The Applied Invention Lab

Create an assumption ledger for the fictional library-kit packing problem. Separate scalar capacity arithmetic from actual geometry, handling, user behavior, and commercial usefulness.

Inputs: starter/docs/problem-brief.md, starter/docs/data-dictionary.md, starter/docs/operator-worksheet.csv, starter/data/capacity.json

Outputs: deliverables/assumption-ledger.csv

Depends on: No prerequisites

• Every capacity, item characteristic, and practical outcome is labeled synthetic or not observed.
• The ledger explicitly excludes physical fit and lifting-safety certification.
• At least one falsifying observation is specified for geometry and one for operator usefulness.

Reconcile row counts, expanded item IDs, units, per-scenario totals, and capacity validity. Preserve source order because it is a deliberate baseline variable, not random observed history.

Inputs: deliverables/assumption-ledger.csv, starter/docs/data-dictionary.md, starter/data/items.csv, starter/data/capacity.json, starter/reference/tote_pack.py

Outputs: deliverables/dataset-audit.json

Depends on: research-assumptions

• Eleven item-type rows expand to 29 instances in four scenarios.
• All expanded IDs are unique within a scenario and every item individually fits the scalar capacities.
• The order-trap, weight-trap, mixed-classroom, and bulky-gaps purposes are identified.

Run the authored tests and baseline CLI with optional small-model proof enabled. Record assignments, counts, lower bounds, and proof method for every scenario; do not omit unfavorable cases.

Inputs: deliverables/dataset-audit.json, starter/reference/tote_pack.py, starter/reference/test_tote_pack.py, starter/reference/expected-baseline.json, starter/data/items.csv, starter/data/capacity.json

Outputs: deliverables/baseline-results.json

Depends on: dataset-audit

• All expected baseline counts match and all assignment invariants pass.
• mixed-classroom reports dominant 7 versus input-order 6 and volume-first 8 bins.
• Proof labels apply only to the abstract model and no physical experiment is claimed.

Create a printable packing-instruction prototype from one baseline scenario. Show tote labels, exact item IDs, both scalar loads, remaining capacity, and a prominent model-only warning.

Inputs: deliverables/baseline-results.json, starter/docs/prototype-brief.md, starter/docs/operator-worksheet.csv, starter/data/items.csv

Outputs: deliverables/packing-instructions.html

Depends on: baseline-computation

• Every chosen scenario item appears once in the printable artifact.
• Both load dimensions and the model-only warning are visible without color dependence.
• No real tote was packed and no physical dimensions were verified merely by rendering the instructions.

Execute the original experiment runner using the supplied seed, 20 permutations per scenario, and all three declared algorithms. Retain every trial and the dataset/configuration/plan hashes.

Inputs: deliverables/baseline-results.json, starter/docs/experiment-design.md, starter/data/experiment-plan.json, starter/data/items.csv, starter/data/capacity.json, starter/reference/run_experiment.py, starter/reference/tote_pack.py

Outputs: deliverables/variant-comparison.json

Depends on: baseline-computation

• The result contains 4 scenarios × 3 algorithms × 20 trial counts.
• The same permutation is used for each algorithm within a trial.
• No measured sample is discarded to claim a universally superior heuristic.

Compare baseline, all trial results, and the operator prototype. Explain why one-dimensional intuition can fail under two constraints and distinguish computational evidence from physical or market evidence.

Inputs: deliverables/variant-comparison.json, deliverables/baseline-results.json, deliverables/packing-instructions.html, deliverables/assumption-ledger.csv, starter/docs/experiment-design.md

Outputs: deliverables/research-interpretation.md

Depends on: variant-experiment, packing-prototype

• The mixed-classroom regression and the bulky-gaps lower-bound gap are discussed.
• The conclusion uses the complete trial set and does not generalize four synthetic scenarios to real libraries.
• At least two decisions remain conditional on later physical or operator observations.

Follow the replication protocol in a fresh interpreter using the same explicit source files. Recompute hashes, baseline assignments, and seeded trials; report exact agreement or discrepancies, including runtime details.

Inputs: deliverables/variant-comparison.json, deliverables/baseline-results.json, starter/docs/replication-protocol.md, starter/reference/tote_pack.py, starter/reference/run_experiment.py, starter/reference/test_tote_pack.py, starter/reference/expected-baseline.json, starter/data/items.csv, starter/data/capacity.json, starter/data/experiment-plan.json

Outputs: deliverables/replication-report.json

Depends on: variant-experiment

• Input hashes and every per-trial count agree or a discrepancy is explicitly reported.
• All assignment constraints are checked again rather than trusting the first report.
• Not-observed physical outcomes remain outside the replication claim.

Design a later owner-approved study using empty or lightweight mock materials, not real transport or lifting limits. Specify geometry, item accessibility, labeling, and operator-comprehension observations without recording invented results.

Inputs: deliverables/research-interpretation.md, deliverables/replication-report.json, deliverables/packing-instructions.html, starter/docs/prototype-brief.md, starter/docs/operator-worksheet.csv

Outputs: deliverables/physical-fit-protocol.csv

Depends on: research-interpretation, replication-run

• The protocol distinguishes planned observation from an actual measured value.
• Mock geometry and readability checks have explicit pass/fail criteria.
• No lifting-safety, structural-load, shipping, or real-world fit guarantee is proposed from scalar units.

Prepare an opportunity brief based on the observed computation and its limitations. Compare an explainable planning worksheet with an integration-heavy service; keep interviews, pricing, procurement, and intellectual-property review as separate future approvals.

Inputs: deliverables/research-interpretation.md, deliverables/assumption-ledger.csv, starter/docs/commercialization-boundary.md, starter/docs/problem-brief.md

Outputs: deliverables/application-hypothesis.md

Depends on: research-interpretation

• A potential user job and disconfirming signal are described without fabricated demand.
• No patentability, novel algorithm, customer, savings, or willingness-to-pay claim is made.
• The brief names the physical and operator evidence needed before external positioning.

Combine replication, interpretation, mock-fit protocol, and application hypothesis into a reviewable stop/continue recommendation. Distinguish a working computation from a proven invention and keep external action owner-gated.

Inputs: deliverables/replication-report.json, deliverables/research-interpretation.md, deliverables/physical-fit-protocol.csv, deliverables/application-hypothesis.md, starter/docs/commercialization-boundary.md

Outputs: deliverables/lab-decision.json

Depends on: physical-fit-protocol, application-hypothesis

• The recommendation states what the code established and what it could not establish.
• All four scenarios and unfavorable evidence inform the recommendation.
• No real study, commercialization, patent filing, purchase, or publication is represented as approved or completed.