{ "schema_version": "1.0.0", "package_id": "W02", "title": "Evidence Before Expression", "language": "en", "phase": "Observe", "essential_question": "Which source can support which claim?", "research_cutoff": "2026-08-24", "status": "structurally_complete_authored_draft_human_review_and_timed_pilot_pending", "source_policy": { "principle": "Every public proposition is atomized and connected to an exact evidence item and versioned source identity; a route is not an endorsement of every source claim.", "prose_reuse": "No external prose is copied into the W02 package.", "figure_reuse": "No external figure is reproduced; all visual specifications are original instructional syntheses.", "claim_boundary": "The atomic proposition, including modality, population, system, time, and outcome, is the unit of adjudication.", "evidence_boundary": "An evidence edge resolves to a passage, table, record, data slice, or observation and states support, partial support, contradiction, qualification, non-information, or scope mismatch.", "source_boundary": "Issuer, type, version, method, artifacts, date, independence, license, and correction path constrain source use.", "closed_system_boundary": "Visible observations and documentation do not disclose hidden candidate sets, model weights, ranking logic, training data, or causal mechanisms.", "review_boundary": "Project evidence cards were used as controlled routes; independent human scientific verification and replication remain pending." }, "ceiling_legend": { "L": "Local controlled artifact; establishes frozen artifact state, not the external validity of its contents.", "N": "Normative or official authority within stated scope; does not by itself establish implementation or effect.", "A": "Primary research bounded by design, data, systems, date, and uncertainty.", "B": "Academic pedagogy or maintained research implementation; method route rather than substitute for primary empirical evidence.", "C": "First-party platform documentation for the named product, surface, locale, version, and date.", "D": "Vendor observational or quasi-experimental evidence bounded by disclosed sampling, measurement, comparator, and conflicts.", "E": "Practitioner or secondary synthesis for vocabulary, workflow examples, and hypotheses; corroboration required for scientific, legal, standards, and causal claims." }, "paper_routes": [ { "id": "P20", "citation_key": "wan2024what", "title": "What Evidence Do Language Models Find Convincing?", "url": "https://arxiv.org/abs/2402.11782", "route": "core", "role": "Bounded two-passage conflict experiment for discussing model stance adoption, passage-feature sensitivity, model-specific filtering, and why semantic relevance is not source truth or authority.", "claim_boundary": "Supports a directional discussion of feature sensitivity within the reviewed controlled setup; displayed results are conditional on generated questions, decontextualized opposing passages, dated models, and model-specific effective samples.", "evidence_boundary": "The reported outcome is model stance adoption, not factual correctness, source quality, retrieval, citation, absorption, user belief, or commercial impact.", "source_boundary": "Reviewed from arXiv:2402.11782v2 through a project evidence card; code was not executed, the repository was not commit-locked, licensing requires separate review, and independent human verification remains pending.", "review_state": "claim_level_review_human_verification_pending" }, { "id": "P40", "citation_key": "wang2026gpe", "title": "GPE: Evaluating Robust Evidence Aggregation for Fact Verification under Controllable GEO-Style Poisoning", "url": "https://arxiv.org/abs/2607.20730", "route": "core", "role": "Controlled three-evidence replacement benchmark for discussing attack-specific degradation, evidence-family risk, clean-versus-poisoned robustness, and aggregation boundaries.", "claim_boundary": "Supports a bounded claim that evaluated verification behavior differs across synthetic attack families and replacement ratios in the benchmark; it does not supply a universal robustness ranking.", "evidence_boundary": "One-run point estimates under a controlled fixed-three-object protocol do not estimate real poisoning prevalence, production source compromise, user harm, or current platform reliability.", "source_boundary": "Reviewed from the available 2026 preprint through a project evidence card whose version refresh, release artifacts, annotation details, and independent human verification remain pending; no reproduction was performed.", "review_state": "claim_level_review_version_refresh_and_human_verification_pending" } ], "notes_routes": [ { "id": "NOTES-CH02", "path": "notes-latex/chapters/02-evidence.tex", "sections": [ "Documents have identities; claims have ceilings", "An evidence ladder for use, not prestige", "The Claim--Evidence--Source graph", "Minimal evidence adequacy", "Reading an empirical GEO paper without flattening it", "Source classes in a mixed research--practice field" ], "role": "Primary conceptual spine for source identity, scoped claims, exact evidence, source classes, contradiction, and visual evidence boundaries.", "boundary": "The chapter is an evolving scholarly synthesis. Its adequacy function and evidence categories are editorial controls, not scientifically calibrated universal scores." }, { "id": "NOTES-SOURCE-LEDGER", "path": "notes-latex/SOURCE_LEDGER.md", "sections": [ "Absorption and evidence contract", "Evidence ceilings L/N/A/B/C/D/E", "Integrated, Candidate, and Context-only states" ], "role": "Canonical project definition of the course catalog ceilings and absorption states.", "boundary": "Catalog inclusion and absorption status do not validate source conclusions; source class and claim disposition remain separate." }, { "id": "NOTES-CLAIM-CONTROL", "path": "notes-latex/CLAIM_REGISTER.md", "sections": [ "C02.003 evidence ceiling", "Chapter 2 claim controls" ], "role": "Editorial gate preventing a claim-relative ceiling from becoming a universal evidence hierarchy.", "boundary": "A register entry is a control record, not independent proof of the claim." } ], "controlled_project_inputs": [ { "id": "DATA-001", "path": "sources/导出的条目_2026-08-24.csv", "sha256": "c149a69140b0dcedc1855f8cbb02fc5cb1e1346470306c76cd68fbafc2e50468", "role": "Canonical 42-record paper-corpus membership and metadata freeze used to retain stable P20 and P40 identities.", "boundary": "A corpus export proves the local record state, not the validity, reproducibility, or present status of paper conclusions." }, { "id": "COURSE-RECORD-W02", "path": "course-site/app/lib/course-data.ts#W02", "role": "Authoritative local W02 outcomes, essential question, before/during/after sequence, visual, interaction, paper route, and low-compute boundary.", "boundary": "The course record is a curriculum specification, not evidence that instruction has been taught, timed, or shown effective." }, { "id": "P20-EVIDENCE-CARD", "path": "notes-latex/evidence-cards/P20-wan2024what.json", "role": "Claim-level method, limitation, external-validity, artifact, license, and forbidden-generalization control for P20.", "boundary": "Automated full-text evidence pass with human verification pending; no independent reproduction." }, { "id": "P40-EVIDENCE-CARD", "path": "notes-latex/evidence-cards/P40-wang2026gpe.json", "role": "Claim-level method, limitation, external-validity, artifact, license, and forbidden-generalization control for P40.", "boundary": "Automated full-text evidence pass with version refresh and human verification pending; no independent reproduction." } ], "synthetic_assets": [ { "id": "NORTHSTAR-ATLAS-W02", "path": "WORKED_CASE.md", "role": "Offline fixture with nine source cards, five source families, six atomic claims, calculations, dispositions, release decisions, and next evidence actions.", "claim_boundary": "Teaches deployment, descriptive change, causal inference, cross-platform transport, assisted-transaction measurement, and commercial-effect distinctions.", "evidence_boundary": "All excerpts, counts, systems, dates, and identities are authored teaching data and support no real-world estimate.", "source_boundary": "No real external source, company, platform, customer, or person is represented." } ], "time_equivalence": { "recorded_media_exists": false, "transcript_planning_rate_wpm": 125, "transcript_target_minutes": { "minimum": 35, "maximum": 45 }, "synchronous_seminar_planned_minutes": 90, "studio_planned_minutes": 75, "timed_rehearsal_completed": false, "boundary": "All durations are authoring estimates until measured from a real class, rehearsal, or finished media file." }, "artifact_contract": [ { "file": "README.md", "role": "Delivery contract, learning outcomes, synchronous and asynchronous routes, evidence legend, assessment, and safety boundary." }, { "file": "LECTURE_NOTES.md", "role": "Stand-alone 3,000–4,500-word conceptual lesson." }, { "file": "SLIDE_SCRIPT.md", "role": "Twenty-four-slide production script with original visuals and text alternatives." }, { "file": "WORKED_CASE.md", "role": "Complete synthetic claim-atomization and evidence-ceiling case." }, { "file": "TRANSCRIPT.md", "role": "No-video readable/recordable equivalent with a planned 35–45-minute delivery mapping." }, { "file": "manifest.json", "role": "Machine-readable provenance, scope, route, and time-equivalence controls." }, { "file": "validate_package.py", "role": "Standard-library structural validator and SHA-256 reporter." } ] }