ONE TASK BETTER — FICTIONAL WEEKLY-PLAN PRACTICE

Prepared from next-episode-editorial-briefs.txt, V05. No video, publication, live AI generation, actual task completion or measured savings is claimed. Roles are teaching inputs, not identified people.

Run offline with Python 3 and the standard library:
python3 build_demo.py --input source-fixture.json --out regenerated --verify-fixture

The base fixture requests 1,080 minutes (18 hours), supplies 600 minutes role capacity (10 hours), reserves 120 minutes (2 hours), and selects 480 minutes (8 hours). Intake supplies 240 minutes minus 60 reserve and selects 180. Operations supplies 180 minus 60 reserve and selects 120. Owner supplies 180 and selects 180. The only explicit customer-work reserves are Intake and Operations; Owner reserve 0 follows that closed constructed-input rule, not a recommendation to assume missing reserves are zero in a real business.

The first four tasks are the supplied selection, not an optimized order or completed work. Handoff approval is planned after file collection and execution_ready=false. All completion flags remain false. FAQ owner/date remain blank; its original soon stays unresolved. Archive 360 minutes remains visible outside selection with owner/date blank. No task deadline is supplied; this weekly bucket is not a calendar booking or guessed deadline.

missing-files-plan.json changes an explicit constructed availability input: approval moves to Waiting, while the two-hour collection task remains planned effort to obtain/check files. Revised selected effort is 300 minutes (5 hours). This does not assert that missing files can be obtained or that a handoff is ready. If collection itself becomes blocked, unselect it or provide an explicit blocking dependency; its successor then stays Waiting.

SourceEvents contain exact editorial excerpts plus a separately labeled constructed scenario condition. Source-linked audit preserves citations and explains every decision. Source fixtures and JSON retain exact original text. Owner assignments additionally require the role and task term together in a supplied comma-delimited role-to-task clause; finding both words somewhere in a full source quote is insufficient. This is a bounded check for this prepared syntax, not complete language interpretation. New proposed facts, assignment scope, dependencies and capacities need source review by the business owner.

CSV import: use Data > From Text/CSV or equivalent. Set task_id and source_id to Text to preserve leading zeros. Set human text and dependency/source ID lists to Text. CSV quotation alone does not establish types. Formula-like human text beginning with equals, plus, minus or at-sign (after common whitespace) gets a display apostrophe; escape-manifest.json preserves exact originals. Unknown fields are empty, distinct from confirmed numeric 0. Minute fields are nonnegative integers when supplied. Native Excel import/calculation has not been tested.

Verification checks integer-minute reconciliation, each role limit, blank owners/dates, source quotations, guessed assignments/deadlines, null versus zero, negative/fractional/boolean inputs, reserve bounds, unknown capacity, dependency cycles/unknown IDs, blocked successors, missing files and CSV preservation. These are deterministic prepared-data checks, not a claim of complete natural-language understanding.

Manifest lists SHA-256 and byte size of payload files; it excludes itself and the ZIP. The ZIP includes the manifest and uses fixed entry timestamps for reproducibility. Keep every unselected or unresolved task visible; record real full-task measurements before claiming savings.
