Seascape roadmap progress

Implemented scope: SS-00 through SS-10; the approved PILOT-01 exploratory family gate passed. SS-11 candidate review/handoff prepared; human and owner gates remain pending. SS-03 hosted acceptance passed; no complete audited regional release is claimed. Specification. SS-00 base/current commit: f2400c13d509ad753d9168ed4e7a07d1ffcfc5a4 (no reset). Initial state: clean main; remote https://github.com/MarineCast/toolkit-seascape.git. Working branch: feature/seascape-repository-organization-updates. Feature-branch pushes authorized on September 27; no merge, tag or package publication. Earlier entries retain their dated evidence. Current next task: ACCEPT-01, continuing SS-11's unavailable human trial and owner version, classic-protection, merge and release decisions. FIX-01/FIX-02/VERIFY-01 remain passed. PILOT-01 used approved cached inputs and explicitly exploratory support; canonical water inputs remain unavailable. No SS-12 started.

Task	Status	Evidence / next gap
SS-00	passed	Baseline classified; notebook PASS (12 code cells, no errors)
SS-01	passed	21 regressions, runtime-only guarded wheel demo; full suite 241 passed / 3 skipped
SS-02	passed	Portable copied notebook; 7 regressions; full suite 248 passed / 3 skipped
SS-03	passed	All three hosted consumer cases passed; run 36322829881, tested commit 6ddd5ca
SS-04	passed	Read-only input preflight; 34 new regressions; guarded runtime-only wheel acceptance
SS-05	passed	CLI-only guidance/debug, side-effect contracts, 29 regressions; runtime-only wheel acceptance
SS-06	passed	Ruff baseline; 209 ASTs unchanged; full suite 322 passed / 3 skipped
SS-07	passed	Local expanded lint/type gates; fresh constrained/range solves and audits; broad ignored-cache secret scan failed
SS-08	passed	Analytic/matrix/publication evidence; local and hosted 352 passed / 3 skipped; run 36324502530, 9/9 jobs
SS-09	passed	Three journeys, guarded source quickstart, docs drift checks; local/hosted 368 passed / 3 skipped; run 36327449869, 9/9 jobs
SS-10	passed, bounded exploratory scope	PILOT-01: two fresh cached-input realizations, exact keyed comparison, measured caps and six inspected maps; no whole release
SS-11	blocked	Exact candidate software checks passed; handoff prepared; human/version/admin/release-owner gates pending
FIX-01	passed	Future GEBCO rights corrected; 32 focused tests; retained fixture/data bytes preserved
FIX-02	passed	113 focused regressions; safe detail, original debug, unchanged gates
VERIFY-01	passed	Corrected 50675db; local package acceptance and 9/9 hosted jobs; dated evidence preserved
PILOT-01	passed, bounded exploratory scope	Installed package 232ac30 at execution HEAD 5d9faaf; two R6/R8 family realizations; see final evidence below
ACCEPT-01	pending	No unfamiliar tester; owner decisions and authorization remain outstanding

SS-00 baseline

Current source matches the reviewed baseline. Entry points (source/tests are authoritative):

Operation	Symbols	Existing tests
Init / CLI	cli.initialize_workspace, cli.main	test_standalone_package.py
Bathymetry	bathymetry.load_bathymetry_config, run_pipeline, build_bathymetry_parquet	test_bathymetry_contracts.py, validation notebook
Planning	workflow.selected_stages, run_domain_layer_build	test_workflow.py
Publication	publication.TransactionalSeascapePublisher, publication.SeascapeReleasePublisher	test_publication.py, test_release.py, test_review_regressions.py
Resolution	products.list_products, list_resolutions, resolve_product	test_products.py, test_review_regressions.py
Matrix	metric_matrix.build_metric_matrix	test_metric_matrix.py

Isolated development venv: /tmp/seascape-roadmap-dev, no inherited site packages. Python 3.14.6, macOS 26.6.2 ARM64; GDAL 3.12.4, PROJ 9.8.1, GEOS 3.13.1. Exact commands/exit codes and dependency/native versions: /tmp/seascape-roadmap-evidence/baseline-commands.json and baseline-environment.{json,txt}. Logs are local machine evidence, not committed artifacts.

• python -m pip install -e '.[test,quality,notebook]': initial sandbox run exit 1 (PyPI DNS, not a demonstrated dependency defect); network-enabled retry exit 0. Install logs under /tmp.
• python -m pip check: exit 0.
• python -m pytest -q: exit 0, 220 passed / 0 failed / 3 skipped. Two feature-catalog tests and one network-consumer test require absent regional artifacts. Ten upstream/deprecation warnings.
• ruff check src tests scripts, python -m mypy, current four-module ruff format --check: exit 0 each.
• seascape --help: exit 0.
• seascape --workspace /tmp/seascape-roadmap-evidence/baseline-workspace init and seascape --workspace /tmp/seascape-roadmap-evidence/baseline-workspace build --dry-run: exit 0 each; fresh workspace, 26 stages, no acquisition/publication.
• python scripts/environment_snapshot.py --output /tmp/seascape-roadmap-evidence/baseline-environment: exit 0.
• Notebook: initial run exit 1 (sandbox loopback bind denied); local-socket retry exit 1 (venv CLI missing from PATH); corrected PATH/local-socket retry exit 0, 12 executed code cells, PASS marker and zero errors. Output seascape-toolkit-validation.ipynb under the evidence directory. Committed notebook untouched.
• GitHub connector fetch_commit_workflow_runs for this exact commit returned an empty list; hosted CI has no available execution evidence. Historical remediation counts are dated evidence.

No production behavior changed during SS-00. No unrelated edits existed to preserve. Live acquisition, regional rebuild, consumer integration and remote release checks remain outside scope.

SS-01 portable demo — passed

Implementation and local acceptance complete. Base: 0d6963b284e51184292cbf591abeeac977cca445; this record's containing implementation commit is the SS-01 handoff. Final Git identity is also saved in /tmp/seascape-roadmap-evidence/final-git-state.json and reported in the task handoff.

• Added demo.run_demo(workspace, *, overwrite=False) / DemoResult and seascape demo. The 48×48 notebook recipe feeds the existing bathymetry.run_pipeline with download/map off. Packaged configuration, transactional family publication, checksums and atomic writers are reused.
• All paths belong to .seascape/demo; ordinary configs/canonical products/releases are preserved. Ownership/symlink/path-type/transaction checks and a POSIX lock guard reruns. Only known files are replaced; previous PASS is invalidated before computation, including metadata failures.
• Reserved provider SYNTHETIC records fixture licensing/warnings and requires acquisition and interactive maps disabled. Existing real-provider metadata defaults are regression-tested. No scientific formulas, thresholds, default regional configs, schemas or release gates changed.
• 14 executed checks cover actual nonempty output, exact H3 r8 support, 2016 eligible pixels, 5 m constant depth, observed zero statistics/fractions, null nodata/outside/sea-level support, finite values, sign/CRS, provenance, family/source/upstream checksums and returned paths.
• 21 focused tests passed / 0 failed / 0 skipped; full suite 241 passed / 0 failed / 3 skipped (same absent regional artifacts as SS-00). Full run: 31 dependency/deprecation warnings.
• Runtime-only wheel acceptance: 14/14 checks, 0 network attempts, pytest/Jupyter absent and their imports forbidden. Installed package imported from consumer site-packages, arbitrary cwd /tmp, source overrides cleared. Audit guard verifies outbound rejection and demo filesystem confinement; two Matplotlib font-discovery child attempts were denied, with successful bundled-font fallback. This is Python process evidence, not an OS-level firewall or a sandbox for native extensions.
• Both static PNGs were opened and visually inspected: extent/orientation, synthetic titles, meter/sign labels and grey unavailable cells. Figures generated without basemap/display server.
• Final existing notebook: 12 executed code cells, PASS marker, zero errors. Notebook/config/resource diffs remain empty. Portable thin-client conversion is SS-02 and was not started.

Commands ran from the owning checkout with /tmp/seascape-roadmap-dev/bin on PATH unless indicated. Machine logs and exact baseline/final static commands: /tmp/seascape-roadmap-evidence/.

Command / evidence	Exit / result
python -m pytest -q tests/test_demo.py (demo-tests-final.log)	0; 21 passed
python -m pytest -q (final-tests.log)	0; 241 passed, 3 regional skips
ruff check src tests scripts; python -m mypy	0 each; typing retains existing four-module scope
Current four-module format gate plus src/seascape/demo.py, src/seascape/cli.py, tests/test_demo.py, scripts/check_demo.py	0; 8 formatted files; full command in final-commands.json
python -m pip check (development and runtime consumer)	0 each
python -m pip wheel . --no-deps --no-build-isolation --wheel-dir /tmp/seascape-roadmap-evidence/wheels	First failed (exit code not captured): setuptools absent in venv; declared build requirements installed; final exit 0
python -m pip install 'setuptools>=80' wheel (development venv)	0; declared build prerequisites
Consumer python -m pip install /tmp/seascape-roadmap-evidence/wheels/toolkit_seascape-0.1.0-py3-none-any.whl	First exit 1: wheel not yet built; final fresh runtime/dependency install exit 0, no inherited packages
Consumer python -m pip install --no-deps --force-reinstall /tmp/seascape-roadmap-evidence/wheels/toolkit_seascape-0.1.0-py3-none-any.whl	0; final wheel refresh after runtime dependencies were already installed normally
From /tmp: env -u PYTHONPATH -u SEASCAPE_WORKSPACE -u SEASCAPE_CANDIDATE_ROOT /tmp/seascape-roadmap-consumer/bin/python /tmp/seascape-roadmap-evidence/check_demo.py --workspace /private/tmp/seascape-roadmap-evidence/consumer-demo-confined2	0; 14/14, confinement and outbound guard; consumer-demo-confined2.log
Copied check_installed_package.py, consumer Python from /tmp, PYTHONPATH cleared	0; 161 modules imported with OrcaCast blocked; consumer-imports.log
Consumer seascape --help	0; consumer-help.log
Consumer seascape --workspace /private/tmp/seascape-roadmap-evidence/consumer-demo-confined2 demo	1 expected; existing output refused, PASS report retained
jupyter nbconvert --to notebook --execute notebooks/validation/01_TOOLKIT_VALIDATION.ipynb --ExecutePreprocessor.timeout=120 --output seascape-toolkit-validation-final.ipynb --output-dir /tmp/seascape-roadmap-evidence	0 with local kernel sockets and development venv PATH; final-notebook.log
pip-audit --disable-pip --no-deps --strict -r /tmp/seascape-roadmap-evidence/baseline-environment.txt	0; no known vulnerabilities; dependency-audit.log
gitleaks git . --redact --no-banner --log-opts="--all"	0; no leaks in history
gitleaks dir . --redact --no-banner	1; pre-existing ignored graphify-out/cache/stat-index.json, generic-api-key false positive on a 64-character file hash under hashes["docs/api.md"]
gitleaks dir /tmp/seascape-roadmap-evidence/review-files --redact --no-banner	0; copied tracked + nonignored files; cache preserved, no ignore added
Local Markdown references in 6 changed guides; git diff --check	0; links exist and diff clean
git diff --exit-code -- notebooks config src/seascape/resources	0; no notebook/template edits

Earlier consumer guard trials failed on metadata's platform.platform() subprocess and font probing (RuntimeError did not permit Matplotlib's expected fallback), then descriptor-relative publisher cleanup was incorrectly interpreted relative to cwd. Final code removes metadata's subprocess, records RUNNING before metadata, denies children with PermissionError, and tracks file-descriptor bases in the acceptance helper. Final guarded execution passed; failures were not counted as passes. Wheel/environment evidence: wheel-build-final.log, consumer-environment.json, consumer-demo-confined2/.seascape/demo/report.json and its figures beneath the evidence directory.

Files: src/seascape/demo.py, src/seascape/cli.py, bathymetry pipeline.py, tests/test_demo.py, scripts/check_demo.py, AGENTS.md, README.md, docs/{API,README,demo}.md, this progress record. No new runtime dependencies. No unrelated initial edits existed; existing configuration, retained products, sibling repositories and local graph cache were preserved. No remote changes performed.

Known limits: local macOS ARM64 Python 3.14.6 evidence only; no available hosted CI result for baseline, no Linux/other-interpreter acceptance, no live providers/regional rebuild/downstream integration, no human newcomer trial. Process-global environment selection is not thread-safe; separate-process writers are guarded. All these external checks remain unrun, not software acceptance blockers. The SS-01 handoff recommended SS-02; its subsequent implementation is recorded below.

SS-02 portable notebook — passed

Base: 4017612214878ce181bde69c909b3c718142d74a; clean initial working tree on the existing feature/seascape-first-run-demo branch. This record's containing commit is the SS-02 handoff; exact final Git identity is in /tmp/seascape-ss02-evidence/final-git-state.json. SS-01 prerequisites were reverified against current run_demo, CLI delegation, packaged resources, production bathymetry and tests. Its unchanged runtime wheel was reused, not rebuilt or relabeled: SHA-256 503b38e2972ff61be2a2a307e77dba319df7fbdffdb33582e683372fab8b1682. No baseline reset, live exploration or runtime-only acceptance redo.

• Replaced checkout discovery and fixture/orchestration cells with installed run_demo. Twelve readable code cells inspect environment, selected workspace, generated configuration, input raster/support, output grain/schema/control values, the API's two figures, actual checks and synthetic provenance. Default temporary products persist after kernel shutdown; optional SEASCAPE_DEMO_WORKSPACE selects a workspace. Explicit overwrite remains false by default.
• Added truthful presentation gates for empty/failed checks, report disagreement and environment restoration. Corrected Data Explorer claims: its actual helper performs live San Juan bathymetry exploration with an exploratory water mask, not completed-release inspection/certification.
• Added copied-notebook acceptance with a temporary interpreter-pinned kernel, no global kernel registration, checkout-read rejection, outbound Python socket/DNS rejection and provider denial. Guard self-test rejects outbound access; loopback remains allowed. This is process evidence, not an OS firewall or native-extension network sandbox.
• Both default and chosen-workspace guarded runs passed 14 production / 15 notebook checks, displayed 2 PNGs, retained synthetic reports after kernel shutdown, restored all four inspected environment variables (including sentinel overrides), and recorded 0 acquisition / 0 outbound attempts. The consumer has notebook extras but still no pytest; imports resolve in its isolated wheel site-packages. Each input directory contains only its copied notebook.
• Plain nbconvert also passed: 12 executed code cells, 0 errors, 2 PNGs, actual PASS marker, explicit interpreter/import paths. Source is unexecuted, contains no machine paths/embedded outputs, and stayed byte-identical through execution. Final source SHA-256: f618fc5df6bc2d68eeda842e6e22f135da80827dde85b1adac1382a1b3c545b3.

Exact command lines, cwd, exit codes and logs: /tmp/seascape-ss02-evidence/commands.json. Development interpreter /tmp/seascape-roadmap-dev/bin/python; consumer /tmp/seascape-roadmap-consumer/bin/python. Python 3.14.6, macOS 26.6.2 ARM64; GDAL 3.12.4 / PROJ 9.8.1 / GEOS 3.13.1. Environment inventories: development-environment.{json,txt}, consumer-environment.json under that evidence directory.

Command / evidence	Exit / result
Consumer python -m pip install 'toolkit-seascape[notebook] @ file:///private/tmp/seascape-roadmap-evidence/wheels/toolkit_seascape-0.1.0-py3-none-any.whl'	0; declared optional extra, no runtime install redo
Development python -m pytest -q tests/test_validation_notebook.py	0; 7 passed, 0 failed/skipped; focused-tests-final.log
Development python -m pytest -q	0; 248 passed, 0 failed, 3 skipped; full-tests-final.log; 31 dependency/deprecation warnings
Consumer copied check_validation_notebook.py, chosen workspace + four sentinel overrides	0; selected-verified-executed.{ipynb,json}, selected-verified-notebook.log
Consumer copied helper, default temporary workspace + four environment variables unset	0; default-verified-executed.{ipynb,json}, default-verified-notebook.log
Consumer python -m jupyter nbconvert --to notebook --execute 01_TOOLKIT_VALIDATION.ipynb --ExecutePreprocessor.timeout=120 --ExecutePreprocessor.kernel_name=seascape-ss02 --output plain-verified-executed.ipynb --output-dir /tmp/seascape-ss02-evidence	0; cwd plain-verified-copy, explicit local kernel/PATH/JUPYTER_PATH; plain-verified-notebook.log
ruff check src tests scripts; current four-module format gate + two new Python files; python -m mypy	0 each; existing typing scope preserved; ruff-check.log, ruff-format.log, mypy.log
Consumer python -m pip check	0; consumer-pip-check.log
Existing environment snapshot helper: consumer / development	1 / 0; helper requires absent consumer pytest; consumer versions captured separately without installing pytest
Evidence source/reference inspection; git diff --check; production/config/dependency/CI/live-notebook diff guard	0 each; 22 local links, unchanged copied bytes and isolated execution verified

The three skips are two feature-catalog tests and one network-consumer test requiring absent materialized regional artifacts. Earlier notebook executions passed before final review exposed Pandas truncation of the import path. Plain-text paths were added and all final executions rerun. Two evidence-inspection trials failed on raw JSON stream-list handling and that path truncation; the normalized final inspection passed. These inspection failures were not counted as passes. Jupyter's local TCP transport warning is retained in logs; no external connections were authorized.

Files: AGENTS.md, README.md, docs/demo.md, notebooks/README.md, notebooks/validation/01_TOOLKIT_VALIDATION.ipynb, scripts/check_validation_notebook.py, tests/test_validation_notebook.py, this progress record. No production API, formula, units/sign, missingness, schema, dependency declaration, default configuration or scientific/release gate change. No unrelated initial edits existed. Production source, canonical products/retained releases, other notebooks, graph cache and sibling repositories were preserved; all demo artifacts remain in demo-owned subtrees. Remote changes: none.

No implementation/acceptance blocker remains. Linux, other interpreters, hosted consumer CI, live providers, regional scientific validity, downstream integration and a human newcomer trial remain not_run. Security/dependency audits already recorded in SS-01 were not rerun for this notebook-only task. Next: SS-03 — clean consumer-install acceptance. Stop before SS-03.

SS-03 clean consumer-install acceptance — blocked on hosted execution

Implementation complete; local acceptance passed. Full task acceptance remains blocked by the unexecuted hosted matrix. Base: c1cab8b44990f9e86d4a4bac26d88bddab0eb30a, clean tree on feature/seascape-first-run-demo. This record's containing commit is the implementation handoff; final Git identity is in /tmp/seascape-ss03-evidence/final-git-state.json. Reverified SS-01/SS-02 APIs, resource loading, notebook source/hash and existing wheel/CI helpers. The inherited-site-package CI wheel check was still present; no prior task or baseline was reset.

• Replaced that check with Linux Python 3.11/3.14 x86_64 and macOS Python 3.14 arm64 (macos-14) consumer jobs. They build an isolated sdist, build the wheel from that sdist, inspect critical files/all packaged resources, then install normally into a fresh venv with no inherited/user site packages. Added only the declared optional build tools; runtime requirements unchanged.
• Runtime layer runs dependency checks, 161 installed-module imports with OrcaCast blocked, packaged-resource checks, the actual installed console script's help/init/26-stage dry run, and the existing guarded demo before pytest/Jupyter or the notebook file is present.
• Four expected nonzero probes passed: import-denied Rasterio (simulated dependency absence), attempted source import rejected, actual installed config removal/restoration rejected by the resource checker, and injected outbound demo activity rejected with a retained FAIL report. Source config and ordinary/canonical/release directories are not used for these probes.
• Only after runtime PASS, installed declared test/notebook extras; 55 copied tests passed (0 failed/skipped, 25 warnings) and the notebook-only copy passed 14 production / 15 notebook checks, 2 figures, no acquisition/outbound attempts, restored environment and retained artifacts. Its source hash remains f618fc5df6bc2d68eeda842e6e22f135da80827dde85b1adac1382a1b3c545b3.
• Python process guards self-test denied network/checkout access and deny child escapes. Jupyter uses the existing loopback-aware kernel guard. This is not an OS/native-extension firewall. CI asserts observed interpreter/system/architecture and uploads concise evidence on failure or success, including hidden demo files, without uploading the consumer venv. Existing notebook/security jobs and quality checks are retained; format targets add only this patch.

Local environment: macOS 26.6.2, Darwin 25.6.0, arm64, Python 3.14.6; GDAL 3.12.4, PROJ 9.8.1, GEOS 3.13.1. Separate runtime/extras inventories and exact consumer argv/cwd/expected/actual exits: acceptance-verified/{runtime-environment.json, extras-environment.json,report.json} under /tmp/seascape-ss03-evidence/. Host commands/exits: commands.json; archive hashes/resource evidence: distribution-report.json.

Executed command / evidence	Exit / result
Development python -m pip install 'build>=1.2'	0; newly declared build tool; existing setuptools/wheel already present
python -m build --outdir /tmp/seascape-ss03-evidence/distributions	0; isolated sdist then wheel from sdist; distribution-build.log
python scripts/check_distribution.py --sdist …tar.gz --wheel …whl --output …/distribution-report.json	0; 9 required files and all 5 packaged resources preserved
python -m pytest -q tests/test_consumer_acceptance.py	0; 11 passed / 0 failed / 0 skipped; focused-tests-verified.log
python -m pytest -q	0; 259 passed / 0 failed / 3 skipped; 31 warnings; full-tests-verified.log
python scripts/check_consumer_install.py --wheel …whl --source <checkout> --forbid-root <MarineCast> --output …/acceptance-verified	0; all 16 steps met expected exits; consumer-run-verified.log, consumer report.json
ruff check src tests scripts; existing four-module + seven changed Python-file format gate; python -m mypy	0 each; existing four-module typing scope retained
Final YAML parse / all shell steps bash -n, guide reference checks, security/quality preservation, copied-file/hash checks, git diff --check	0; evidence review-check.json, source-reference-ci-check.json

The same two feature-catalog tests and one network-consumer test skip for absent materialized regional artifacts. Earlier consumer runs exited 1: first the helper incorrectly used python -m seascape.cli (no module entry point), then its outbound probe patched an unbound pipeline alias. Init-file and expected-denial assertions caught both. Corrected console dispatch and bound-call injection have regressions; fresh final acceptance passed. Earlier failures remain in acceptance/ and acceptance-final/; they were not relabeled or counted as complete passes.

Files: .github/workflows/ci.yml, AGENTS.md, pyproject.toml, docs/{DEVELOPMENT,demo}.md, notebooks/README.md, scripts/{check_installed_package,check_demo,check_distribution, check_consumer_install,check_consumer_failures,consumer_guard}.py, tests/test_consumer_acceptance.py, this progress record. Scientific/public API behavior changed: no; formulas, missingness, schemas, release gates, production source, configurations, notebook source, canonical products/retained releases and siblings preserved. No unrelated initial edits. No live acquisition, remote mutation, release publication or history rewrite.

Hosted Linux 3.11/3.14 and macOS-14 3.14 arm64 jobs: not_run. No local Linux/container runtime or Python 3.11 was available; new hosted jobs cannot execute without an authorized remote update. Local macOS-26 evidence does not establish the hosted macOS-14 job. Security/dependency audits were not rerun; their existing CI gates remain unchanged. Regional/live/downstream/human checks remain outside scope. Next: SS-04 — read-only prerequisite checks, with SS-03 hosted acceptance still pending before integration. Stop before SS-04. Remote changes: none.

SS-04 read-only prerequisite checks — passed

Base: 368ee1b2400415a298dd742f4d6cad3fe4990af8; the containing implementation commit is this handoff (exact final identity also in /tmp/seascape-ss04-evidence/final-git-state.json). Clean initial tree; retained feature/seascape-first-run-demo. SS-00 prerequisites reverified against current CLI, stage registry, family loaders, candidate rebasing and scientific contracts. SS-03 implementation is present; its hosted acceptance remains blocked, not relabeled passed.

• Added build --dry-run --check-inputs [--json]; flags require --dry-run. Plain dry-run stays unchanged. JSON-only planning explicitly reports inspection not_run; checked failures emit parseable JSON and exit 1. Report schema 1 includes workspace/config/candidate, stage expansion, configured/default destinations, publication intent, checks, required/optional status and actions.
• Execution and preflight share plan_domain_layer_build and candidate configuration rendering. Existing family loaders inspect a context-local copy in memory; common/workspace overrides are restored. No second stage registry, data hashes, acquisition, builder, cleanup or candidate writes. Exact upstream output paths are generated_by_plan; absent external inputs remain distinct.
• Existing CRS/unit/sign/H3/scale validation is reused. Native raster header checks were factored out of the slope function and reused without reading pixels; formulas, thresholds, masks, defaults, schemas and release validation remain unchanged. Header inspection accepts local TIFF signatures/GTiff only, disables GDAL auxiliary writes and rejects disguised remote VRT sources.
• Readability is not schema/coverage validation. Vector/Parquet headers, source values, datum, checksum/reuse identity, directory/archive usability and release approval remain unverified. Required skip reuse, kelp annual-layer usability and reef partial-inventory usability fail as unverified; optional limits cannot turn them into PASS. All 26 packaged stages report their selected prerequisites without execution. Configured fluvial barrier dependencies are included.
• 43 focused passed / 0 failed / 0 skipped (34 new plus 9 existing workflow tests); final full suite 293 passed / 0 failed / 3 skipped, 65 dependency/deprecation warnings. Same two feature-catalog and one network-consumer tests skip for absent materialized regional artifacts.
• Final wheel built from its sdist, required resources compared, then force-reinstalled into the fresh runtime-only consumer (initial normal wheel install resolved all declared dependencies; no inherited site packages). Guarded imports: 162 modules, OrcaCast/source access blocked. Ready/missing cases passed with pytest/Jupyter absent and imports blocked, Python outbound/child guards active, zero Python write attempts, identical workspace bytes and environment after each call. Installed console help/negative JSON case also ran. This is Python guard evidence, not an OS/native firewall. Fixtures explicitly label synthetic path/header acceptance only.
• Validation notebook/demo/configuration resources are unchanged; this additive planning command needs no thin-client notebook change. Their existing production regressions ran in the full suite; copied-notebook/headless/SS-03 matrix acceptance was not redone. No real data or release created.

Commands below ran in the owning checkout, with /tmp/seascape-roadmap-dev/bin as the development interpreter/tool prefix. Evidence root E=/tmp/seascape-ss04-evidence; C=$E/consumer/bin/python. These prefixes abbreviate absolute paths only. Final static argv/exit records are in static-commands.json; installed-console argv/cwd/expected/actual exits in consumer-commands-final.json.

Command	Exit / evidence
python -m pytest -q tests/test_preflight.py tests/test_workflow.py	0; focused-tests-final.log, 43 passed
python -m pytest -q	0; full-tests-final.log, 293 passed / 3 skipped
ruff check src tests scripts	0
python -m mypy	0; existing 4-module scope
ruff format --check src/seascape/products.py src/seascape/core/geo/crs.py src/seascape/core/artifacts/confinement.py src/seascape/seafloor_physiography/depth.py src/seascape/preflight.py src/seascape/cli.py src/seascape/core/config/data.py tests/test_preflight.py	0; 8 files
python scripts/environment_snapshot.py --output /tmp/seascape-ss04-evidence/environment	0; environment.{json,txt}
pip-audit --disable-pip --no-deps --strict -r /tmp/seascape-ss04-evidence/environment.txt	0; no known vulnerabilities; dependency-audit.log
gitleaks git . --redact --no-banner --log-opts=--all	0; 13 base commits; gitleaks-history.log
gitleaks dir . --redact --no-banner	1; ignored local cache finding, below
gitleaks dir . --redact --no-banner --report-format json --report-path /tmp/seascape-ss04-evidence/gitleaks-tree-report.json	1; redacted diagnostic
gitleaks dir /tmp/seascape-ss04-evidence/tracked-scan --redact --no-banner	0; current tracked source plus the two new task files, copied read-only; gitleaks-tracked.log
python -m build --no-isolation --outdir /tmp/seascape-ss04-evidence/distributions-final	0; sdist then wheel from sdist using installed declared build tools; build-final.log
python scripts/check_distribution.py --wheel /tmp/seascape-ss04-evidence/distributions-final/toolkit_seascape-0.1.0-py3-none-any.whl --sdist /tmp/seascape-ss04-evidence/distributions-final/toolkit_seascape-0.1.0.tar.gz --output /tmp/seascape-ss04-evidence/distribution-final.json	0; 9 required files, 5 resources
python -m venv /tmp/seascape-ss04-evidence/consumer then $C -m pip install /tmp/seascape-ss04-evidence/distributions/toolkit_seascape-0.1.0-py3-none-any.whl	0; fresh consumer, install.log
$C -m pip install --force-reinstall --no-deps /tmp/seascape-ss04-evidence/distributions-final/toolkit_seascape-0.1.0-py3-none-any.whl	0; final wheel, install-final.log
$C -m pip check	0; pip-check-final.log
$C $E/consumer_guard.py --forbid-root /Users/tylerstevenson/Documents/Code_Repos/MarineCast --script $E/check_installed_package.py -- --snapshot $E/consumer-environment-final.json	0; imports-final.log, 162 imports
$C $E/consumer_guard.py --forbid-root /Users/tylerstevenson/Documents/Code_Repos/MarineCast --script $E/preflight_acceptance_final.py	0; consumer-preflight-final.log, ready/missing reports
$C $E/consumer_guard.py --forbid-root /Users/tylerstevenson/Documents/Code_Repos/MarineCast --script $E/consumer/bin/seascape -- --workspace $E/workspace-final build --only seascape-geomorphometry --dry-run --check-inputs --json	1 expected; missing raster, parseable JSON, console-missing.log
$C $E/consumer_guard.py --forbid-root /Users/tylerstevenson/Documents/Code_Repos/MarineCast --script $E/consumer/bin/seascape -- build --help	0; both flags exposed, console-help.log
git diff --check	0

Consumer commands clear PYTHONPATH/PYTHONHOME and set PYTHONDONTWRITEBYTECODE=1; scripts are copied outside the checkout. Python 3.14.6, macOS 26.6.2 arm64; GDAL 3.12.4, PROJ 9.8.1, GEOS 3.13.1. Final wheel SHA-256 f68ef744a9adea081280205cd4393ec4cd411a3390889b6e9ce688a99cc96867; sdist 0f32cbf1ecc871acb81747eeddf40731b07f32293b647c6e5db3d0a36fa802f8. An earlier focused run exited 1 (one assertion counted the newly added destination-access check as a source input: 41 passed / 1 failed); corrected assertion scopes source inputs. Initial archive helper invocation omitted required --output (exit 2); corrected inspection passed. Earlier 34/42-test runs and initial wheel acceptance passed, but final counts/artifact above supersede them.

Broad directory secret scan is failed, not passed: one generic-api-key finding at ignored graphify-out/cache/stat-index.json:1, a 64-hex documentation hash under docs/API.md -> hashes -> docs/api.md. This pre-existing developer cache is unmodified. The tracked-source scan passed; no security gate or ignore/allowlist was weakened. Its redacted report is local evidence, not a published artifact. Hosted Linux/macOS SS-03 checks remain not_run because remote updates are not authorized. Regional/live/downstream/human acceptance is not_run.

Changed files: AGENTS.md, docs/{API,WORKFLOWS}.md, src/seascape/{cli,workflow,preflight}.py, src/seascape/core/config/data.py, src/seascape/seafloor_physiography/geomorphometry/build.py, tests/test_preflight.py, this progress record. Scientific behavior changed: no. Additive CLI planning/report behavior; existing producer/consumer signatures and numerical behavior preserved. No unrelated tracked edits existed; configurations, canonical products, retained releases, notebook source, ignored cache and sibling repositories preserved. Next: SS-05, CLI diagnostics and side-effect contracts. Stop before SS-05. Remote changes: none.

SS-05 CLI diagnostics and operation effects

passed (local implementation/acceptance). Base 7ebdb999d80ecce934169a4cb1b52de107be245b; current commit is the SS-05 commit containing this record. Preserved feature/seascape-first-run-demo. SS-01/SS-04 prerequisites were satisfied; source reinspection confirmed family forwarding/restoration, existing config validators, stage re-raise behavior, candidate runners disabling acquisition, immutable release checks and transaction recovery.

Identified CLI failures now report operation/stage, reason, nonsecret path/setting, corrective action and guide on stderr, with exit 1. Translation uses known validation sites/resource failure types; unlisted calculation/programming failures still raise. Global --debug before the command exposes original chained tracebacks. Family help (0), parser errors (2), nonzero family returns, workspace/argv restoration and Python exception types/causes remain intact. An internal workflow callback changes only CLI stage-failure presentation; normal API re-raise/results are unchanged. JSON preflight schema/stdout are unchanged; failure guidance is additive stderr. The workflow guide now lists prerequisites, network/write/replacement/reuse behavior and candidate/family/whole-release effects, including direct-family and inspection hazards.

Final verification below ran in the owning checkout unless the command uses the copied helpers under E (those ran from E, outside the checkout). Exact command prefixes:

D=/tmp/seascape-roadmap-dev/bin
R=/tmp/seascape-ss04-evidence/consumer/bin
E=/tmp/seascape-ss05-evidence
F=/Users/tylerstevenson/Documents/Code_Repos/MarineCast
$D/python -m pytest -q tests/test_cli_diagnostics.py tests/test_products.py tests/test_demo.py tests/test_preflight.py tests/test_workflow.py
$D/python -m pytest -q
$D/ruff check src tests scripts
$D/python -m mypy
$D/ruff format --check src/seascape/products.py src/seascape/core/geo/crs.py src/seascape/core/artifacts/confinement.py src/seascape/seafloor_physiography/depth.py src/seascape/cli.py src/seascape/_cli_diagnostics.py tests/test_cli_diagnostics.py
$D/python -m build --no-isolation --outdir $E/distributions-final
$D/python scripts/check_distribution.py --sdist $E/distributions-final/toolkit_seascape-0.1.0.tar.gz --wheel $E/distributions-final/toolkit_seascape-0.1.0-py3-none-any.whl --output $E/distribution-final.json
$R/python -m pip install --force-reinstall --no-deps $E/distributions-final/toolkit_seascape-0.1.0-py3-none-any.whl
$R/python -m pip check
$R/python $E/consumer_guard.py --forbid-root $F --script $E/check_installed_package.py -- --snapshot $E/installed-final.json
$R/python $E/consumer_guard.py --forbid-root $F --script $E/cli_acceptance_final.py
$R/python $E/check_demo.py --workspace $E/demo-workspace-final --forbid-root $F
$D/python scripts/environment_snapshot.py --output $E/environment
$D/pip-audit --disable-pip --no-deps --strict -r $E/environment.txt
git diff --check
gitleaks dir $E/tracked-scan --redact --no-banner
gitleaks git . --redact --no-banner --log-opts=--all

All final commands above exited 0. Focused 103 passed / 0 failed / 0 skipped; full 322 passed / 0 failed / 3 skipped, 65 warnings. Skips: two feature-catalog and one network-consumer test require absent materialized regional artifacts. Added 29 regressions. Lint, four-module mypy and seven-file format checks passed; no repository-wide formatting. Build used provisioned declared build tools, producing wheel from sdist. Distribution inspection passed (9 required files, 5 resources). Final wheel SHA-256 e7b947f0cf62ccc7b538c1e4e061ef097d22d8dd877e94f806977058e4d841e4; sdist e356f47837efff0e52869f56fc21eee692348c34e50dd9f4bc1b9aed13aeb307.

Runtime-only acceptance reused the isolated SS-04 consumer after force-reinstalling this wheel; no inherited site packages, pytest or Jupyter supplied dependencies. Guarded imports: 163 modules. Copied CLI helper: 13 cases with exit/JSON/diagnostic/restoration/output-preservation checks. Copied demo: 14 production acceptance checks, synthetic provenance, demo-owned artifacts only. Python checkout/outbound/child guards passed; these are not an OS/native-extension firewall claim. Console wrapper commands using $R/python $E/consumer_guard.py --forbid-root $F --script $R/seascape -- also ran: --help exit 0; --workspace $E/cli-workspace --debug inspect bathymetry --config missing.yaml exit 1; --workspace $E/cli-workspace build --only seascape-bathymetry --dry-run --check-inputs --json exit 1. Failure exits were expected; JSON parsed and debug stderr contained the original traceback with empty stdout.

Evidence: /tmp/seascape-ss05-evidence/ (commands.json, final test/build/import/demo logs, CLI cases, distribution, installed/environment snapshots, audit/security logs; initial intermediate logs also retained). Python 3.14.6/macOS ARM64; GDAL 3.12.4, PROJ 9.8.1, GEOS 3.13.1. Audit exit 0, no known vulnerabilities. Initial focused runs exited 1 (4 failed/83 passed, then 1 failed/99 passed): test API calls needed the selected workspace for relative includes and the original CRSError type for an invalid CRS. Corrected tests passed. Intermediate focused/full checks also passed (102/321 respectively); final review added a publication-path refusal regression and revalidated the final source/wheel. One read-only inspection shell typo (/ tmp) exited 126, no mutation.

Broad gitleaks dir . --redact --no-banner --report-format json --report-path /tmp/seascape-ss05-evidence/gitleaks-tree.json exited 1: the same ignored, untouched graphify-out/cache/stat-index.json:1 generic-api-key finding, a 64-hex documentation cache hash at docs/API.md -> hashes -> docs/api.md. Broad scan failed, not waived/passed. Tracked-source and local-history scans passed; no ignore/allowlist/gate weakened. Debug/provider tracebacks can contain sensitive detail. Unlisted validation failures remain diagnosable tracebacks. Hosted SS-03 Linux/macOS checks remain not_run, remote updates unauthorized. Live acquisition, regional rebuilds, downstream integration and human release acceptance are not_run. Notebook was assessed: it still calls unchanged run_demo; hash unchanged, no edit or redundant headless run.

Changed: AGENTS.md, docs/{API,WORKFLOWS}.md, this record, src/seascape/{cli,workflow,_cli_diagnostics}.py, tests/{test_cli_diagnostics,test_products}.py. Scientific behavior changed: no; formulas, defaults, product/release schemas, Python exception contracts and scientific/release gates retained. No unrelated tracked changes existed. Existing configurations, canonical/retained products, notebook, ignored cache and siblings preserved. Next: SS-06, repository-wide formatting only. Stop before SS-06. Remote changes: none.

SS-06 formatting baseline

passed (local implementation/acceptance). Base 818df353a418ee3ec963466228c6589851599e72; current commit is the SS-06 commit containing this record. Clean starting tree; preserved feature/seascape-first-run-demo. SS-00 prerequisite complete; source edits were serialized. Rechecked existing Ruff configuration, CI gates, source-byte code identity and prior acceptance limits. Used Ruff 0.16.9, existing defaults/Python 3.11 target, without additional exclusions.

Formatted 124 Python files under src/, tests/, scripts/; no lint autofixes, import reordering, symbol/type changes, moves or algorithm edits. All 209 Python ASTs match exactly after ignoring location metadata (including TypeIgnore.lineno), retaining type comments and string/docstring values. All 56 protected file hashes match, including configuration, resources, schemas and notebook. All 209 files also parse with Python 3.14's feature_version=(3,11); this is syntax evidence, not execution on Python 3.11. CI and contributor instructions now enforce ruff format --check src tests scripts. A structural YAML comparison confirms the CI format command is the only workflow change; other test/scientific/security/consumer gates are retained.

Exact verification commands (owning checkout, except copied helpers run from E):

D=/tmp/seascape-roadmap-dev/bin
R=/tmp/seascape-ss04-evidence/consumer/bin
E=/tmp/seascape-ss06-evidence
F=/Users/tylerstevenson/Documents/Code_Repos/MarineCast
$D/ruff format src tests scripts
$D/python $E/verify_format.py
$D/ruff format --check src tests scripts
$D/ruff check src tests scripts
$D/python -m mypy
$D/python -m pytest -q
$D/python -m build --no-isolation --outdir $E/distributions
$D/python scripts/check_distribution.py --sdist $E/distributions/toolkit_seascape-0.1.0.tar.gz --wheel $E/distributions/toolkit_seascape-0.1.0-py3-none-any.whl --output $E/distribution.json
$R/python -m pip install --force-reinstall --no-deps $E/distributions/toolkit_seascape-0.1.0-py3-none-any.whl
$R/python -m pip check
$R/python $E/consumer_guard.py --forbid-root $F --script $E/check_installed_package.py -- --snapshot $E/installed.json
$R/python $E/check_demo.py --workspace $E/demo-workspace --forbid-root $F
$D/python scripts/environment_snapshot.py --output $E/environment
$D/pip-audit --disable-pip --no-deps --strict -r $E/environment.txt
gitleaks dir $E/tracked-scan --redact --no-banner
gitleaks git . --redact --no-banner --log-opts=--all
git diff --check

Commands above exited 0. Full suite: 322 passed / 0 failed / 3 skipped, 65 warnings. Two feature-catalog and one network-consumer test skip because materialized regional artifacts are absent. No new tests or changed assertions; test ASTs are identical. Full format check: 230 files already formatted (209 tracked Python files); existing four-module mypy and bug lint passed. Initial ruff format --check src tests scripts exited 1, reporting 124 files needing formatting; corrected by the formatter. One inspection command ended with a stray / tmp (exit 126, read-only typo); expected absent old evidence-path probes were not acceptance.

Wheel built from sdist and inspected (9 required files, 5 resources). SHA-256: wheel cc60508ccc2c7db7115cba67b38b7cf90dc86bcd391cb89f1b50b2a1b259bb21, sdist 5f55b80dbbf99be8a74084d2fe61976f34196eade487688e13d496b3165d42e4. Reused isolated runtime-only SS-04 consumer after force-reinstalling this wheel; no inherited packages, pytest or Jupyter dependencies. Guarded 163 module imports and 14 demo checks passed; synthetic artifacts only in E/demo-workspace/.seascape/demo. Guards establish Python checkout/outbound/child-process denial, not an OS/native firewall. Dependency audit found no known vulnerabilities. Python 3.14.6/macOS ARM64; GDAL 3.12.4, PROJ 9.8.1, GEOS 3.13.1.

Evidence: /tmp/seascape-ss06-evidence/ (command ledger, before/after ASTs, verification-final.json, protected hashes, scope/CI verification, changed-file inventory and test/build/runtime/security logs). Broad gitleaks dir . --redact --no-banner --report-format json --report-path /tmp/seascape-ss06-evidence/gitleaks-tree.json exited 1 on the same untouched ignored graphify-out/cache/stat-index.json:1 documentation-hash finding. Broad scan failed; tracked source/history passed. No cache deletion, allowlist or weakened gate. Hosted SS-03 Linux/macOS checks remain not_run (remote updates unauthorized). Notebook assessed: unchanged file and production API ASTs; no edit or repeated headless run. Live acquisition, regional rebuilds, downstream integration and human release acceptance remain not_run.

Changed files: 124 Python files (complete inventory in evidence), .github/workflows/ci.yml, AGENTS.md, docs/DEVELOPMENT.md, docs/environments/README.md, this progress record. Behavior/API/scientific changes: none. Source bytes/code identity change, so existing resume state may invalidate; validation is preserved. Ruff remains range-declared; version reproducibility and broader lint/typing belong to SS-07. No unrelated tracked changes existed. Existing configs, canonical products, retained releases, notebook, ignored cache and siblings preserved. Next: SS-07, static checks and environment reproducibility. Stop before SS-07. Remote changes: none.

SS-07 static checks and reproducibility

passed (local SS-07 implementation/static/reproducibility acceptance; broad tree secret scan still failed). Base 7f2832e086e5dda513ebf5bd207a56d38fa0dbd5; current commit is the SS-07 commit containing this record. Clean starting tree; retained feature/seascape-first-run-demo. Reverified SS-06, current package/CI gates, public interfaces and snapshot traversal against source.

Added narrow Ruff F401,F841,F811,F601,F602,B006,I001 rules, preserving prior rules and formatting. Corrected 52 import-order findings and two unused test imports; no exclusions/suppressions added. Mypy now checks 12 documented interface modules, retaining product/CRS/confinement/depth targets. Added demo fixture and preflight schema-1 types, generator/resource-copy annotations, and separate CLI result locals. Existing exception, serialized report, scientific/product/release contracts stay unchanged. Only external NumPy stubs are skipped explicitly: installed stubs contain Python 3.12+ syntax incompatible with the retained 3.11 type target. Existing imported-code/missing-stub limits remain documented; scientific array internals are not claimed strictly typed.

Extended the existing snapshot's explicit extras/marker traversal, including transitive extras, cycles and missing-active-distribution failure. It records versions/native/platform evidence, omitting package URLs, executable/install/compiler paths; invalid versions raise a sanitized error. Six new regressions cover closure, extras, cycles, markers, privacy and failure. Added separate runtime/test/quality constraints/JSON; retained historical evidence and runtime metadata ranges. CI quality now exercises Linux 3.11 unconstrained and macOS ARM64 3.14 constrained installs, compares baseline closure/native versions, and audits both quality closures. Other CI jobs/settings are structurally unchanged. The expanded audit found bootstrap pip 26.1.2 affected by PYSEC-2026-3721 (two advisory rows, one package). Added quality-only pip>=26.2, upgraded to 26.2.1, rebuilt and exercised two new isolated installs; final audits have no known vulnerabilities. No vulnerability exclusions, vulnerable downgrades or new runtime dependencies.

Exact final acceptance commands, in this checkout unless noted:

D=/tmp/seascape-roadmap-dev/bin
R=/tmp/seascape-ss04-evidence/consumer/bin
E=/tmp/seascape-ss07-evidence
F=/Users/tylerstevenson/Documents/Code_Repos/MarineCast
$D/python scripts/environment_snapshot.py --extra test --extra quality --output docs/environments/quality-python314-macos-arm64
$D/python -m build --no-isolation --outdir $E/final-distributions
$D/python -m venv $E/baseline
$D/python -m venv $E/range
$E/baseline/bin/python -m pip install --cache-dir $E/pip-cache --retries 1 -c docs/environments/quality-python314-macos-arm64.txt "$E/final-distributions/toolkit_seascape-0.1.0-py3-none-any.whl[test,quality]"
$E/range/bin/python -m pip install --cache-dir $E/pip-cache-unconstrained --retries 1 "$E/final-distributions/toolkit_seascape-0.1.0-py3-none-any.whl[test,quality]"
# Each command below ran in both baseline and range; actual argv/exits are in *-commands.json.
for V in "$E/baseline/bin" "$E/range/bin"; do
  "$V/python" -m pip check
  "$V/ruff" check src tests scripts
  "$V/ruff" format --check src tests scripts
  "$V/python" -m mypy
  "$V/python" -m pytest -q
  "$V/python" scripts/environment_snapshot.py --extra test --extra quality --output "$E/$(basename "$(dirname "$V")")-environment"
  "$V/pip-audit" --disable-pip --no-deps --strict -r "$E/$(basename "$(dirname "$V")")-environment.txt"
done
$D/python scripts/check_distribution.py --sdist $E/final-distributions/toolkit_seascape-0.1.0.tar.gz --wheel $E/final-distributions/toolkit_seascape-0.1.0-py3-none-any.whl --output $E/final-distribution.json
$R/python -m pip install --no-cache-dir --force-reinstall --no-deps $E/final-distributions/toolkit_seascape-0.1.0-py3-none-any.whl
$R/python -m pip check
# Copied helpers below ran from E, outside checkout; source/overrides cleared by the helpers.
$R/python $E/consumer_guard.py --forbid-root $F --script $E/check_installed_package.py -- --snapshot $E/final-installed.json
$R/python $E/check_demo.py --workspace $E/final-demo-workspace --forbid-root $F
gitleaks dir $E/tracked-scan --redact --no-banner
gitleaks git . --redact --no-banner --log-opts=--all
git diff --check

Final commands above exited 0. Each final installed-wheel suite: 328 passed / 0 failed / 3 skipped, 65 warnings (baseline 96.44s; range 101.27s). Two feature-catalog and one network test require absent regional artifacts. Initial development full suite also 328/0/3; focused suite 103/0/0. Static checks: 231 formatted files; 12 typed modules. Both isolated environments use CPython 3.14.6, macOS 26.6.2 ARM64, GDAL 3.12.4, PROJ 9.8.1, GEOS 3.13.1, Ruff 0.16.9, mypy 2.3.1, pip-audit 2.10.1 and pip 26.2.1. Exact constrained snapshot matches the tracked baseline; range solve differs only in platformdirs 4.12.0 vs 4.11.15. No inherited site packages. Quality environments intentionally include pytest; separate reused runtime-only consumer has no pytest/Jupyter packages. Guarded 163 imports and 14 demo checks pass on the final wheel; no network attempts, two denied font-discovery child attempts. Synthetic artifacts stay in E/final-demo-workspace/.seascape/demo. Guards are Python-level denial, not a native/OS firewall.

Final sdist SHA-256 79b0a3951d131ac3f071588ec7d8390d6c678ce583eb27b01eebd4e121fca67b; wheel ac3c4a2959b34acfa66e4b8d867e812dd5dd1f241558f61ae02760afcdbc6b21. Distribution inspection passed (9 required files, 5 resources). Reviewed 45 import-only changed Python ASTs with imports removed; other source changes are annotations/types and CLI local names. 14 protected config/resource/notebook files match HEAD; tracked Python parses with 3.11 syntax (which is not execution on 3.11). Notebook assessed: unchanged notebook/public runtime behavior; no edit or repeated Jupyter run. Source-byte/resume identities may change; validators retained.

Failures retained in evidence: exploratory unused/import lint exit 1 (54 findings), initial expanded mypy exit 2 (NumPy stub syntax), then exit 1 (five CLI result-type errors), corrected; first constrained install exit 1 (sandbox PyPI DNS; network-enabled retry passed); initial closure-equality probe exit 1 (expected unconstrained platformdirs drift, corrected comparison). Initial pre-security constrained/range audits exited 1, then final fresh fixed installs/audits passed. One helper-copy/import batch ran from the wrong cwd and exited 2 (missing copied helper); corrected owning-cwd copy and outside-cwd acceptance passed. A patch-tool invocation was rejected for duplicate operations on one file before any edit; corrected. These are not passed checks.

gitleaks dir . --redact --no-banner --report-format json --report-path /tmp/seascape-ss07-evidence/gitleaks-tree.json exited 1: the same untouched ignored graphify-out/cache/stat-index.json:1 generic-api-key documentation-hash finding. Broad scan failed; tracked-tree/history scans passed. No cache deletion/allowlist/gate weakening. Hosted Linux/macOS acceptance and new quality matrix remain not_run: remote changes unauthorized. Live data, regional builds, downstream integration and human usability/release trials not_run.

Changed files: 59 files; complete inventory in E/changed-files.txt. Functional/type/tooling changes: pyproject.toml, CI, scripts/environment_snapshot.py, src/seascape/{cli,demo,preflight}.py, core/config/data.py, new tests/test_environment_snapshot.py, new quality constraints/JSON, AGENTS.md, README.md, environment guide and this progress record. Remaining Python changes are import cleanup only. Evidence: /tmp/seascape-ss07-evidence/ (final per-environment command JSON/logs, solve/native snapshots/difference, full/audit/build/distribution/runtime/security/scope evidence). Behavior/API: stronger static contracts and reproducible development tooling; snapshot gains explicit extras and sanitized evidence; runtime Python APIs, CLI outputs and report schemas retained. Scientific behavior changed: no. No unrelated starting edits existed; configs, packaged resources, canonical/retained products, notebook, ignored cache and siblings preserved. Next: SS-08, pending hosted SS-03 prerequisite; stop before SS-08. Remote changes: none.

SS-03 acceptance closure recheck — blocked

Selected the earliest incomplete task with satisfied prerequisites: SS-03 (SS-01/SS-02 passed). Base 721a7199436ece870cceea17b3b2e312378e24ac; current commit is this documentation handoff. Clean starting tree; retained feature/seascape-first-run-demo. Read current instructions, roadmap/progress, consumer CI, check_consumer_install.py and its regression tests. Runtime-first installation, copied outside-checkout helpers, negative probes, declared test/notebook layer, three Linux/macOS interpreter/architecture cases and scientific/security gates remain implemented. No independent unfinished local implementation was identified; verified work was not repeated.

Read-only GitHub checks at 2026-09-27 13:19 UTC:

• github_fetch_commit for current 721a7199436ece870cceea17b3b2e312378e24ac and SS-03 368ee1b2400415a298dd742f4d6cad3fe4990af8: HTTP 422, "No commit found for SHA" for both.
• github_search_branches(owner=MarineCast, repo_name=toolkit-seascape, query=feature/seascape-first-run-demo): successful read, no matching branches.
• Direct GET https://api.github.com/repos/MarineCast/toolkit-seascape/actions/runs?head_sha=721a7199436ece870cceea17b3b2e312378e24ac&per_page=100 and the same endpoint with head_sha=368ee1b2400415a298dd742f4d6cad3fe4990af8: both successful reads, total_count=0. These queries include all trigger types; an empty pull-request-only wrapper was not used to infer absence of other runs.
• GET https://api.github.com/repos/MarineCast/toolkit-seascape/branches/main: successful read, remote main remains f2400c13d509ad753d9168ed4e7a07d1ffcfc5a4.

Precise blocker: the consumer workflow changes are local; the required hosted Linux 3.11/3.14 x86_64 and macOS 3.14 ARM64 execution/results do not exist for these commits. Executing them requires an explicitly authorized remote update followed by inspection of real job results. Current scope forbids push/remote mutations; no dispatch, rerun, PR, settings change or remote update was performed. SS-08 depends on SS-03; SS-09 onward likewise remain ineligible. Local macOS success and configured jobs do not satisfy the missing hosted evidence.

Commands: git status --short, git branch --show-current, git rev-parse HEAD, git remote -v, git log -10 --oneline, git branch -vv, reference/diff checks and git diff --check: exit 0. command -v docker, command -v podman, command -v python3.11: exit 1 each (not on PATH). /tmp/seascape-roadmap-dev/bin/python -c 'import platform; print(platform.python_version(), platform.system(), platform.machine())': exit 0, Python 3.14.6 / Darwin / arm64. Hosted acceptance not_run; pytest/build/demo/ notebook/audits not_run this recheck, preserving previously verified evidence rather than claiming fresh passes. Tests this turn: not_run, no new pass/fail/skip counts.

Changed: this progress record only. Behavior/API/scientific changes: none. No unrelated starting changes; production source, CI, tests, configs, resources, notebooks, products/releases, ignored caches and sibling repositories preserved. Evidence: /tmp/seascape-ss03-recheck-evidence/. Next: finish SS-03 hosted acceptance, then SS-08; stop before SS-08. Remote changes: none.

SS-03 hosted acceptance closure — passed

User explicitly authorized the feature-branch push to test SS-03 and publish the local code changes to Git. Preserved all earlier commits on feature/seascape-repository-organization-updates. Base/tested commit 6ddd5ca7d4ce25b229bf0dd50bbd3e3708afc7f8; current commit is this documentation handoff. Clean starting tree. No new production, CI, configuration, test or scientific changes. This result supersedes the earlier blocked SS-03 entries without relabeling those historical runs.

Run 36322829881, triggered by push on September 27, completed success: 9/9 jobs. Inspected job conclusions and decoded logs; consumer environment JSON and expected-exit results are printed in those logs. All three consumer jobs built/inspected an sdist and its wheel, passed 16/16 consumer steps, verified 163 installed modules, and asserted the observed interpreter/system/architecture:

Actual consumer	GDAL / PROJ / GEOS	Job ID
Ubuntu 24.04.5, Linux x86_64, Python 3.11.16	3.10.3 / 9.5.1 / 3.13.1	108629839889
Ubuntu 24.04.5, Linux x86_64, Python 3.14.7	3.12.4 / 9.8.1 / 3.13.1	108629839917
macOS 14.8.9, Darwin arm64, Python 3.14.7	3.12.4 / 9.8.1 / 3.13.1	108629839834

Each consumer installed the wheel with dependencies in a fresh, non-inherited runtime environment; pip/import/resource/help/init/dry-run/demo checks passed before test/notebook extras. Four expected negative probes per case exited 1 with required markers (missing Rasterio, source import, missing packaged resource, attempted outbound activity); these are successful rejection tests, not concealed failures. Runtime demo, copied external tests and copied notebook steps exited 0. Python process guards do not claim OS/native isolation or regional scientific validity.

Both Linux full-suite jobs: 328 passed / 0 failed / 3 skipped, 65 warnings each. Skips remain two feature-catalog and one network-consumer test requiring absent materialized regional products. The separate notebook job passed. Both quality jobs passed lint, 231-file format checks, 12-module mypy, dependency checks and advisory audits (no known vulnerabilities). The constrained macOS job compared dependency/native versions successfully; interpreter patch/OS differences from the local baseline are recorded above. Hosted Gitleaks history and broad fresh-checkout scans passed. Earlier local ignored-cache findings remain historical failed scans, not waived.

Exact commands/actions: git push --set-upstream origin feature/seascape-repository-organization-updates (exit 0); git rev-parse HEAD origin/feature/seascape-repository-organization-updates (exit 0, matching SHA). Read-only GitHub run/jobs/logs/artifacts requests succeeded. Executed CI commands are retained verbatim in .github/workflows/ci.yml and the downloaded decoded job logs. Documentation reference checks, git diff --check, commit and documentation push exited 0. No additional local pytest/build/notebook execution was needed for this documentation-only closure.

Five non-expired Actions artifacts exist: three consumer bundles and two scientific-environment snapshots. Recorded IDs/digests/expiry in /tmp/seascape-ss03-hosted-evidence/run-36322829881.json; decoded logs are job-<id>.log there. Artifact ZIP contents were not downloaded: nested copied-test counts and figure visual QA are not newly claimed. Reports/logs/distributions remain available in Actions artifacts. Native versions and all 16 expected-exit results were verified from job logs.

Changed: this record, docs/environments/README.md, docs/DEVELOPMENT.md. Behavior/API/scientific changes: none. No unrelated starting edits; source, CI, tests, configs, resources, notebooks, canonical/retained products, ignored caches and siblings preserved. Remote changes: authorized feature-branch pushes only; no PR, merge, tag, settings change, package publication or dataset acquisition. Live/regional/downstream/human release acceptance remain not_run. Next: SS-08, now eligible; stop before SS-08.

SS-08 scientific, publication and export acceptance — passed

Base e3e831bfc63ef5578838308b2e82322d26d74dac; current commit is the SS-08 commit containing this record. Branch feature/seascape-repository-organization-updates; initial tree clean. Rechecked roadmap, progress, root/owner AGENTS, contracts, source and named tests. Read-only GitHub lookup reconfirmed prerequisite run 36323147359 succeeded on the base commit. SS-03/SS-05 and the SS-07 static boundary are satisfied; no prior implementation repeated.

Contract map: P = present at base, N = missing evidence added/extended here. All names below executed in the focused/full suite; test quantity is not a scientific certification.

Contract	Named evidence	Classification
Production transformation, family/checksum validation, constant field, nodata/unavailable, observed zero, nonempty support, sign/CRS, explicit synthetic provenance	test_normal_run_nonempty_expected_values_and_provenance, test_empty_or_all_null_output_cannot_pass, test_arbitrary_cwd_environment_restoration_and_confinement	P; first test extended N for gradient
Analytic gradient and finite-value rejection	test_demo_rejects_corrupted_gradient_values, test_demo_finite_policy_rejects_infinity	N
Keyed row invariance, identical support/cardinality, unique/non-null exact-resolution keys	test_composite_is_invariant_to_each_input_order, test_composite_rejects_incompatible_identity; matrix test_matrix_rejects_duplicate_h3_keys, test_matrix_rejects_invalid_identity_without_replacing_output, test_matrix_row_order_invariance_and_collision_free_field_names	P composite/duplicate; N matrix boundaries
Depth sign/missingness, metric projected axes, Q90; flat/coastal nodata and controlled plane slope	test_depth_values_reject_contradiction_and_preserve_missingness, test_scientific_configuration_rejects_nonmetric_crs, test_terrain_rejects_unsupported_depth_sign_before_io, test_q90_schema_rejects_other_quantiles, test_native_slope_masks_land_nodata_and_preserves_flat_edges, test_native_slope_matches_north_south_plane_including_edges	P
Actual publication A/B, retained historical A, manifest-last ordering, interrupted recovery/rollback	test_two_publications_retain_prior_product_and_manifest_bytes, test_generation_is_rolled_back_when_canonical_promotion_fails, test_release_publisher_promotes_manifest_last, test_snapshot_recovers_interrupted_release_before_read, test_release_rolls_back_failure_at_each_promotion	P
Exact resolution, tampering, unsafe paths, invalid release IDs	test_missing_product_and_resolution_fail_without_fallback, test_checksum_mismatch_and_incomplete_release_fail, test_historical_release_id_is_validated, test_product_resolver_rejects_paths_outside_generation, test_matrix_rejects_unsafe_catalog_paths	P parser checks; N unsafe product/catalog cases
Resume invalidation after included config, package identity and file-backed source changes	test_resume_invalidates_when_included_domain_config_changes, test_resume_invalidates_when_package_code_identity_changes, test_resume_invalidates_when_file_backed_source_changes	P
Export freezes A while actual publisher installs B; table, catalog and checksum metadata cannot mix	test_matrix_pins_real_publication_during_release_switch	N, existing pinning passed unchanged
Null/zero, QC/evidence, units, original types, field names, explicit legacy limits	test_matrix_aligns_by_h3_and_preserves_zero_null_and_qc, test_matrix_row_order_invariance_and_collision_free_field_names	P values/metadata; N shuffled/collision boundaries
Empty support, failed write/replace, existing destinations, aliases/overwrite and retained generations	test_matrix_rejects_empty_support_without_resolution_column, test_matrix_failed_write_preserves_valid_destination, test_matrix_output_cannot_replace_input, release-switch test; test_existing_export_is_preserved	N except existing CLI refusal P

Reproduced failures and fix: initial matrix run: 3 failed / 17 passed, exit 1. Empty tables without a resolution column were accepted; overwrite could replace the input catalog or canonical release manifest. Export now rejects empty support and those aliases, and cannot write into any retained generation. Python signatures, output schema, physical values and exception types remain unchanged; these invalid inputs now raise ValueError. Initial collection exit 2 was a test fixture import error, fixed by making copied tests a package and reusing the existing publisher fixture. Initial focused exit 1 (108 passed / 2 failed / 16 fixture errors) exposed an arithmetic mistake in the new analytic expectation; corrected the row sum from 452 to 453, without changing calculations.

Demo adds gradient to report controls and one real check (15 total), using the original raster, production bathymetry API, packaged config and family validator. The 18 interior pixels have sum(row)=453, sum(column)=426; d=5+(145*c+80*r)/47 gives mean 5680/47, min 5280/47, max 6085/47, range 805/47 meters. Absolute 3e-5 m accounts for float32 fixture rounding; constant depth retains 1e-6 m and band partitions 1e-12, no relative tolerance. Existing slope plane assertions retain NumPy's tolerance for double-precision angular-to-meter arithmetic. Same-environment keyed tables repeat exactly; timestamps/run IDs/plotting bytes are not compared. The notebook remains a thin client; only its explanation changed. Scientific behavior changed: no. Publisher tests reuse a test-only pre-audited fixture; they exercise publication/resolution, not a full regional audit. The one-family demo is never promoted or labeled a complete release.

Executed commands/results (all from this checkout; exact expanded argv, exit codes and logs in /tmp/seascape-ss08-evidence/commands.json; consumer subcommands in acceptance/report.json):

Command	Result
/tmp/seascape-roadmap-dev/bin/python -m pytest -q	exit 0, 352 passed / 0 failed / 3 skipped, 69 warnings
Focused python -m pytest -q over demo/matrix/products/review/workflow/consumer/publication/release/bathymetry contracts	exit 0, 126 passed, 29 warnings; exact file list in focused-final.json
ruff check src tests scripts; ruff format --check src tests scripts; python -m mypy	exit 0 each; 232 formatted files, 12 checked interface modules
python -m pip check	exit 0; no broken requirements
python -m build --no-isolation --outdir /tmp/seascape-ss08-evidence/distributions	exit 0; wheel built from sdist, declared provisioned build tools
python scripts/check_distribution.py --sdist … --wheel … --output …	exit 0, 9 required files / 5 resources survive; full args in distribution.json
python scripts/check_consumer_install.py --wheel … --source … --forbid-root … --output …	exit 0, 16/16 expected-exit steps; fresh isolated runtime first, then declared extras
Copied guarded pytest; copied headless notebook	exit 0 each; 83 passed / 0 skipped, 29 warnings; notebook 15 production / 16 total checks, 2 embedded figures, no outbound/acquisition attempts
python scripts/environment_snapshot.py --extra test --extra quality --output /tmp/seascape-ss08-evidence/environment	exit 0, Python/native/dependency closure recorded
pip-audit --disable-pip --no-deps --strict -r /tmp/seascape-ss08-evidence/environment.txt	exit 0, no known vulnerabilities; no ignore/downgrade
gitleaks dir /tmp/seascape-ss08-evidence/tracked-review --redact --no-banner; git diff --check	exit 0 each; tracked/new source scan only, ignored cache preserved

Environment: fresh consumer and development CPython 3.14.6, macOS 26.6.2 ARM64; GDAL 3.12.4, PROJ 9.8.1, GEOS 3.13.1. Source/OrcaCast reads, outbound Python activity and child processes denied during consumer execution; installations permit declared dependency downloads and notebook loopback traffic. No OS firewall claim. Wheel SHA-256 and distribution checks are in distribution-report.json; portable execution/logs/notebook in acceptance/.

The three skips remain the two feature-catalog checks and one network-consumer check requiring absent regional artifacts. No new skips. Prior broad ignored-cache Gitleaks finding is not waived or relabeled passed; that scan was not repeated. Real acquisition, regional equality/accuracy, downstream integration, visual figure QA and human onboarding trials not_run. Hosted run 36324502530 passed 9/9 jobs on implementation commit aace671d74c565b4f141805dbe66505fb055f239: Linux Python 3.11 and 3.14 each 352 passed / 3 skipped, 69 warnings; all three clean consumer cases (Linux 3.11/3.14, macOS ARM64 3.14) 16/16 expected-exit steps; notebook, both quality jobs and history/tree secret scans passed. Decoded job logs and exact tested SHA are in /tmp/seascape-ss08-evidence/hosted/. Hosted copied-test counts/artifact interiors were not downloaded or inspected; their executed steps passed. This progress-only closure changes no code. Initial unrelated tracked changes: none; configs, canonical/retained products, caches and sibling repositories preserved. Files: demo/matrix APIs, consumer helper, three contract test files plus test-package marker, demo/matrix guides, notebook explanation and this record. Remote actions: normal authorized feature push of implementation and this progress closure; no merge, tag, package publication, settings or dataset changes. Next: SS-09; stop before it.

SS-09 three user journeys — acceptance passed

September 27. Base 0d307de978c98358af6dbfce31a034dc80bed91c; implementation 3022ace6b6a868c65e56f0efb6ddf8a9f84bb9d9 on feature/seascape-repository-organization-updates. Clean starting tree. Re-read owner/ancestor instructions, roadmap/progress and actual CLI/config, preflight, packaging and publication fixtures. SS-08 closure run 36324758385 was rechecked completed/success; no reset, source acquisition or prior task reimplementation.

README now leads with purpose, a verified source-install route and the actual synthetic PNG. Demo, bounded candidate processing and audited-release consumption have separate prerequisites. Configuration edits name real keys; output schema/units/sign/counts/null-zero/QC and failure limits are explicit. The consumer example freezes one release ID and reads its retained R6 product. Dated migration/remediation evidence is labeled; the original generated product block is intact. A 26-stage input reference comes from existing preflight metadata in disposable templates. Offline CI checks local links/headings, command/stage/family names, catalog-reference drift and four packaged YAML pairs. Three-platform CI also runs the exact marked quickstart blocks. No new producer framework, formula, public API, product schema or dependency declaration.

Evidence: /tmp/seascape-ss09-evidence; exact argv/cwd/exits in *.command.json, corresponding logs, distribution reports, quickstart-accepted/report.json, consumer/report.json, environment snapshot and preservation.json. Local CPython 3.14.6, macOS 26.6.2 ARM64, GDAL 3.12.4, PROJ 9.8.1, GEOS 3.13.1; no inherited site packages in new consumer/source-install environments. For commands below, D=/tmp/seascape-roadmap-dev/bin, E=/tmp/seascape-ss09-evidence, R=/Users/tylerstevenson/Documents/Code_Repos/MarineCast/Toolkits/toolkit-seascape, G=/Users/tylerstevenson/Documents/Code_Repos/MarineCast (absolute paths, no placeholders).

Exact check command	Result
$D/python scripts/check_docs.py	0; 39 documents, 139 local links, 26 stages; external URL availability deliberately unrun
$D/python -m pytest -q tests/test_documentation.py tests/test_consumer_acceptance.py	0; 27 passed (16 new documentation cases plus 11 existing consumer cases)
$D/python -m pytest -q	0; 368 passed / 3 skipped / 69 warnings
$D/ruff check src tests scripts	0
$D/ruff format --check src tests scripts	0; 235 files
$D/python -m mypy	0; 12 interface modules
$D/python -m pip check	0
$D/python scripts/environment_snapshot.py --extra test --extra quality --output $E/environment	0
$D/pip-audit --disable-pip --no-deps --strict -r $E/environment.txt	0; no known vulnerabilities
$D/python -m build --no-isolation --outdir $E/accepted-distributions	0; sdist then wheel from sdist
$D/python scripts/check_distribution.py --sdist $E/accepted-distributions/toolkit_seascape-0.1.0.tar.gz --wheel $E/accepted-distributions/toolkit_seascape-0.1.0-py3-none-any.whl --output $E/distribution-accepted-report.json	0; 9 required files / 5 packaged resources
$D/python scripts/check_quickstart.py --sdist $E/accepted-distributions/toolkit_seascape-0.1.0.tar.gz --source $R --forbid-root $G --output $E/quickstart-accepted	0; all 12 documented shell lines; 15 production demo checks; normal runtime-only installation, both source trees guarded
$D/python scripts/check_consumer_install.py --wheel $E/distributions/toolkit_seascape-0.1.0-py3-none-any.whl --source $R --forbid-root $G --output $E/consumer	0; 16/16 expected-exit steps; 163 imports, 83 copied tests passed, copied notebook 15 production/16 total checks and 2 embedded figures
/opt/homebrew/bin/gitleaks dir /tmp/seascape-ss09-evidence/tracked-tree --redact --no-banner	0; current tracked files only; ignored local cache excluded explicitly
git diff --check / git diff --cached --check	0

The accepted wheel differs from the clean-consumer-tested wheel only in README-derived METADATA and RECORD; every packaged code/resource byte matches (preservation.json). All 196 tracked source/config/package/notebook files under protected paths equal the base commit. The README PNG is copied from SS-08's actual demo; visual inspection confirmed its synthetic title, units and grey unavailable cells. This is illustration QA, not regional scientific validation.

Initial docs-check failures (prose mistaken for a CLI command, stale TODO link) and one import-order lint failure were corrected, with succeeding checks above. The first source quickstart passed; a stricter rerun failed its evidence-directory assertion although all shell commands exited zero. Using an explicit mktemp template corrected that portability issue; quickstart-accepted passed. Failed logs/reports are retained, never relabeled PASS. Distribution hashes are in the reports. Guarded installed CLI family help returned 0; the documented bathymetry input preflight returned the expected 1 (missing_external/unverified), not a ready/PASS release. Exact interpreter, runner, workspace and arguments are in family-help.command.json and missing-input-preflight.command.json. The broad local ignored-cache Gitleaks false positive remains the SS-07/SS-08 limitation; no cache cleanup or allowlist change. Current tracked-tree/history and hosted results are recorded below.

Three regional artifact skips remain test_feature_catalog.py:107,130 and test_network_consumer_contracts.py:207; inputs are intentionally absent. No live sources, regional rebuild/comparison/audit, full real-data release, downstream application integration, Windows-native publication or unfamiliar-human usability trial ran. Python guards are not an OS firewall; dependency installation permits index access. Longer processing/publishing/export examples require their explicitly documented existing inputs and authorization.

Changed: .github/workflows/ci.yml, AGENTS.md, README.md, docs/{API,CONFIGURATION,DEVELOPMENT, MIGRATION,README,WORKFLOWS,demo,hardening-review,metric-matrix,products,review-remediation}.md, docs/environments/README.md, docs/stage-inputs.md, docs/assets/demo-bathymetry.png, notebooks/README.md, scripts/{check_docs,check_quickstart}.py, tests/test_documentation.py, this record. Unrelated work, configurations, canonical products, retained releases, source calculations and notebook bytes are preserved; no sibling edits. Next: SS-10, requiring explicit approved inputs, AOI, network and resource limits. Stop before it. Feature-branch commit/push use the user's existing authorization and message Repository Organization Updates; no merge/tag/package publication/settings/data changes. Hosted evidence is recorded below.

SS-09 implementation 3022ace6b6a868c65e56f0efb6ddf8a9f84bb9d9 was committed and pushed with message Repository Organization Updates (both commands exit 0). GitHub run 36327449869 completed success, 9/9 jobs. Linux Python 3.11.16 and 3.14.7 each ran 368 passed / 3 skipped / 69 warnings, and doc checks each reported 39 documents / 139 local links / 26 stages. All three consumers completed 16 expected-exit wheel steps plus all 12 source-quickstart lines with 15 demo checks: Linux x86_64 3.11.16 (GDAL 3.10.3 / PROJ 9.5.1 / GEOS 3.13.1), Linux x86_64 3.14.7 and macOS ARM64 3.14.7 (GDAL 3.12.4 / PROJ 9.8.1 / GEOS 3.13.1). Both quality jobs, original notebook and history/working-tree secret scan jobs passed. Hosted copied-test/notebook steps passed; their artifact ZIPs were not downloaded, so their internal counts are not separately asserted here. Actual run/jobs JSON, five decoded job logs and hosted-acceptance-summary.json are under the evidence root. Local gitleaks git . --redact --no-banner --log-opts=--all also returned 0 (22 commits). The current handoff commit is this evidence record's containing documentation-only commit; its identity, final status and remote synchronization are saved in final-git-state.json. Documentation-only closure checks: python scripts/check_docs.py and git diff --check both returned 0; no Python/scientific implementation changed after the successful hosted run. No real-data or SS-10 work was performed. The existing feature-branch push authorization was reused.

SS-10 bounded San Juan pilot — blocked execution, runbook complete

September 27. Base 515e60de133fcb4346d976a683978f45c2d64101; current implementation is this record's containing commit on feature/seascape-repository-organization-updates. Clean starting tree. SS-09 closure run 36327837094 was re-read for that exact base: completed/success. Reverified owner instructions, actual explorer, helper, production bathymetry/downloader/publication code, preflight contracts and cached inputs. No earlier task reimplementation or reset.

The bounded runbook is SS-10's explicit blocked-runbook alternative. It contains the existing San Juan bounds/R6/R8 recipe, pinned source identities and rights, CRS/datum/alignment/sign/nodata, proposed limits, exact configuration-only/preflight commands, future numerical/visual acceptance and resource-measurement limitations. The real-data gate is unsatisfied; execution, output validation, measurements, repeatability and map QA are not_run. No scientific formula, API, config default, source/retained manifest or notebook changed.

Local GEBCO 2026 and Natural Earth 5.1.1 inputs exist; safe local reads require no acquisition. All seven historical family-manifest artifact checks matched production checksum_path. The 46-file cache is 22,717,798 bytes. This is inventory/identity evidence, not scientific acceptance. Official GEBCO 2026/terms and Natural Earth terms pages were read; no provider metadata queue, source download, extraction, pixel processing or redistribution was performed.

Execution blockers: (1) production source/attribution metadata and the historical manifest say CC BY 4.0, conflicting with the GeoTIFF and current GEBCO public-domain terms; the producer needs a focused correction/regression before new trusted manifests; retained history remains untouched; (2) six canonical water inputs are absent in the fresh workspace, while the Natural Earth proxy cannot be relabeled canonical or used to bypass reuse validation; (3) family publication cannot be disabled in this pipeline, and SS-10 step 5's explicit disposable-workspace publication/resource scope is not approved. The runbook makes a concrete bounded proposal; no unbounded default ran.

Local evidence: /tmp/seascape-ss10-evidence. CPython 3.14.6, macOS 26.6.2 ARM64, GDAL 3.12.4, PROJ 9.8.1, GEOS 3.13.1; existing isolated development environment, no dependency changes. For commands below, D=/tmp/seascape-roadmap-dev/bin, E=/tmp/seascape-ss10-evidence, W=$E/accepted-preflight-workspace; cwd is this owning checkout. Preparation's exact Python code/argv is saved in accepted-preparation.command.json; preflight argv/exits are in accepted-preflight.command.json. Both initial and final preparation reports are retained; both returned 0, and both guarded preflights returned the expected 1, not a ready/PASS build.

Executed acceptance command	Exit / result
$D/python -c <exact runbook preparation code in accepted-preparation.command.json> with SEASCAPE_PILOT_ROOT=$W	0; bbox/R6/R8/workers/sign/output confinement and cached raster checksum assertions passed
$D/python scripts/consumer_guard.py --forbid-root $W/forbidden --module seascape -- --workspace $W build --only seascape-bathymetry --dry-run --check-inputs --json --candidate-root $W/candidate with common/candidate overrides unset	1, expected block; six missing_external, source raster ready, intermediates generated_by_plan, resume/publication not_applicable
$D/python scripts/check_docs.py	0; 40 documents / 148 local links / 26 stages; external URL availability not checked by this offline tool
$D/python -m pytest -q tests/test_documentation.py tests/test_preflight.py tests/test_workflow.py	0; 59 passed / 0 failed / 0 skipped, 34 warnings
$D/python -m pytest -q -rs	0; 368 passed / 0 failed / 3 skipped, 69 warnings
$D/python scripts/environment_snapshot.py --extra test --extra quality --output $E/environment	0
git diff --check	0

Final preparation's production resolved-config hash is 21245ccfbf85677c8da6e9054c9aee32419c87e4d118e6967f3b5c1e76ee934c. Its archived resolved config includes machine paths; those reports stay untracked. This hash differs from history and another fresh workspace because paths participate in production hashing. preflight-acceptance.json confirms all configured outputs under the fresh candidate and no candidate/artifacts created. The guard deliberately allows checkout/cache reads and denies Python outbound/child calls; it is not an OS firewall or evidence of installed-wheel isolation.

cached-input-inventory.json and preservation.json record all 46 cached files byte-identical and all 257 tracked files outside docs/ byte-identical. No unrelated initial edits existed; existing configs, canonical products, retained releases, explorer outputs and sibling repositories are preserved. Three suite skips remain test_feature_catalog.py:107,130 and test_network_consumer_contracts.py:207: clean-checkout regional artifacts intentionally absent. No new executable source/tests: existing docs/preflight/workflow regressions and executed recipe provide the relevant acceptance. Fresh wheel/consumer/notebook/static/security gates were not rerun for these documentation-only edits; the successful exact-base SS-09 closure is prior evidence.

Changed files: docs/pilots/san-juan.md, docs/README.md, this record. Feature-branch commit/push use existing authorization and message Repository Organization Updates; final identities and remote synchronization are saved in final-git-state.json and the handoff. No merge, tag, package publication, settings, dataset or release changes. Next eligible task: SS-11, explicitly limited to an offline research candidate unless SS-10's execution gates are resolved first. Stop before SS-11.

SS-11 release-candidate review — owner gates pending

September 27. Base and tested candidate: ee41524b92ea6a6ef643488f903a1d69248e6d8b. Clean initial tree; retained feature/seascape-repository-organization-updates. Current handoff commit is this record's containing documentation-only commit, separate from the tested artifact identity. Re-read instructions/roadmap/progress and actual CI, metadata, distribution/consumer helpers and scientific/publication/export evidence. SS-11 prerequisites are satisfied for an offline research candidate with SS-10's blocked real-data path explicit; no prior task rewritten.

Prepared candidate notes and maintainer checklist: capabilities, fixed export/installation defects, additive interfaces and stricter invalid-export validation, scientific invariants, known limits, artifact hashes, version decision and go/no-go gates. No code/config/schema/default/dependency/version change. No scientific calculation changed. Review implementation is complete; human/version/classic-protection/release-owner gates are blocked/pending, not PASS. The owner explicitly replied that no unfamiliar tester is available; that human trial is pending, not replaced by automated acceptance. SS-10 remains blocked, with no acquisition/real processing, measured operating envelope, regional maps or release audit.

Candidate GitHub run 36329187229 was freshly inspected: completed/success, nine jobs, exact candidate SHA. Both test jobs ran 368 passed / 3 skipped / 69 warnings. Three clean consumers built/inspected sdist-derived wheels, ran all 16 expected-exit runtime/extras steps, copied notebook and documented source quickstart. Actual runtime records: Linux x86_64 Python 3.11.16 (GDAL 3.10.3 / PROJ 9.5.1 / GEOS 3.13.1), Linux x86_64 3.14.7 and macOS ARM64 3.14.7 (GDAL 3.12.4 / PROJ 9.8.1 / GEOS 3.13.1). Both quality jobs, original notebook and history/tree secret jobs passed. This is verified exact-SHA hosted evidence, not a new workflow dispatch or a transfer of the local wheel to Linux. Artifact ZIPs were not downloaded; hosted copied-test/notebook internal counts are not independently asserted. Nine decoded logs are retained (consumer logs are explicitly bounded tails).

Local evidence: /tmp/seascape-ss11-evidence, *.command.json with exact argv/cwd/exits and corresponding stdout/stderr; distribution/candidate reports, consumer logs/reports/notebook, source-quickstart reports, environment, API observations and preservation evidence. Existing development CPython 3.14.6, Darwin ARM64; fresh build/consumer/source-install environments have no inherited site packages. Local GDAL 3.12.4 / PROJ 9.8.1 / GEOS 3.13.1. Installation can reach declared dependency indexes; runtime is Python-guarded, not native/OS-firewalled.

Commands below ran from this owning checkout. D=/tmp/seascape-roadmap-dev/bin, E=/tmp/seascape-ss11-evidence, B=$E/build-env/bin, R is the absolute owning checkout and G the absolute MarineCast workspace. Exact expanded paths are in the command JSON files; machine paths/logs remain untracked.

Executed command	Exit / result
$D/python -m venv $E/build-env; $B/python -m pip install '.[build]'	0 each; clean declared-dependency build environment
$B/python -m build --outdir $E/distributions	0; isolated sdist, then isolated wheel from sdist; setuptools 84.0.0 / wheel 0.48.0
$B/python scripts/check_distribution.py --sdist $E/distributions/toolkit_seascape-0.1.0.tar.gz --wheel $E/distributions/toolkit_seascape-0.1.0-py3-none-any.whl --output $E/distribution-report.json	0; 9 required files / 5 resources; archive/metadata/source-byte checks passed
$B/python scripts/check_consumer_install.py --wheel $E/distributions/toolkit_seascape-0.1.0-py3-none-any.whl --source $R --forbid-root $G --output $E/consumer	0; 16 expected-exit steps, 83 copied tests passed, copied notebook 15 production / 16 total checks, two embedded figures
$B/python scripts/check_quickstart.py --sdist $E/distributions/toolkit_seascape-0.1.0.tar.gz --source $R --forbid-root $G --output $E/quickstart	0; exact install/demo/plan blocks; 15 synthetic demo checks; automated simulation only
$D/python -m pytest -q -rs	0; 368 passed / 0 failed / 3 skipped, 69 warnings; includes scientific/publication/release/matrix regressions
$D/ruff check src tests scripts; $D/ruff format --check src tests scripts; $D/python -m mypy	0 each; 235 formatted files / 12 interface modules
$D/python scripts/environment_snapshot.py --extra test --extra quality --output $E/environment	0
$D/pip-audit --disable-pip --no-deps --strict -r $E/environment.txt	0; no known vulnerabilities
/opt/homebrew/bin/gitleaks git . --redact --no-banner --log-opts=--all	0; 24 commits
/opt/homebrew/bin/gitleaks dir $E/tracked-tree --redact --no-banner	0; tracked candidate snapshot only; local ignored cache excluded explicitly
$D/python scripts/check_docs.py; git diff --check	0; final 41 documents / 152 local links / 26 stages; external URL availability not checked by this offline tool
$D/python -m pytest -q tests/test_documentation.py	0; 16 passed / 0 failed / 0 skipped on the final handoff documentation
/opt/homebrew/bin/gitleaks dir $E/handoff-tracked-tree --redact --no-banner	0; intended final tracked documentation/source snapshot, including new handoff; ignored caches excluded explicitly

Local wheel SHA-256 463aa070c2a4d5d22d8383eb55b66b3bc4b453b13c78dbce78cdc7248c062df3; sdist a7ba7fa13feb35a6f6c5c244bf05629b6c6e925f3ff6ebe134e3a2727abea8bd. Metadata: toolkit-seascape 0.1.0, Python >=3.11, Apache-2.0. All packaged source/resource bytes match the exact candidate; no datasets/retained products/graph caches entered either archive. Both synthetic demo PNGs were opened and inspected for titles, orientation, sign/meters and grey unavailable support. This is synthetic illustration QA, not regional coastline/data validation.

Three full-suite skips are still test_feature_catalog.py:107,130 and test_network_consumer_contracts.py:207, requiring absent regional materialized products. The older broad ignored-cache Gitleaks failure remains historical/unresolved; no cleanup or allowlist change. Current tracked/history scans and clean hosted tree scans pass their stated scopes.

Administration inspection: repository rulesets including parents returned []; the connector rejected the effective-branch endpoint as unsupported, then a public read-only GET to that endpoint returned HTTP 200 / []. Classic main protection returned integration 403, so that check is blocked and empty rulesets are not proof of absent classic protection. Repository metadata enables squash/merge/rebase, disables auto-merge; these are options, not a chosen merge policy. Nine actual observed check names are listed in the handoff; no settings change was proposed/applied. No version policy/tags were found. Existing tested 0.1.0 is retained as the proposed base; public version/prerelease policy needs owner approval, with rebuild/revalidation if metadata changes.

Changed: docs/release-candidate.md, docs/README.md, this record. All other tracked files and all 46 cached explorer files remain byte-identical; configs, scientific code, canonical products, retained releases, notebook and siblings preserved. Detailed final preservation and Git state are saved under the evidence root. No unrelated starting edits. Normal feature commit/push use existing authorization and message Repository Organization Updates; no merge/tag/registry, release upload, settings, dataset or application changes. Next: owner resolution of the pending SS-11 gates; SS-10 execution requires its separate scope/correction/input gates. No SS-12 started.

Focused review fixes (September 27)

Base ec78b57878c4a93b28892132df7c1edf900d6da3, clean existing feature branch; no reset. FIX-01 changes future source/attribution rights only: GEBCO public-domain terms/acknowledgement reference, synthetic fixture licensing unchanged, unknown-provider rights explicitly unverified. Official terms/release documentation rechecked; runtime metadata needs no network. Production fixture tests verify release/checksum, identical keyed values/nulls/resolution and retained publisher-created manifest/product bytes. Historical source caches/manifests remain untouched. python -m pytest -q tests/test_demo.py tests/domains/environment/seascape/test_bathymetry_contracts.py: exit 0, 32 passed, 27 warnings. Initial suggested top-level bathymetry test path did not exist: exit 4, no tests run; corrected to the actual nested path. Logs/exact argv are in /tmp/seascape-review-fixes-evidence/{fix01,fix01-focused}.command.json and matching logs. FIX-02 pre-change reproduction: malformed YAML and fixed-Q90 violation each exit 1 in text, JSON and debug+JSON, but useful details and tracebacks are absent. Reports saved under the same evidence root; debug files stay local/private. FIX-02 now retains safe reason/type and document/setting context with additive optional schema-1 fields. Existing classifier/sanitizer reused; malformed YAML excludes snippets, quoted user values are redacted while fixed scientific enum requirements remain. Explicit debug receives the original caught exception on stderr; unfiltered debug logs remain private. Unknown defects re-raise, and preflight remains read-only with environment restoration. No scientific validation rule changed. python -m pytest -q tests/test_preflight.py tests/test_cli_diagnostics.py tests/test_workflow.py tests/test_products.py: exit 0, 113 passed, 62 warnings (fix02-confirm logs). Earlier regression development runs exit 1 (12 then 10 failures): omitted existing positive-down classifier and a test mutation of an unused field, both corrected without weakening validation. Text/JSON/debug tests cover both caught paths, YAML/include/reference/Q90, secrets, legacy fields and unexpected exceptions. Actual Q90 JSON exit 1 supplies the workflow example. Notebook stays unchanged: its existing production-demo thin-client coverage remains appropriate; copied installed-wheel execution follows in VERIFY-01.

VERIFY-01 tested 50675db2bef48d3e48de56ec02a5c98a992808ca. Separate fixes: f98a36b (metadata), 9755f94 (diagnostics). README pins immutable 9755f94, containing both fixes; all 189 package files at that revision and the tested child match the new wheel exactly. The child adds the pin and two existing test modules to external consumer acceptance. No notebook refactor or scientific change. A later repository-docs-only closure refers to this exact candidate. Evidence root: /tmp/seascape-review-fixes-evidence; {name}.command.json stores exact argv, cwd, cleared environment and exit code; matching stdout/stderr, consumer/quickstart reports and nine decoded hosted logs retain details. Debug reproduction logs stay private, excluded from Git. Development CPython 3.14.6 / macOS 26.6.2 ARM64; GDAL 3.12.4, PROJ 9.8.1, GEOS 3.13.1.

Executed check (from owner checkout unless helper specifies outside cwd)	Exit / actual result
python -m pytest -q tests/test_demo.py tests/domains/environment/seascape/test_bathymetry_contracts.py tests/test_preflight.py tests/test_cli_diagnostics.py tests/test_documentation.py tests/test_consumer_acceptance.py tests/test_products.py tests/test_metric_matrix.py tests/test_review_regressions.py tests/test_workflow.py	0; 216 passed / 0 failed / 0 skipped, 93 warnings (verify-focused)
python -m pytest -q	0; 398 passed / 0 failed / 3 skipped, 99 warnings (verify-suite)
ruff check src tests scripts; ruff format --check src tests scripts; python -m mypy	0 each; 235 formatted files, 12 typed modules
python scripts/check_docs.py; python -m pip check; git diff --check	0 each; 41 docs/152 local links/26 stages, external URLs offline/not checked
python scripts/environment_snapshot.py --extra test --extra quality --output /tmp/seascape-review-fixes-evidence/environment	0; exact test/quality/native closure captured
pip-audit --disable-pip --no-deps --strict -r /tmp/seascape-review-fixes-evidence/environment.txt	0; no known vulnerabilities
Existing isolated build interpreter python -m pip check, then python -m build --outdir /tmp/seascape-review-fixes-evidence/distributions	0 each; isolated sdist then wheel from sdist, setuptools 84.0.0/wheel 0.48.0 backends
Guarded installed-wheel exact metadata comparison (one-off controlled-fixture check)	0; both 2026/2025 branches exactly preserve all 153 keyed rows/values/nulls, source checksum and H3 R8. Initial attempts exit 1 on the existing /tmp symlink safeguard and an incorrect check-script key; corrected to resolved fresh workspace and actual h3_resolutions; no producer-code change
Existing scripts/check_distribution.py --sdist ... --wheel ... --output .../distribution-report.json	0; 9 required files / 5 resources; 189 package source files byte-matched independently
Existing scripts/check_consumer_install.py --wheel ... --source <owner> --forbid-root <MarineCast> --output .../consumer	0; runtime-only first, 163 imports/15 demo checks; 4 negative cases expected exit 1; extras later; 175 external tests / 0 failed / 0 skipped; copied notebook 15 production/16 notebook checks, 2 embedded figures, no acquisition/network attempts
Existing scripts/check_quickstart.py --sdist ... --source <owner> --forbid-root <MarineCast> --output .../quickstart	0; marked source install/demo/init/stages/dry-run, 15 demo checks
Gitleaks 8.30.1 git . --redact --no-banner --log-opts=--all; dir <fresh tracked-tree> --redact --no-banner	0 each; no leaks. Ignored local Graphify cache broad-scan finding remains historical; no cache deletion/allowlist change or claim that local ignored-tree scan passed
git push origin feature/seascape-repository-organization-updates	0; authorized ec78b57 → 50675db
Hosted run 36345268484	success, 9/9 jobs; both Linux full suites 398 passed/3 regional skips. Linux consumers 3.11.16/3.14.7 x86_64; macOS consumer 3.14.7 ARM64; two quality/audit jobs, copied/source notebooks, source quickstarts and clean-checkout broad secret/history scans passed

Three regional skips remain test_feature_catalog.py:107/:130 and test_network_consumer_contracts.py:207: materialized regional artifacts are absent from a clean checkout. No added skip/validation relaxation. Dependency installs use network; guarded Python computation forbids checkout access/outbound/source acquisition/children, not an OS/native firewall. New runtime demo report and both figures actually inspected: explicit SYNTHETIC, elevation/depth in meters, H3 R8, unavailable gray; 153 cells. No real-data numerical/map acceptance implied. Artifact identities are in candidate-artifacts.json and the current release note. All 46 original cache files (22,717,798 bytes), configs, canonical/retained products and other tracked files outside the 14 intended changes remain byte-identical. Retained publisher-created fixture bytes are also regression-tested. No sibling changes, reset, acquisition, merge, tag, registry/settings/data upload.

Stop after VERIFY-01. PILOT-01 still needs valid canonical inputs or the existing explicitly validated exploratory support route; source/bounds/R6/R8 identities; approved fresh workspace, resource and local-family-publication scope; two executed realizations with numerical/provenance, wall/storage/memory and inspected-map evidence. Future rights correction removes only that blocker. ACCEPT-01 still needs an unfamiliar user's install/demo/output/units-resolution/provenance/null trial (owner confirmed none available), observed friction, owner version/merge/release decisions and classic-protection review (prior integration 403, not proof of no protection). Neither gate passed.

PILOT-01 continuation — execution approval pending (September 27)

Started clean at 8eaae5fc1ed8e2359726527c528f55f45913e3fb on the existing authorized feature/seascape-repository-organization-updates; no reset/repeated roadmap implementation. Read owner/workspace instructions, current roadmap/progress/runbook and actual producers. PILOT-01 is the next incomplete task. Its existing explorer mask was explicitly exploratory, but downstream H3/network manifests hard-coded canonical territorial-water provenance. The helper now transactionally publishes a schema-3 exploratory mask manifest with actual Natural Earth release/ZIP checksum, optional-credit public-domain rights, partial completeness, cartographic scale and model-ineligible caveats. Prepared workspace support-version overrides are explicitly exploratory. The three existing production builders verify available selected-mask manifest identity and retain lineage/completeness; old San Juan masks with their explicit exploratory AREA declaration and no manifest fail before publication. Ordinary legacy absence behavior remains. Historical manifests are not rewritten. No formulas, CRS/datum/sign, nodata, quantiles, neighborhoods, H3 selection, parent bands, scientific settings or release gates changed. Offline fixtures compare all support/network products and bathymetry values exactly and reject bad lineage before writes, preserving existing destinations. No public signature/schema change; additive metadata only, plus stricter validation of declared provenance. Validation notebook's existing production-demo coverage stays appropriate; its committed bytes are unchanged.

Evidence: /private/tmp/seascape-pilot-01-readiness; command JSON retains exact argv/cwd/exits and logs. prepare.py exits 0: cached GEBCO 2026 production checksum and Natural Earth 5.1.1 ZIP identity reverified; 7,204,615 extracted ZIP bytes; raster header and valid/nodata/land/zero pixel counts recorded without writes. Fresh configuration fingerprint/hash in pilot-plan.json. No real producer or source acquisition ran. Guarded canonical preflight exits 1 with valid failed JSON and the same six missing external water inputs; no READY claim. Proposed new roots: /private/tmp/seascape-pilot-01-20260927/replicate-{1,2}. Exact existing-API phase recipe and outer process-group watchdog prepared outside Git; compile check exits 0 and unapproved runner refusal exits 1 before workspace creation. Caps proposed: zero source downloads, 32 MiB inputs including extraction, 256 MiB workspace, 10 minutes / 2 GiB sampled aggregate process-tree RSS per realization, four workers. Sampler interval 0.25 seconds; overshoot possible, not a hard OS memory reservation or native-network firewall. Actual regional monitoring/stress, two-run comparison and map generation/inspection remain not_run. SS-10 step 5's explicit disposable workspace/publication approval was requested with this concrete plan and remains pending; existing feature push/local software-check authorizations were preserved.

Executed using /tmp/seascape-roadmap-dev/bin (CPython 3.14.6, macOS 26.6.2 ARM64; GDAL 3.12.4/PROJ 9.8.1/GEOS 3.13.1):

Command	Exit / actual result
python -m pytest -q tests/test_water_geometry_provenance.py	initial 1: 9 passed/1 failed, test copy omitted logical-name checksum update; corrected, no validator relaxation; subsequent 0: 11 passed before final extra rejection regression
python -m pytest -q tests/test_water_geometry_provenance.py tests/test_demo.py tests/domains/environment/seascape/test_bathymetry_contracts.py tests/domains/environment/seascape/spatial_support/test_water_network.py	0; 55 passed/0 skipped at that development step
python -m pytest -q	final 0; 410 passed/3 existing regional skips/101 warnings, including all 12 new regressions; previous intermediate 409 passed/3 skipped
ruff check src tests scripts; ruff format --check src tests scripts; python -m mypy	0 each; 237 formatted files / 12 typed modules
jupyter nbconvert --to notebook --execute notebooks/validation/01_TOOLKIT_VALIDATION.ipynb --ExecutePreprocessor.timeout=120 --output seascape-toolkit-validation.ipynb --output-dir /private/tmp/seascape-pilot-01-readiness	sandbox 1 on loopback bind; authorized local-loopback retry 0; output copy only
python scripts/check_docs.py; python -m pip check; git diff --check	0 each; 41 documents/152 local links/26 stages, external URLs not checked by offline checker
Existing isolated build interpreter python -m build --outdir /private/tmp/seascape-pilot-01-readiness/distributions	0; sdist then wheel from sdist, setuptools 84.0.0/wheel 0.48.0
python scripts/check_distribution.py --sdist ... --wheel ... --output .../distribution-report.json	0; 9 required files/5 resources; actual full paths in command JSON

Three regional skips remain feature-catalog lines 107/130 and network-consumer line 207; no new skip. Read-only refresh of prior hosted run 36345837141 confirms success on starting 8eaae5f; that is not hosted evidence for this new change. Installed-wheel acceptance and final preservation/commit evidence follow below when actually run. PILOT-01 remains blocked only on the exact requested local execution scope for the exploratory route (canonical inputs still unavailable). Its real numerical/resource/visual gates are not passed. Next after resolving PILOT-01: ACCEPT-01; the unfamiliar tester remains unavailable/pending and owner version/merge/release/classic-protection decisions remain open. No ACCEPT-01 work started.

Final package acceptance: existing scripts/check_consumer_install.py --wheel ... --source <owner checkout> --forbid-root <MarineCast> --output .../consumer exits 0. Fresh runtime-only consumer first: 164 imports, unchanged 15-check synthetic demo, four negative cases expected exit 1; declared extras only afterward, 175 external tests passed/0 skipped, copied notebook passed with two embedded figures. The new 190-file package matches this tested working tree byte-for-byte. Wheel SHA-256 79abf3642b3b2b2dab2189f036648b45743a22e7829295a0d02b0b79c0b5e3a7; sdist ecbfa454aadff6a72e35e450ab8641c38e7ff228c691027042cdd507a2c95780. python scripts/environment_snapshot.py --extra test --extra quality --output .../environment, pip-audit --disable-pip --no-deps --strict -r .../environment.txt, Gitleaks 8.30.1 history and fresh tracked-tree scans all exit 0. Ignored-cache broad-scan finding remains historical, excluded explicitly; no cleanup/allowlist/gate change. Final docs checker exits 0 (41 documents/154 local links/26 stages), documentation tests 16 passed. No current-run source quickstart or regional producer/visual acceptance claim; previous VERIFY-01 evidence remains identified separately.

Tested package/source commit 232ac308f7f2229c2f4047ab9fe204d95b8a6bf9, message Repository Organization Updates. Exact 190 packaged files match that commit. All 274 other tracked files and the 46-file / 22,717,798-byte original cache remain byte-identical; both proposed real-pilot roots still absent. Exact commands and full handoff are in the evidence root's commands.json, HANDOFF.md, preservation.json, distribution and consumer reports. Authorized git push origin feature/seascape-repository-organization-updates could not complete: observed git-credential-osxkeychain waiting, native UI inventory confirms Mac locked and automatic unlock unavailable. Only that task's Git/helper descendants were terminated; child Git exit -15 (SIGTERM), recorder shell exit 241. Remote branch was read-only verified still at 8eaae5f. Manual Mac unlock is requested; no credential is requested in chat. No push success or new hosted CI claim. A docs-only closure records these facts; feature push can retry after unlock. No merge, tag, registry/settings changes or data upload. PILOT-01 real execution and ACCEPT-01 remain pending as stated above; do not infer either approval from successful software checks.

PILOT-01 execution — passed within approved exploratory scope (September 27)

User approved the exact prepared plan with “Let's do it”; the original plan and its approval hash are retained in /private/tmp/seascape-pilot-01-readiness/{pilot-plan,execution-approval}.json. Started clean at execution HEAD 5d9faaf83aaccffe61a7eb97acd27c01ec6c62a7; same feature branch. Production used the installed wheel for tested code 232ac308f7f2229c2f4047ab9fe204d95b8a6bf9 (SHA-256 79abf3642b3b2b2dab2189f036648b45743a22e7829295a0d02b0b79c0b5e3a7), with all 190 installed/source package files reverified byte-identical. Research helper/configs were read from this checkout. No source downloads, sibling changes or science implementation edits.

Two new owned roots /private/tmp/seascape-pilot-01-20260927/replicate-{1,2} now contain actual exploratory mask/H3/network/bathymetry family publications and offline figures. Same source pins, bounds, H3 R6/R8 and production scientific settings. Four schema-3 family manifests per run pass existing artifact/lineage validators; support, edges, neighborhoods, radius operator and reachable area pass their existing consumer validators. Canonical preflight remains failed on its six missing inputs; it was not made READY. No complete release audit/promotion, retained release change or real-data upload. All 46 original cache files / 22,717,798 bytes remain byte-identical.

Each run has 4,660 R8 cells (4,384 mean-depth values / 276 nulls) and 134 R6 cells (127 values / 7 nulls). R8 means span 1–354 m; R6 means 1.7647058824–243.4271844660 m, positive down. Source 24,024 pixels: 18,478 negative elevations, 5,515 positive, 31 zero and zero nodata; 18,226 marine samples assigned to selected support in each run. Copies retain EPSG:4326, native affine/alignment, meters, source vertical EPSG:5831 and nodata -32767. Header identity is not vertical-datum accuracy or survey-coverage certification. Depth bands preserve observed zero versus unavailable null; fractions sum to 1 within absolute 1e-12 and existing parent-band recomputation is exact. One R6 boundary cell 8628d1047ffffff has null direct mean but one child-derived composition sample: production point-assignment helpers verify that its R8 child 8828d10425fffff maps to that hierarchical parent while the same pixel maps directly to R6 8628d1057ffffff. This is the existing direct-mean / child-composition support distinction; no fill, relabeling or formula change.

All 23 Parquet tables match exactly by their declared H3 keys, including numeric/null values and geometry. The NPZ radius operator and scientific configuration also match exactly. Of 24 recorded product artifacts, 22 match byte checksums; H3_GRIDS_6 and H3_GRIDS_8 differ only in physical row order from the existing threaded producer. Key-sorted Arrow values including original geometry WKB and schema metadata match exactly. Both raw identities remain separately verified/recorded; no claim of universally bit-identical products. Path/time/run-ID/config-hash differences are retained, with only workspace paths normalized for config/source comparisons. No scientific values normalized.

Measured realization	Producer phases including preparation	Complete attempt / completion	Sampled aggregate peak RSS	Final owned workspace
1	13.3264 s	Initial 16.0276 s ended on report serialization; acceptance-only retry 2.6996 s; completion including intervention 115.5035 s	446,021,632 bytes	17,387,048 bytes
2	10.8242 s	13.5285 s, all phases exit 0	444,366,848 bytes	17,387,323 bytes

Original deadline retained for run-1 reporting retry; no cap termination. All stayed below the approved 600 s / 2 GiB sampled process-tree RSS / 256 MiB workspace / 32 MiB inputs including extraction / four workers. RSS/disk sampled every 0.25 s; each phase also retains macOS /usr/bin/time -l maximum RSS in bytes. Sampler can miss peaks/overshoot, not a hard OS memory reservation. Shared machine/OS/filesystem caches limit timing comparison and regional extrapolation. Python socket/child guard and PROJ_NETWORK=OFF used; no acquisition, but no OS/native network firewall or network-byte metering claim. Actually inspected all six source/R8/R6 PNGs with view_image: north-up extent, positive-up source versus positive-down means, meters/legend, coarse shoreline alignment and gray unavailable support. Cartographic coastline generalization is visible; no legal/navigation/model readiness or coastal accuracy claim. No live tile/CDN map opened.

Exact commands/exits and phase argv are in commands.json, replicate-*-resources*.json and HANDOFF_EXECUTION.md under the evidence root; summary, hashes, QC and visual record in execution-summary.json, repeatability.json, additional-acceptance.json, parent-availability-evidence.json and visual-inspection.json. Execution command used the existing recorder and installed interpreter: python run.py real-replicate-N python bounded.py <approved-root>. First aggregate exit 1 was pandas-NA JSON serialization in the one-off evidence script after all producer phases exited 0; reporting-only correction/retry exits 0, failed logs retained. Initial comparison exit 1 used relative source paths as absolute; next extra byte-identity comparison exit 1 exposed physical row order. Final comparison exits 0 after exact Arrow proof and explicit distinct byte identities. Additional availability check exit 1 assumed R6 mean/composition shared support; corrected check uses the actual documented distinction and production-helper evidence, exit 0. No production validator relaxed and no scientific values corrected to force PASS.

Authorized feature push exits 0 (8eaae5f → 5d9faaf), remote independently verified. First retry's automatic permission review timed out before its command ran; direct retry succeeded. Hosted run 36352109541 is completed/success, all nine jobs on exact execution HEAD: Linux 3.11/3.14 tests and consumers, macOS consumer, both quality jobs, source notebook and secrets. Prior full suite 410 passed/3 regional skips and installed-wheel 175 external tests remain identified at the tested package revision; no redundant source suite or packaging rebuild for this docs-only execution closure. No remote merge, tag, registry/settings or data publication. Stop after PILOT-01; next ACCEPT-01 remains pending on its actual unfamiliar-human evidence and owner decisions, not implied by this pilot or software CI.

Python 3.14-only support — September 29

At clean starting HEAD 49b5feb558f5749ddc2ed61e8b84cfdf4ba65267 on feature/seascape-scientific-hardening, the package advertised Python >=3.11, which explains the 3.11 test, consumer and quality jobs. Per the owner's new support decision, metadata now requires >=3.14,<3.15; test, notebook, consumer and quality CI jobs all select 3.14. Current-use README, contributor, demo and platform docs and the executable quickstart reflect that policy. Historical 3.11 runs/candidate evidence are retained as dated records. Ruff/mypy target 3.14; the old NumPy-stub override is gone. Six source files received only Python-3.14 Ruff formatting (AST-identical to HEAD); demo fixture gained type annotations, with its calculations unchanged. Package APIs, source data, canonical products, retained releases and scientific gates are unchanged.

Local interpreter: CPython 3.14.6, macOS ARM64. Evidence root: /private/tmp/seascape-python314-only.rgzcw5 (not a release or uploaded artifact).

Command/check	Result
python -m pytest -q tests/test_standalone_package.py tests/test_consumer_acceptance.py tests/test_documentation.py	exit 0; 69 passed (after metadata-fixture correction)
python -m pytest -q	exit 0; 446 passed, 3 skipped for absent regional artifacts; 110 warnings
python scripts/check_docs.py; python -m pytest -q tests/test_documentation.py	exits 0; 46 documents/198 links/26 stages; 16 passed
ruff check src tests scripts; ruff format --check src tests scripts; python -m mypy; python -m pip check; git diff --check	exits 0; 242 Python files formatted; 12 typed modules; no broken requirements or whitespace errors
python -m build --outdir <evidence>/distributions; python scripts/check_distribution.py --sdist <sdist> --wheel <wheel> --output <report>	exits 0; isolated sdist-to-wheel build; both archives declare Python 3.14 only; 9 required files and 5 resources checked
python scripts/check_consumer_install.py --wheel <wheel> --source <checkout> --forbid-root <MarineCast> --output <evidence>/consumer	exit 0; 16 expected-exit steps, runtime-only demo and negative probes, 179 external tests, copied notebook with 2 figures
python scripts/check_quickstart.py --sdist <sdist> --source <checkout> --forbid-root <MarineCast> --output <evidence>/quickstart	exit 0; documented source install/plan/demo, 15 synthetic checks
python -m pip install --dry-run --no-deps --ignore-installed --python-version 3.11 <wheel>	expected exit 1: Requires-Python excludes 3.11
Same pip dry run with --python-version 3.14	exit 0: would install 0.1.0

The first isolated build attempt exited 1 because sandbox DNS blocked PyPI build dependencies; the authorized network retry above succeeded. The first distribution inspection found setuptools' reordered equivalent metadata clauses; the inspector now compares those clauses and its regression also rejects a wheel that falsely advertises 3.11. Focused checks were rerun after this correction. Hosted CI for this exact support-policy change remains unrun until a feature push/run completes; historical 3.11 CI success is not current-support evidence. No 3.11 execution was claimed here.