Fictional offline mission-planning dossier with four parallel workstreams

Create an auditable offline mission-planning dossier for a FICTIONAL field observatory. This is a multi-file analytical deliverable, not a website. Use only the source data below, Python 3 standard library and Markdown/JSON/CSV. No network, packages, fabricated sources or operational safety advice. The four workstreams are independent: execute them in PARALLEL branches, each owning ONLY its named directory. None may import, read or modify another stream during this phase. After ALL four finish, perform a sequential integration and verification phase. Preserve the source data separately from derived results. Each stream must deliver input.json, solve.py, result.json and report.md; solve.py must reproduce its result deterministically from its input, with a --check mode that exits nonzero for an incorrect existing result. 1. packing/: Assign indivisible items to boxes X(capacity 9) and Y(capacity 10), or leave them out. Items ID:weight:value are A:6:11 B:5:10 C:4:8 D:4:7 E:3:7 F:3:6 G:2:5 H:2:4 I:1:2. A and H are mandatory. B and E may both be selected but may not share a box. Maximize total value, then minimize total selected weight. Report one optimal assignment, all omitted IDs, box loads and both objective values. Prove global optimality by exhaustive enumeration of assignments, not a greedy claim. 2. schedule/: Six nonpreemptive jobs on identical benches X,Y, integer starts >=0: A=3, B=2, C=4, D=2, E=3, F=1 time units. Precedence: A before D; B before D and E; C and D before F. C cannot start before 1. Jobs D and E share one technician and cannot overlap even on different benches. Intervals are half-open, so touching is allowed. Minimize makespan. Solve baseline and a disruption where bench Y is unavailable on [3,5). Provide a feasible assignment/start/end for every job and a checkable proof no shorter schedule exists in each case. 3. routing/: Undirected graph, positive travel times: DP=4,DQ=3,PQ=2,PR=5,QR=2,QS=6,RS=2,RT=5,ST=3,TD=6. Start/end D, visit P,Q,R,S,T at least once. Revisits are allowed; distance is sum of actual traversed edges. Find a globally shortest closed walk, then solve again with edge RS removed. Return actual edge-by-edge walks, costs and an exhaustive shortest-path-plus-order proof. Do not report a metric-closure jump as a physical edge. 4. quality/: Records (sensor,time,revision,raw): (S1,0,1,10),(S1,0,2,12),(S1,1,1,8),(S2,0,1,3),(S2,0,1,3),(S2,1,1,4),(S2,1,1,5),(S3,0,1,7),(S3,1,1,bad),(S3,2,1,12). Group by sensor,time; take highest revision only; identical repeats at that revision collapse; conflicting values at that revision quarantine the whole group. A surviving nonnumeric raw value is excluded. Calibrations S1=(raw-2)/2, S2=3*(raw+1), S3=10-raw. Inclusive valid band [0,10]. Deliver sorted cleaned.csv with sensor,time,revision,raw,calibrated,in_band; separately list conflicts, invalid records, superseded rows and duplicates in result.json. Report per-sensor retained-row count, calibrated min/max/mean, in-band count; excluded groups never enter aggregates. Integration: create README.md, dossier.md, manifest.json and verify_all.py outside those directories. Explain inputs, algorithms, optima, disruption impacts and data exclusions, without mixing independent assumptions. manifest.json lists every delivered file except itself, with SHA-256. verify_all.py must check all four results against their source inputs, detect deliberate corruption of EACH stream's result in disposable copies, and confirm the original files were not altered. Execute the verification and record real results. Final answer: concise results, file paths, verification evidence and any limitations.

Repair the publication-integrity defect in this project's existing offline mission dossier. This is a narrow maintenance task on the seeded files, not a rebuild of the four analyses. Observed defect in the previous delivered run: manifest.json names .atoma-probes.json although that internal journal is intentionally excluded from Git publication. The 20 ordinary entries have correct SHA-256 values. verify_all.py passes without validating manifest.json. Preserve every byte under packing/, schedule/, routing/ and quality/ and preserve all mathematical findings. Read the existing integration files. Repair manifest generation and verify_all.py so the delivered package is independently verifiable after publication, without an Atoma journal or other runtime state. Define the deliverable inventory independently of the manifest entries; check that the manifest lists every delivered regular file except itself exactly once, with its actual SHA-256. Do not assume that being present in the workspace makes a file deliverable. Atoma excludes .git, node_modules, __pycache__, .atoma, any path component starting .atoma-, .github/workflows and credential-like paths. For this existing dossier the published input set is the four stream directories plus README.md, dossier.md, verify_all.py and manifest.json; any additional maintenance files you deliver must also be inventoried. Never publish or require internal journals. Keep the original four stream checks and disposable result-corruption checks. Add persistent regression tests for the package boundary: a clean publication-shaped copy with no internal journal passes; a manifest referencing an internal journal, a missing delivered file, an altered file/hash, an omitted entry, a duplicate entry, a path escaping the package, and an unexpected ordinary file each fail for the intended manifest-integrity reason. Perform negative tests only in disposable copies; prove the original delivery files are unchanged by verification. Manifest checks must read bytes and never run commands from manifest content. Update reproduction instructions to explain when/how to regenerate the inventory and how to verify a downloaded checkout. Regenerate the manifest after all final edits. Run the existing and new checks against a copy of exactly the publishable files, not merely the live workspace. Record actual evidence; do not claim independent proof from replaying an old successful command. Do not change the platform's exclusion policy or expose .atoma-probes.json. Final answer: concise changes, checks and limitations.

Fix one remaining regression in the existing verify_all.py. The previous repair made manifest.json correct, and all original analysis files remain unchanged. However the published commit 9928280dec39e9804ed9be1b8d416cced93ade7f fails its README command in a normal Git checkout: "manifest integrity: unexpected ordinary file: .git/HEAD". A disposable package with local runtime state also fails on .atoma/cache. Make a narrow patch. Use ONE shared path-exclusion predicate for publishable_files(), the ordinary-file inventory in validate_manifest(), manifest-entry rejection, and regression fixtures. Match excluded path components case-insensitively: .git, node_modules, __pycache__, .atoma, and anything starting .atoma-. Also exclude root .github/workflows and credential-like files (.env, .env.* except *.example, .npmrc, .netrc, .pypirc, credentials, credentials.json, secrets, .secrets, id_rsa, id_ed25519, secret/secrets json/yaml/yml/toml/txt, and pem/key/p12/pfx/keystore extensions). Excluded workspace files must be ignored by the delivered inventory, but a manifest entry naming one must be rejected. Never read or execute their contents for verification. Preserve rejection of unexpected ordinary files, missing/altered/omitted/duplicate/escaping entries. Add a persistent POSITIVE regression: a disposable publication copy with harmless dummy .git/HEAD, .git/config, .atoma/cache, .atoma-probes.json, node_modules/example/index.js and quality/__pycache__/example.pyc still passes manifest validation and full verification. Its manifest must remain unchanged. Add a negative regression explicitly listing an excluded file in the manifest. Keep existing clean-copy and corruption tests. Confirm full verification does not change original files. Keep the four analysis directories and dossier.md byte-identical. Keep maintenance scope to verify_all.py, README.md only if needed, and the regenerated manifest.json. Regenerate manifest last, run the complete verifier, read back the final changed files, and report concrete evidence. No new analyses or website.