramming Oct 30, 2026
c942491
test: exercise preview updates and diagnose cancellation timeouts
EvanProgramming Oct 30, 2026
1e0bf5d
fix: require working plans before complex native tools
EvanProgramming Oct 31, 2026
082ada1
fix: preserve blocked plan completion status
EvanProgramming Oct 31, 2026
c988001
fix: retain permitted plan revisions in compact mode
EvanProgramming Oct 31, 2026
8294ed5
test: make provider deadline checks deterministic
EvanProgramming Oct 31, 2026
29a5213
fix: archive prior native tool results
EvanProgramming Nov 2, 2026
caa7f44
fix: archive legacy tool results between rounds
EvanProgramming Nov 2, 2026
24617f1
fix: protect active database during history rollback
EvanProgramming Nov 3, 2026
24005eb
test: isolate bugfix workflow from background learning
EvanProgramming Nov 3, 2026
7905fb6
chore: prepare 3.0.0 stable release candidate
EvanProgramming Nov 3, 2026
389fe2b
fix: keep public downloads on the published stable release
EvanProgramming Nov 6, 2026
ed8e442
test: trace Windows cancellation completion failures
EvanProgramming Nov 6, 2026
c37a1fe
fix: block unsafe V3 evaluation and meter DeepSeek in CNY
EvanProgramming Nov 6, 2026
1c1df7f
fix: avoid asynchronous diagnostic watchdog deadlocks
EvanProgramming Nov 6, 2026
5b8ee3b
docs: record V3 finalization evidence and blockers
EvanProgramming Nov 6, 2026
289c0bf
fix: retain benchmark cancellation and efficiency evidence
EvanProgramming Nov 6, 2026
f3e0ddf
test: isolate workflow acceptance from user state
EvanProgramming Nov 6, 2026
bcae0c9
fix: complete native planning and preserve execution evidence
EvanProgramming Nov 6, 2026
fda6688
fix: verify matched cohort behavior and dispatched provenance
EvanProgramming Nov 6, 2026
60df1e6
docs: record native V3 repairs and unresolved release gates
EvanProgramming Nov 6, 2026
38dd672
test: synchronize cancellation assertions with manager state
EvanProgramming Nov 6, 2026
0318e68
test: add bounded Windows cancellation load controls
EvanProgramming Nov 6, 2026
9e7e7ab
fix: require complete measured evidence for learning promotion
EvanProgramming Nov 6, 2026
aa00f37
fix: retain scoped process uncertainty across cancellation and restart
EvanProgramming Nov 6, 2026
2536983
fix: derive final effect claims and child reports from verified outcomes
EvanProgramming Nov 6, 2026
469d76e
docs: record remaining formal release evidence boundaries
EvanProgramming Nov 6, 2026
1d3c9e0
fix: keep protection probes isolated from parent credentials
EvanProgramming Nov 6, 2026
d96159f
fix: retain parent-bound worker evidence before qualifying evaluation…
EvanProgramming Nov 7, 2026
833fcea
docs: bind independent evaluation review and full search coverage
EvanProgramming Nov 7, 2026
8f92f5f
test: preserve provider safety distinction with complete worker fixtures
EvanProgramming Nov 7, 2026
eb8c606
test: reconstruct original Windows cancellation environment
EvanProgramming Nov 7, 2026
99fdb1f
feat: bind benchmark processes to ephemeral kernel identities
EvanProgramming Nov 7, 2026
252b0e1
test: inspect effective Darwin groups through the kernel API
EvanProgramming Nov 7, 2026
54be922
test: distinguish cancellation checkpoint latency with paired controls
EvanProgramming Nov 7, 2026
0951cda
fix: retain owned SQLite connections through runtime shutdown
EvanProgramming Nov 7, 2026
f97eaba
fix: preserve substantive routing signals across task wording
EvanProgramming Nov 7, 2026
d97b716
feat: isolate matched benchmark actors behind a privileged parent
EvanProgramming Nov 7, 2026
80a0620
test: validate legacy runtime and canonical Windows diagnostics
EvanProgramming Nov 7, 2026
f77474e
fix: preserve delegated history and close actor launch races
EvanProgramming Nov 7, 2026
b3f1187
test: retain namespace bootstrap failure diagnostics
EvanProgramming Nov 7, 2026
39ee61d
fix: preserve strict actor startup across platform sandboxes
EvanProgramming Nov 7, 2026
ab27bab
test: capture native namespace capability failures
EvanProgramming Nov 7, 2026
24050df
fix: complete Linux capability cleanup before actor exec
EvanProgramming Nov 7, 2026
52df6c4
test: trace native Windows cancellation waits with paired controls
EvanProgramming Nov 7, 2026
dfe975f
fix: default DeepSeek to Flash while retaining model fallback intent
EvanProgramming Nov 7, 2026
c87ecef
fix: preserve parent-observed timeout and process state snapshots
EvanProgramming Nov 7, 2026
51e83df
test: retain one bounded Windows trace across cancellation controls
EvanProgramming Nov 7, 2026
0818783
test: retain only native scheduling and file I/O trace signals
EvanProgramming Nov 7, 2026
027f382
test: complete isolated routing and Git acceptance evidence
EvanProgramming Nov 7, 2026
4f6bf9c
docs: distinguish stable installers from unpublished V3 source
EvanProgramming Nov 7, 2026
e065481
fix: read child cancellation from the live owning supervisor
EvanProgramming Nov 7, 2026
03a7db5
fix: track Windows soak process creation identities
EvanProgramming Nov 7, 2026
0e3f470
test: isolate the soak observer process family
EvanProgramming Nov 7, 2026
0109494
fix: keep benchmark assertions outside edited source processes
EvanProgramming Nov 7, 2026
9e9ec70
fix: align DeepSeek smoke with production tool requests
EvanProgramming Nov 7, 2026
a802a3a
fix: separate benchmark outcomes and declare file scope
EvanProgramming Nov 7, 2026
c0e2fc6
fix: continue after rejected native plan batches
EvanProgramming Nov 7, 2026
d668c70
fix: align benchmark callers with scoped prompts
EvanProgramming Nov 7, 2026
f0f437a
fix: verify working plan completion before live evaluation
EvanProgramming Nov 8, 2026
360ac2f
test: use typed content in strategic read receipt
EvanProgramming Nov 8, 2026
44625e6
fix: retain completed plan identity in native context
EvanProgramming Nov 8, 2026
8c595f9
test: include archive reader in routing plan schema
EvanProgramming Nov 8, 2026
02aa70d
fix: clarify read-only plan completion arguments
EvanProgramming Nov 8, 2026
13e3154
fix: refresh current native tool availability
EvanProgramming Nov 9, 2026
842588d
fix: keep native discovery outside task evidence
EvanProgramming Nov 9, 2026
6632e8b
fix: bound structured working plan continuation per turn
EvanProgramming Nov 9, 2026
b7f6164
fix: include Gemini thinking tokens in charged output
EvanProgramming Nov 9, 2026
5a0141f
fix: meter Gemini usage with model-specific estimates
EvanProgramming Nov 9, 2026
72ac9ec
feat: add guarded Gemini automatic evaluation
EvanProgramming Nov 9, 2026
5368395
fix: complete native Gemini evaluation compatibility
EvanProgramming Nov 9, 2026
6ebc93f
fix: preserve native acceptance provider flag
EvanProgramming Nov 9, 2026
22c182b
feat: support uncapped Gemini evaluation accounting
EvanProgramming Nov 10, 2026
426b2d5
docs: retire the owner-waived historical macOS attribution blocker
EvanProgramming Nov 11, 2026
9f31f44
fix: preserve native Panstar auto evaluation and accounting
EvanProgramming Nov 11, 2026
e86da2a
fix: request explicit native smoke continuation confirmation
EvanProgramming Nov 11, 2026
eca3d3a
fix: retain bounded native tool evidence without recursive archives
EvanProgramming Nov 11, 2026
aadaa56
test: cover inline and explicitly archived planning evidence
EvanProgramming Nov 11, 2026
b3f3f62
fix: certify benchmark safety after durable provider settlement
EvanProgramming Nov 11, 2026
97ab81b
fix: use signed Ubuntu mirror for CI dependencies
EvanProgramming Nov 11, 2026
f8fb2c9
fix: verify investigation fixtures against structured source facts
EvanProgramming Nov 12, 2026
7efd301
fix: separate physical source reads from legacy workflow heuristics
EvanProgramming Nov 12, 2026
8448683
fix: align matched fixture evidence with physical task contracts
EvanProgramming Nov 12, 2026
1fd3f27
test: synchronize recovery fixture and control streaming deadline
EvanProgramming Nov 12, 2026
2543346
fix: verify planned command checks using exact execution receipts
EvanProgramming Nov 14, 2026
16203b4
fix: retain streamed usage after benchmark client disconnect
EvanProgramming Nov 14, 2026
487b368
fix: bound active benchmark responses streams
EvanProgramming Nov 14, 2026
6fe7f15
fix: let disconnected benchmark actors finish teardown
EvanProgramming Nov 15, 2026
fa39c13
fix: retain benchmark actor channel through interpreter exit
EvanProgramming Nov 15, 2026
46102f5
fix: qualify Terra for the V3 Panstar campaign
EvanProgramming Nov 15, 2026
0b5949a
fix: request final text after a completed native plan
EvanProgramming Nov 15, 2026
ac7651e
fix: declare source receipt and plan verification requirements
EvanProgramming Nov 16, 2026
d51ae85
fix: carry active tool guidance in native plan replies
EvanProgramming Nov 16, 2026
e0b6fa4
test: read the full archived native plan reply
EvanProgramming Nov 16, 2026
465d910
test: make downstream drain accounting deterministic
EvanProgramming Nov 16, 2026
81019dd
fix: support metered DeepSeek Flash release evaluation
EvanProgramming Nov 16, 2026
97d1aba
fix: bound DeepSeek Pro to the model routing comparison
EvanProgramming Nov 16, 2026
8fe366a
fix: reject separately billed model tool usage
EvanProgramming Nov 16, 2026
e52eea1
fix: admit approved DeepSeek ledger modes at relay startup
EvanProgramming Nov 16, 2026
a33b140
fix: keep active tool guidance after native continuation
EvanProgramming Nov 16, 2026
037eb50
fix: defer active tool guidance only for selected tasks
EvanProgramming Nov 16, 2026
3e91bbf
test: isolate gateway and streaming diagnostics
EvanProgramming Nov 16, 2026
2a50f98
fix: stage material plan revisions after verified work
EvanProgramming Nov 16, 2026
d163f00
fix: recover from mistaken pre-plan artifact lookups
EvanProgramming Nov 16, 2026
76d2312
test: specify the expected source condition expression
EvanProgramming Nov 16, 2026
b1bec82
test: record Windows soak process identity with Win32
EvanProgramming Nov 16, 2026
20f0625
fix: retain legacy plan evidence without synthetic fingerprints
EvanProgramming Nov 16, 2026
f85164b
fix: seed verifiable plans and distinguish task progress
EvanProgramming Nov 16, 2026
20fa482
fix: score plan replacement using ordered verification receipts
EvanProgramming Nov 16, 2026
c9c5921
fix: recognize merged stderr in failed fixture reproduction
EvanProgramming Nov 16, 2026
ebe7173
fix: verify fixture test identity across display options
EvanProgramming Nov 16, 2026
6d85922
fix: verify commits in native working plans
EvanProgramming Nov 17, 2026
ffa0ab8
fix: preserve reproduction exit evidence in fixture prompts
EvanProgramming Nov 17, 2026
20678da
fix: use strict native tools in the DeepSeek release benchmark
EvanProgramming Nov 17, 2026
e462912
fix: satisfy DeepSeek strict schema validation
EvanProgramming Nov 17, 2026
fdcef65
fix: decode canonical DeepSeek strict tool arguments
EvanProgramming Nov 17, 2026
786fc65
fix: defer incomplete TUI acceptance events
EvanProgramming Nov 17, 2026
7f1a29f
fix: retain typed validation for strict tool calls
EvanProgramming Nov 17, 2026
331089a
fix: validate complete strict tool responses before execution
EvanProgramming Nov 17, 2026
1a4115d
fix: preserve compatible Chat completions without a finish reason
EvanProgramming Nov 17, 2026
dd4e245
fix: simplify strict plan inputs and confirm verified tasks consistently
EvanProgramming Nov 17, 2026
b12a7ff
fix(ci): fall back across signed Ubuntu archive mirrors
EvanProgramming Nov 17, 2026
39c9f29
fix: align V3 verification prompts with completed plan state
EvanProgramming Nov 18, 2026
b090b0c
fix: bind complete native reads to selected path criteria
EvanProgramming Nov 18, 2026
b0e7f7a
fix: carry whole-file verification through native receipts
EvanProgramming Nov 18, 2026
7ee998b
fix: use the workspace tool boundary for read provenance
EvanProgramming Nov 18, 2026
7586d93
fix: close V3 offline acceptance gaps before metered evaluation
EvanProgramming Nov 18, 2026
58e2f8d
fix: carry verified reads through approved plan criteria
EvanProgramming Nov 18, 2026
f4e42eb
fix: align advertised tools with approved plan bounds
EvanProgramming Nov 18, 2026
df40021
fix: verify delegated native evidence from scoped receipts
EvanProgramming Nov 18, 2026
7d44309
fix: handle native contexts without delegation metadata
EvanProgramming Nov 18, 2026
5eaa3c7
test: wait independently for captured process streams
EvanProgramming Nov 18, 2026
8a5292b
fix: align native plan edits with verified progress
EvanProgramming Nov 18, 2026
9129e6f
test: use native Windows counters for production soak
EvanProgramming Nov 18, 2026
5b68913
fix: align native plan creation schema and guidance
EvanProgramming Nov 18, 2026
353f352
fix: expose exact native plan verification mismatches
EvanProgramming Nov 18, 2026
0b733da
fix: align native planning tools with completion state
EvanProgramming Nov 18, 2026
ff38ede
fix: scope active native plan edits to the current revision
EvanProgramming Nov 18, 2026
939148f
test: isolate caller deadline from socket idle timing
EvanProgramming Nov 18, 2026
b898ba6
fix: align strict plan arguments and seeded verification
EvanProgramming Nov 18, 2026
8bc7b9b
fix: preserve archived read verification and bound process handles
EvanProgramming Nov 19, 2026
60d5a8a
fix: recover from malformed native tool responses
EvanProgramming Nov 19, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
242 changes: 238 additions & 4 deletions .github/workflows/ci.yml

Large diffs are not rendered by default.

2 changes: 2 additions & 0 deletions .github/workflows/release.yml
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,7 @@ on:
push:
tags:
- "v*"
- "!v3.*"
workflow_dispatch:
inputs:
release_tag:
Expand All @@ -16,6 +17,7 @@ permissions:

jobs:
build-release:
if: ${{ !startsWith(inputs.release_tag || github.ref_name, 'v3.') }}
runs-on: ubuntu-latest
env:
RELEASE_TAG: ${{ inputs.release_tag || github.ref_name }}
Expand Down
93 changes: 93 additions & 0 deletions .github/workflows/v3-benchmark-isolation.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,93 @@
name: V3 benchmark isolation diagnostics
on:
push:
branches: ['Evan/v3-finalization']
paths: ['benchmarks/**', 'tests/test_v3_protected_worker.py', '.github/workflows/v3-benchmark-isolation.yml']
workflow_dispatch:
permissions:
contents: read
jobs:
isolation:
strategy:
fail-fast: false
matrix:
os: [ubuntu-latest, macos-latest]
python: ['3.12', '3.13']
runs-on: ${{ matrix.os }}
timeout-minutes: 20
steps:
- uses: actions/checkout@v4
with:
fetch-depth: 0
- uses: actions/setup-python@v5
with:
python-version: ${{ matrix.python }}
- name: Install Linux backend without changing host security policy
if: runner.os == 'Linux'
run: |
sudo "$(command -v python)" scripts/ci_ubuntu_mirror.py
sudo apt-get update
sudo apt-get install --no-install-recommends -y bubblewrap
sysctl kernel.apparmor_restrict_unprivileged_userns user.max_user_namespaces > isolation-policy-before.txt
- name: Install project dependencies
run: python -m pip install -e '.[all]'
- name: Record diagnostic identity
shell: bash
run: |
git rev-parse HEAD > isolation-commit.txt
python -m pip freeze > isolation-dependencies.txt
if [[ "$RUNNER_OS" == Linux ]]; then
{
bwrap --version
ls -l /usr/bin/bwrap
uname -r
cat /proc/self/attr/current
if [[ -f /etc/apparmor.d/bwrap ]]; then cat /etc/apparmor.d/bwrap; fi
} > isolation-backend.txt
fi
- name: Verify actor isolation and cleanup without persistent accounts
if: always()
shell: bash
run: |
set -euo pipefail
interpreter="$(command -v python)"
umask 022
stage_root="$(mktemp -d /var/tmp/openkyrozen-isolation.XXXXXX)"
trap 'sudo -n rm -rf -- "$stage_root"' EXIT
chmod 755 "$stage_root"
git clone --no-checkout --no-hardlinks "$PWD" "$stage_root/source"
git -C "$stage_root/source" checkout --detach "$GITHUB_SHA"
test "$(git -C "$stage_root/source" rev-parse HEAD)" = "$GITHUB_SHA"
sudo -n chown -R 0:0 "$stage_root"
private_home="$stage_root/parent-home"
sudo -n mkdir -p "$private_home"
sudo -n chmod 700 "$private_home"
cd "$stage_root/source"
sudo -n env -i PATH="$PATH" HOME="$private_home" TMPDIR=/var/tmp PYTHONPATH="$PWD" \
PYTHONDONTWRITEBYTECODE=1 KYROZEN_DISABLE_VECTOR_INDEX=1 \
"$interpreter" -m unittest tests.test_v3_process_owner tests.test_v3_linux_namespace \
tests.test_v3_actor_supervisor tests.test_v3_actor_worker tests.test_v3_actor_dispatch \
tests.test_v3_actor_native_acceptance \
tests.test_v3_routing_supplement \
tests.test_v3_verifier_isolation \
tests.test_v3_fixture_strength \
tests.test_v3_matched_live.V3MatchedLiveTests.test_fixture_tests_verify_the_actual_workspace_source \
tests.test_v3_shell_cleanup tests.test_v3_protected_worker -v 2>&1 | tee "$GITHUB_WORKSPACE/isolation-tests.log"
- name: Verify host security policy was preserved
if: always() && runner.os == 'Linux'
shell: bash
run: |
sysctl kernel.apparmor_restrict_unprivileged_userns user.max_user_namespaces > isolation-policy-after.txt
diff -u isolation-policy-before.txt isolation-policy-after.txt
- name: Preserve diagnostics
if: always()
uses: actions/upload-artifact@v4
with:
name: isolation-${{ matrix.os }}-py${{ matrix.python }}
path: |
isolation-commit.txt
isolation-dependencies.txt
isolation-backend.txt
isolation-tests.log
isolation-policy-before.txt
isolation-policy-after.txt
151 changes: 151 additions & 0 deletions .github/workflows/v3-windows-reliability.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,151 @@
name: V3 Windows cancellation investigation
on:
push:
branches: ['Evan/v3-windows-cancellation-investigation-20261107']
paths: ['.github/workflows/v3-windows-reliability.yml', 'scripts/v3_windows_reliability.py', 'tests/test_v3_windows_reliability.py']
workflow_dispatch:
inputs:
candidate_sha:
description: Exact 40-character product commit to test (diagnostic code remains separate)
type: string
required: true
default: '7905fb634fe447243668c2d01179ce5cb2cdb275'
control:
description: Load control; paired-anchor compares equal CPU load with/without read-only anchor (1–10 pairs)
type: choice
options: [baseline, cpu, sqlite, combined, paired-anchor, native-wait, all]
default: baseline
iterations:
description: Samples per control and Python version (1–50)
type: number
default: 50
environment:
description: Current packages, or exact original Python 3.12.10/packages with reviewer interleaving
type: choice
options: [current, historical]
default: current
env:
PRODUCT_SHA: ${{ inputs.candidate_sha || '7905fb634fe447243668c2d01179ce5cb2cdb275' }}
ITERATIONS: ${{ github.event_name == 'workflow_dispatch' && format('{0}', inputs.iterations) || '50' }}
HISTORICAL_ENVIRONMENT: ${{ inputs.environment == 'historical' && '1' || '0' }}
permissions:
contents: read
jobs:
cancellation:
runs-on: windows-latest
timeout-minutes: 60
strategy:
fail-fast: false
matrix:
python: ${{ fromJSON(inputs.environment == 'historical' && '["3.12.10"]' || '["3.12","3.13"]') }}
control: ${{ fromJSON(inputs.control == 'all' && '["baseline","cpu","sqlite","combined"]' || format('["{0}"]', inputs.control || 'baseline')) }}
env:
LOAD_CONTROL: ${{ matrix.control }}
steps:
- name: Validate exact candidate identity
shell: pwsh
run: |
if ($env:PRODUCT_SHA -cnotmatch '^[0-9a-f]{40}$') { throw 'candidate_sha must be a full lowercase commit SHA' }
if ($env:ITERATIONS -cnotmatch '^[0-9]+$' -or [int]$env:ITERATIONS -lt 1 -or [int]$env:ITERATIONS -gt 50) { throw 'iterations must be 1–50' }
if ($env:HISTORICAL_ENVIRONMENT -eq '1' -and [int]$env:ITERATIONS -gt 20) { throw 'historical campaign is bounded at 20 iterations' }
if ($env:LOAD_CONTROL -eq 'paired-anchor' -and [int]$env:ITERATIONS -gt 10) { throw 'paired-anchor is bounded at 10 pairs' }
if ($env:LOAD_CONTROL -eq 'native-wait' -and ($env:HISTORICAL_ENVIRONMENT -ne '1' -or [int]$env:ITERATIONS -ne 4)) { throw 'native-wait requires historical environment and exactly four triples' }
- name: Checkout diagnostic harness
uses: actions/checkout@v4
with:
path: diagnostic-source
fetch-depth: 0
- name: Checkout exact product
uses: actions/checkout@v4
with:
ref: ${{ env.PRODUCT_SHA }}
path: product
fetch-depth: 0
- name: Checkout original product for native wait comparison
if: matrix.control == 'native-wait'
uses: actions/checkout@v4
with:
ref: '0f172f9fe692ff8fafcc0a15a72fe51eaf67e977'
path: historical-product
fetch-depth: 0
- uses: actions/setup-python@v5
id: python
with:
python-version: ${{ matrix.python }}
- name: Retain unavailable historical runtime evidence
if: failure() && steps.python.outcome == 'failure'
shell: pwsh
run: |
New-Item -ItemType Directory -Force windows-evidence | Out-Null
'{"status":"INFRASTRUCTURE_FAILURE","reason":"Requested Python runtime could not be provisioned"}' | Set-Content -Encoding utf8 windows-evidence/historical-environment.json
- name: Verify checked-out product
shell: pwsh
working-directory: product
run: |
if ((git rev-parse HEAD) -cne $env:PRODUCT_SHA) { throw 'Checked-out candidate differs from requested commit' }
if (git status --porcelain --untracked-files=no) { throw 'Product checkout has tracked edits' }
- name: Install exact product dependencies
shell: pwsh
run: |
New-Item -ItemType Directory -Force windows-evidence | Out-Null
if ($env:HISTORICAL_ENVIRONMENT -eq '1') {
python diagnostic-source/scripts/v3_windows_reliability.py --write-historical-requirements windows-evidence/historical-requirements.txt
if ($LASTEXITCODE -ne 0) { throw 'Unable to prepare historical requirements' }
python -m pip install -r windows-evidence/historical-requirements.txt -e './product[all]' *> windows-evidence/dependency-install.log
if ($LASTEXITCODE -ne 0) {
'{"status":"INFRASTRUCTURE_FAILURE","reason":"Original pinned dependencies could not be installed; no substitutions allowed"}' | Set-Content -Encoding utf8 windows-evidence/historical-environment.json
throw 'Historical dependency installation failed; inspect dependency-install.log'
}
} else {
python -m pip install -e './product[all]'
if ($LASTEXITCODE -ne 0) { throw 'Dependency installation failed' }
}
- name: Record dependency versions
shell: pwsh
run: |
New-Item -ItemType Directory -Force windows-evidence | Out-Null
python -m pip freeze | Set-Content -Encoding utf8 windows-evidence/dependencies.txt
$env:GITHUB_SHA | Set-Content -Encoding utf8 windows-evidence/diagnostic-commit.txt
- name: Trace unchanged cancellation test under recorded load control
shell: pwsh
run: |
$historical = @()
if ($env:HISTORICAL_ENVIRONMENT -eq '1') { $historical = @('--historical-environment', '--interleave-reviewer') }
$native = @()
if ($env:LOAD_CONTROL -eq 'native-wait') { $native = @('--historical-repository', 'historical-product') }
python diagnostic-source/scripts/v3_windows_reliability.py --repository product --output windows-evidence --iterations $env:ITERATIONS --control $env:LOAD_CONTROL --cpu-workers 2 @historical @native
- name: Verify diagnostic harness independently of historical product
if: always() && steps.python.outcome == 'success'
working-directory: diagnostic-source
shell: python
run: |
import json, os, sys, tempfile
from pathlib import Path
root = Path.cwd().resolve()
sys.path.insert(0, str(root))
from scripts.v3_windows_reliability import isolated_environment, run_process
output = root.parent / 'windows-evidence'
output.mkdir(exist_ok=True)
code = ("from pathlib import Path; import openkyrozen, unittest; "
"print('Diagnostic product source:', openkyrozen.__file__, flush=True); "
"assert Path(openkyrozen.__file__).resolve().is_relative_to(Path.cwd()); "
"unittest.main(module='tests.test_v3_windows_reliability', verbosity=2)")
with tempfile.TemporaryDirectory(prefix='kyrozen-diagnostic-tests-') as home:
env = isolated_environment(Path(home))
env['PYTHONPATH'] = str(root)
env['KYROZEN_DIAGNOSTIC_TEST_OUTPUT'] = str(output / 'diagnostic-test-evidence')
result = run_process([sys.executable, '-c', code], output / 'diagnostic-unittest.log', env, timeout=300)
returncode = result['returncode']
(output / 'diagnostic-unittest.json').write_text(json.dumps({
'status': 'INFRASTRUCTURE_FAILURE' if returncode == 124 else 'PASS' if returncode == 0 else 'FAIL',
**result, 'diagnostic_commit': os.environ['GITHUB_SHA'],
'python': sys.version, 'scope': 'diagnostic-source only; separate from historical product acceptance',
}, indent=2), encoding='utf-8')
raise SystemExit(returncode)
- name: Upload investigation evidence
if: always()
uses: actions/upload-artifact@v4
with:
name: windows-cancellation-${{ matrix.python }}-${{ env.PRODUCT_SHA }}-${{ matrix.control }}
path: windows-evidence/
if-no-files-found: error
2 changes: 2 additions & 0 deletions README.ja.md
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,8 @@

## インストール

V3 は未リリースです。以下のコマンドは公開済み安定版 **v2.0.8** をインストールします。

OpenKyrozen は Python **3.12 と 3.13** をサポートします。Python 3.14 は対象外です。

### macOS または Linux
Expand Down
2 changes: 2 additions & 0 deletions README.ko.md
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,8 @@

## 설치

V3는 아직 출시되지 않았습니다. 아래 명령은 공개된 안정 버전 **v2.0.8**을 설치합니다.

OpenKyrozen은 Python **3.12와 3.13**을 지원합니다. Python 3.14는 지원 대상이 아닙니다.

### macOS 또는 Linux
Expand Down
2 changes: 1 addition & 1 deletion README.md
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,7 @@

## Install

The current published stable release is **v2.0.8**. It supports Python 3.12 and 3.13; Python 3.14 is outside the supported range.
V3 is not yet released. The commands below install the published stable **v2.0.8**. It supports Python 3.12 and 3.13; Python 3.14 is outside the supported range.

macOS or Linux:

Expand Down
2 changes: 2 additions & 0 deletions README.zh-CN.md
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,8 @@

## 安装

V3 尚未发布。下方命令安装已发布的稳定版 **v2.0.8**。

OpenKyrozen 支持 Python **3.12 和 3.13**。Python 3.14 暂不支持。

### macOS 或 Linux
Expand Down
Loading
Loading