Skip to content

RUN_LOG.md

Latest commit

 

History

History
198 lines (166 loc) · 16.7 KB

RUN_LOG.md

File metadata and controls

198 lines (166 loc) · 16.7 KB
Raw

PhaseProbe v0.1.0 run log

Evidence is recorded only from executed commands. Private research paths, PDFs, extracted text, credentials, and machine-specific absolute paths are excluded from this public log.

Scope and research — 2026-08-01

  • Created only the PhaseProbe project root, then .gitignore, AGENTS.md, and this log; read all three back before research or implementation.
  • Created the ignored private research ledger and source directories before retrieval. Four requested PDFs were retrieved privately, SHA-256 hashed, and excluded from Git. The unidentified PDF was identified from internal metadata rather than its URL. No PDF, extracted text, quotation, illustration, or chapter substitute is tracked.
  • Verified prior art and public technical claims against primary papers or official project documentation. PRIOR_ART.md records direct links and a dated comparison.
  • Exact-name registry checks on 2026-08-01 returned no phaseprobe GitHub repository and HTTP 404 from PyPI and npm project endpoints. The name decision is documented as a practical collision audit, not legal clearance.
  • Installed uv 0.11.10 from its official release artifact under ignored .tools/; downloaded artifact SHA-256: 7A0C424C7BC55A74751F13592235953EBBE182FA00355F7AE3FB7AB734A51638.
  • Installed GitHub CLI 2.97.0 from its official release artifact under ignored .tools/; downloaded artifact SHA-256: 35D7FE05C4DD1411FFDA1E73DFC7C6F44B75C936CA51FA6595C657FDC0350CEC.
  • Installed CPython 3.12.13, environments, dependencies, and caches only within ignored project directories. uv lock --check and uv sync --frozen --all-extras passed.

Architecture freeze

v0.1.0 uses a zero-runtime-dependency Python src/ package with typed model adapters, canonical JSON configuration, explicit seeds, deterministic fixed-step execution, finite-state validation, capped traces, exact trace hashes, bounded one-dimensional scan/perturb search, repeatability confirmation, policy checks, integrity-protected replay fixtures, a fixed pytest generator, and terminal/JSON/offline-HTML reports.

The public commands are scan, perturb, check, replay, generate-test, and report. The stable exit contract is 0 completed, 1 policy/explicit fail-on-finding, 2 invalid input, 3 numerical failure, and 4 internal defect.

Measured numerical evidence

Command suite: all eight built-in positive and negative-control commands, each with JSON output and its expected exit code.

example observed result baseline changed/control exit
logistic qualitative transition found period-2 period-4 0
logistic-negative no qualitative transition found period-2 n/a 0
lorenz finite-time trajectory divergence found two-lobe finite-time trajectory same qualitative class; separation predicate triggered 0
lorenz-negative no sensitive perturbation found two-lobe finite-time trajectory threshold not reached 0
predator-prey check policy passed bounded positive oscillation no invariant violation 0
predator-prey-negative check policy failed bounded positive oscillation coarse-step invariant drift 1
toggle qualitative state switch found u-dominant v-dominant 0
toggle-negative no sensitive perturbation found u-dominant u-dominant 0

All eight commands completed in 12.238 seconds. The logistic classification bracket was 3.449477539 .. 3.449478027, with a 4.882812501e-07 final endpoint separation. This is finite-time classifier evidence, not an exact bifurcation point. The Lorenz result is explicitly a finite-time divergence rate, not a Lyapunov exponent.

The predator–prey refined RK4 configuration passed its declared 1e-7 invariant-drift tolerance; the deliberately coarse negative control failed that same policy. Invalid-state tests covered hard-limit failure, and the engine checks NaN, infinity, arithmetic/domain failures, and state-dimension mismatch.

Performance evidence

  • Instrumented phaseprobe scan --example logistic: exit 0, 10.114 seconds wall time, 34.81 MiB peak process-tree working set, and 20.90 MiB peak process-tree private bytes. Monitoring followed the Windows virtual-environment launcher and its child processes.
  • Repository-native demo generation: 14.957 seconds total. Its embedded real-command transcript measured scan 9.879 seconds, replay 1.328 seconds, generation 1.330 seconds, and generated pytest 1.631 seconds.
  • Full example suite: 12.238 seconds.
  • Final coverage suite: 38 tests in 64.86 seconds. The quick-start target of 30 seconds, full-example target of three minutes, and 512 MiB memory target were met on this measured environment.

These are single-environment observations, not cross-machine benchmark claims.

Validation gates

gate exact command or method observed result
format python -m ruff format --check . 28 files already formatted
lint python -m ruff check . passed
strict typing python -m mypy src tests passed; 25 source files
tests and branch coverage python -m pytest --cov=phaseprobe --cov-report=term-missing 38 passed in 64.86 seconds; 87.92% coverage; 85% gate met
generated test python -m pytest -q tests/generated 1 passed; committed fixture replays exact class and trace hashes
configuration lock uv lock --check passed
frozen development install uv sync --frozen --all-extras passed
version python -m phaseprobe --version phaseprobe 0.1.0
examples eight positive/negative commands expected statuses and exits; 12.238 seconds
quick start instrumented positive logistic scan 10.114 seconds; 34.81 MiB peak working set
replay positive logistic fixture both endpoints matched classification, model identity, and exact retained trace hashes
test generation positive logistic fixture fixed-template test and copied integrity fixture created; pytest passed
reports terminal, JSON, HTML regeneration tests passed; HTML contains no script or CDN reference
demo python scripts/generate_demo.py GIF, static PNG, sanitized real-command transcript created
documentation links python scripts/check_links.py 15 Markdown files, 22 external links inventoried, no missing local links
privacy/secret/large file python scripts/hygiene.py 77 tracked public files scanned, no issues, no file over 1 MB
package contents wheel/SDist archive inspection wheel 32 files, source archive 76 files, no PDF/private research/tool/environment path
runtime dependencies wheel metadata plus clean uv environment no runtime Requires-Dist; only PhaseProbe installed before test tooling
packed install wheel into isolated project-local environment version, positive scan, strict replay, test generation, and generated pytest passed
packed quick start isolated wheel environment 9.131 seconds

Package artifacts

  • Wheel: phaseprobe-0.1.0-py3-none-any.whl, 42,313 bytes, final SHA-256 207F303D70ABB5D9CCBC019BCF1AB9AF51D769838054283DF3A7C5D5D79CBA18.
  • Source distribution: phaseprobe-0.1.0.tar.gz, 352,805 bytes, final SHA-256 4BC793351FBEAC740A4ECC15A681AFE567ACBE3E94D0DCD54262667BAECDA6A5.

Failures and repairs

  1. The first editable build failed because package metadata referenced a README that had not yet been written. Added the README and reran successfully.
  2. The first broad Ruff invocation traversed ignored project-local tool/cache files. Added explicit .tools, .cache, environment, private research, output, and build exclusions; all subsequent ruff . gates were bounded to public project content.
  3. The first generated-test subprocess passed an absolute Windows test path; pytest misinterpreted it and inspected a protected compatibility directory. Ran the subprocess from the generated directory with a relative filename.
  4. Initial measured coverage was 82.19%, below the declared 85% gate. Added command-level replay, test-generation, report, failure-code, and integrity tests; final coverage is 87.92%.
  5. A nested generated-pytest process inherited coverage-control variables and produced incompatible combine data. Removed only those instrumentation variables from the nested test environment; functional behavior remained unchanged.
  6. The first packed-smoke script used an unavailable old-PowerShell parameter. Replaced it with an explicit artifact count and compatible selection.
  7. A uv-created smoke environment intentionally had no pip module, so python -m pip freeze was not a valid dependency query. Repeated the gate with uv pip freeze --python; wheel metadata independently confirmed zero runtime dependencies.
  8. The first committed generated test failed Ruff import-block normalization. Updated the generator template, regenerated the test, and restarted the full final gate.
  9. The first staged git diff --check found extra blank lines at EOF and CRLF-sensitive demo/fixture lines. Normalized the reported text endings, added repository-level LF attributes, and reran the full local gates; the final staged diff check passed.

Remote publication evidence

  • Verified GitHub authentication as the required aliengineering-byte account before initializing Git.
  • Created the public aliengineering-byte/phaseprobe repository with main as its default branch, then pushed agent/phaseprobe-v0.1.0.
  • Opened draft pull request #1 from the release branch to main.
  • GitHub Actions run 30719800789 passed all seven jobs: package-and-hygiene plus Windows and Ubuntu on Python 3.10, 3.12, and 3.14.
  • Tag creation, release publication, and the unauthenticated public-clone verification necessarily occur after this release commit is frozen; their evidence is reported in the final release handoff rather than retroactively changing the tagged source.

Skipped or bounded checks

  • External documentation links are inventoried in CI but not fetched there to avoid network-flaky builds; the prior-art/name audit used live official/primary sources during this release run.
  • No formal trademark or legal clearance was performed.
  • No PyPI or npm publication was attempted or authorized.
  • The tagged release remains immutable; post-tag release and fresh-public-clone evidence is therefore external to this tracked pre-release log.

v0.2.0 SciPy integration log (append-only)

This section records the PhaseProbe 0.2.0 SciPy integration and release mission. Entries are appended in execution order. Measurements are observations from the named environment, not cross-platform performance claims.

2026-08-02 pre-edit baseline

  • Project root: the public PhaseProbe repository root; neighboring projects were not inspected.
  • Branch and HEAD: clean main tracking origin/main at 92ed4a14c3eecca82fc990c2cf2eb28b6a3b6a82.
  • Remote: public https://github.com/aliengineering-byte/phaseprobe.git for fetch and push.
  • Latest and only tag: annotated v0.1.0; local and public remote dereference to 92ed4a14c3eecca82fc990c2cf2eb28b6a3b6a82. The tag object is 4e12bbbafc1215cea6f369f8e3e34a50ffe1f77c and will not be changed.
  • Environment: CPython 3.12.13, Windows, PhaseProbe 0.1.0 from the project virtual environment.
  • Base wheel audit: phaseprobe-0.1.0-py3-none-any.whl has no unconditional Requires-Dist; only the existing dev and demo extras declare packages.
  • Formatting: 28 files already formatted; exit 0 in 2.046 seconds.
  • Ruff lint: all checks passed; exit 0 in 0.189 seconds.
  • Strict mypy: no issues in 25 source files; exit 0 in 10.707 seconds.
  • Tests and branch coverage: 38 passed in 70.47 seconds (73.153 seconds wall time), with 87.92% total coverage and the 85% gate met.
  • Generated pytest: 1 passed in 1.13 seconds (1.844 seconds wall time).
  • Package build: wheel and source archive built successfully in 27.571 seconds.
  • Link inventory: 15 Markdown files and 22 external links; no missing local links; exit 0 in 0.410 seconds.
  • Hygiene: 77 tracked public files scanned; no secret, personal-data, or file-over-1-MiB issue; exit 0 in 0.226 seconds.
  • All eight existing examples completed in 14.358 seconds: logistic positive found a qualitative transition (exit 0, 11.172 seconds); logistic negative found none (exit 0, 0.263 seconds); Lorenz positive found finite-time trajectory divergence (exit 0, 0.862 seconds); Lorenz negative found none (exit 0, 0.422 seconds); predator–prey positive passed policy (exit 0, 0.319 seconds); predator–prey negative failed its declared policy as expected (exit 1, 0.226 seconds); toggle positive found a qualitative state switch (exit 0, 0.602 seconds); toggle negative found none (exit 0, 0.413 seconds).
  • Release target: v0.2.0, because v0.1.0 remains the latest public release tag.
  • Release tooling observation: gh is not installed in this environment. No global tool or Git configuration was changed.

2026-08-02 integration and pre-release validation

  • Current upstream audit: SciPy 1.18.0 documentation and the public solve_ivp implementation were reviewed, including solver methods, evaluation grids, events, vectorization, tolerances, termination status, invalid values, complex-domain caveats, and cross-platform replay limits. Only the public scipy.integrate.solve_ivp API is used.
  • Compatibility policy: the optional extra declares NumPy >=1.23.5,<2.5 and SciPy >=1.15,<1.19. The lock resolves supported SciPy releases for Python 3.10 through 3.14 while the base wheel retains no unconditional runtime dependency.
  • Architecture: the existing step-level protocol remains intact. A backward-compatible trajectory protocol dispatches one bounded adaptive solve while sharing scan, perturbation, invariant, artifact, reporting, and test-generation machinery.
  • Replay: schema 2.0 adds declared exact and tolerance modes and retains integrity hashes. Existing schema 1.0 fixtures remain readable and use exact trace comparison; adaptive SciPy fixtures record tolerances and environment evidence without claiming byte-identical replay.
  • Installed-wheel core validation: 46 tests passed and three SciPy-dependent tests skipped in an isolated environment where NumPy and SciPy were absent. Core import, CLI execution, exact replay, and the dependency audit passed.
  • Installed-wheel SciPy validation: SciPy 1.18.0 and NumPy 2.4.6 were installed solely through the wheel's scipy extra. The predator-prey policy and tolerance replay passed, and all three committed generated pytest regressions passed.
  • Full validation: 81 tests passed in 109.91 seconds with 86.37% branch coverage, exceeding the unchanged 85% gate. Formatting covered 40 files; Ruff passed; strict mypy passed for 35 source files; the lockfile check passed.
  • All 12 built-in and SciPy positive/negative example commands matched their declared exit outcomes in 35.815 seconds. The SciPy Lorenz positive case found finite-time trajectory divergence; its short-window control found none. The tight DOP853 predator-prey configuration passed its first-integral policy, while the deliberately coarse RK23 configuration failed it.
  • Instrumented Windows measurements using the real CPython 3.12 process: the base logistic-map quick start completed in 10.962 seconds with a 19.92 MiB peak working set; the SciPy Lorenz example completed in 7.830 seconds with a 73.54 MiB peak working set. These are observations of this host, not performance guarantees.
  • Documentation/demo validation: the verified terminal transcript, GIF, static fallback, HTML report, replay fixtures, generated tests, five-minute quick start, and unposted upstream proposal are present. Link inventory passed for 17 Markdown files and 29 external links.
  • Public-tree hygiene: 101 tracked or untracked candidate files were scanned; no secret, personal-data, workstation-path, PDF, or file-over-1-MiB issue was found. git diff --check passed. Neighboring projects and global Git configuration were not touched.

2026-08-02 pre-release CI portability repair

  • The first GitHub matrix run exposed two CI-environment defects: strict mypy could not resolve NumPy type aliases in the intentionally SciPy-free environment, and top-level packed-smoke virtual environments were correctly treated as unignored public candidates by the expanded hygiene scanner.
  • Static checks were moved into the full SciPy matrix, where the optional type namespaces are installed on both operating systems and all supported Python versions. The core matrix still executes all dependency-free tests and examples without NumPy or SciPy.
  • Packed-smoke environments now live under the ignored .cache directory in CI and tagged release validation. No hygiene exclusion was weakened, and the 85% full-install branch-coverage gate remains unchanged.
  • Feature-branch push triggering was removed because pull-request triggering already covers the branch; this prevents duplicate full matrix runs while retaining main and release-branch pushes.