{ "schema": "project-telos.learning-forge/youtube-research-seed/v1", "generated_at": "2026-06-30T00:00:00.000Z", "tool": "telos.learning.forge", "purpose": "Turn operator-provided AI research videos, channels, papers, and official pages into verified learning objects: source leads, concept cards, toy labs, failure cases, measurements, and Crucible verdicts.", "validation": { "verdict": "MATCH", "validated_by": [ "gather.status", "index.map", "forum.doctor", "crucible.status", "telos.catalog" ], "notes": [ "YouTube links are stored as source leads until Gather captures metadata, captions, transcript hashes, channel identity, upload date, and digest seals.", "Current source receipts are lawful public source references for the research frame, not full-paper or raw-transcript imports.", "Learning objects may promote a claim only after a source receipt and a Crucible gate both join to the same claim card." ] }, "source_boundary": { "raw_transcripts_stored": false, "raw_video_stored": false, "youtube_claims_promoted": false, "shadow_full_text_allowed_as_provenance": false, "operator_notes_are_source_leads": true, "fresh_resolution_required": true, "claim_promotion_rule": "A YouTube segment starts UNVERIFIABLE. It can become MATCH only after Gather records metadata/transcript receipts and Crucible joins the segment to lawful current sources or runnable measurements." }, "youtube_seed_corpus": { "receipt_state": "UNVERIFIABLE_UNTIL_GATHER_TRANSCRIPT", "curation_rule": "Treat the supplied videos and channels as leads for AI education, reasoning, agents, reproducibility, interpretability, coding agents, AI systems, alternative compute, and AI-for-science. Do not invent titles, transcripts, upload dates, or claims.", "inputs": [ "https://www.youtube.com/watch?v=5pieVHmlbyk&t=668s", "https://www.youtube.com/watch?v=1WH9_edB538", "https://www.youtube.com/watch?v=3AhJNyj_X98&t=238s", "https://www.youtube.com/watch?v=0gvHqRv8gTg", "https://www.youtube.com/watch?v=Ihv542rCUns", "https://www.youtube.com/watch?v=w-GugZv044I", "https://www.youtube.com/watch?v=Mw60FH5iflI&t=107s", "https://www.youtube.com/watch?v=VlP-12yc2f8&t=668s", "https://www.youtube.com/@lobais/videos", "https://www.youtube.com/@MachineLearningStreetTalk" ], "required_gather_fields": [ "source_type", "provided_url", "video_id_or_channel_handle", "timestamp_hint", "channel", "upload_date", "metadata_sha256", "transcript_sha256", "description_sha256", "digest_seal", "availability", "claim_state" ], "gather_receipt_ledger": { "ref": "demo/research/youtube-learning-forge-receipts.json", "schema": "project-telos.learning-forge/youtube-research-seed/v1", "captured_at": "2026-06-30T07:27:57.550385+00:00", "video_count": 8, "channel_count": 2, "corpus_digest_seal": "acfdc2e9ef532620579f221fccd59f8ba3e3505c1d28084aa7e893280bef88d5", "receipt_state": "GATHER_RECEIPTS_CAPTURED", "research_claim_state": "UNVERIFIABLE_UNTIL_CROSS_CHECK", "privacy_boundary": "Receipt ledger stores metadata hashes, transcript hashes, channel-list hashes, titles, refs, and counts only; raw transcript bodies stay in the ignored local corpus." }, "segment_model": [ "intro_claim_setup", "core_concept", "technical_mechanism", "failure_modes", "research_directions" ] }, "current_source_receipts": [ { "id": "source-neurips-2026-education-track", "title": "NeurIPS 2026 Call for Educational Resources", "url": "https://neurips.cc/Conferences/2026/CallforEducationalResources", "source_kind": "official-conference-call", "resolved_at": "2026-06-30T00:00:00Z", "provenance_class": "lawful_source", "receipt_hash": "sha256:40b3deff2a7f483b4a9387e527ab85922341b3542755eca0131d6e957495398f" }, { "id": "source-neurips-2026-reproducibility-track", "title": "NeurIPS 2026 Call for Reproducibility Papers", "url": "https://neurips.cc/Conferences/2026/CallForReproducibility", "source_kind": "official-conference-call", "resolved_at": "2026-06-30T00:00:00Z", "provenance_class": "lawful_source", "receipt_hash": "sha256:9e567b2826828f4eb69e14fe09530ef32e12ed31d33b0394f1c7d56ca643c743" }, { "id": "source-neurips-2026-evaluations-datasets-track", "title": "NeurIPS 2026 Evaluations & Datasets Track Call for Papers", "url": "https://neurips.cc/Conferences/2026/CallForEvaluationsDatasets", "source_kind": "official-conference-call", "resolved_at": "2026-06-30T00:00:00Z", "provenance_class": "lawful_source", "receipt_hash": "sha256:177ed968844e58602b1b426c51ed3eadb5efca2dd513cc529ca45659e7811ee3" }, { "id": "source-ibm-planning-era-language-models", "title": "Planning in the Era of Language Models for NeurIPS 2025", "url": "https://research.ibm.com/publications/planning-in-the-era-of-language-models", "source_kind": "official-research-publication-page", "resolved_at": "2026-06-30T00:00:00Z", "provenance_class": "lawful_source", "receipt_hash": "sha256:a7031628995dc702d3bdc4f9575d6f62984963fd54469fc11e4758951ad2e77c" }, { "id": "source-deepseek-r1-arxiv", "title": "DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning", "url": "https://arxiv.org/abs/2501.12948", "source_kind": "open-preprint", "resolved_at": "2026-06-30T00:00:00Z", "provenance_class": "lawful_source", "receipt_hash": "sha256:2ea080196440647dc94320dc3b68eca373e4d1e07a25260890b756b71d9f3765" }, { "id": "source-swe-skills-bench-arxiv", "title": "SWE-Skills-Bench: Do Agent Skills Actually Help in Real-World Software Engineering?", "url": "https://arxiv.org/abs/2603.15401", "source_kind": "open-preprint", "resolved_at": "2026-06-30T00:00:00Z", "provenance_class": "lawful_source", "receipt_hash": "sha256:0c9cacc24c93cf0cf648f9e7067aa74f6f4266263e399ddffeea964e388ac801" }, { "id": "source-swe-bench-pro-arxiv", "title": "SWE-Bench Pro: Can AI Agents Solve Long-Horizon Software Engineering Tasks?", "url": "https://arxiv.org/abs/2509.16941", "source_kind": "open-preprint", "resolved_at": "2026-06-30T00:00:00Z", "provenance_class": "lawful_source", "receipt_hash": "sha256:7c2774e96d6f3df9534a0727a20f4f86e4b887ae2e9d86d5fc6d97ea13c25011" }, { "id": "source-mcp-2025-06-18-spec", "title": "Model Context Protocol specification 2025-06-18", "url": "https://modelcontextprotocol.io/specification/2025-06-18", "source_kind": "official-protocol-specification", "resolved_at": "2026-06-30T00:00:00Z", "provenance_class": "lawful_source", "receipt_hash": "sha256:7844c9ee267b08c069d5f11734277cede29e283314f5d9b5eec76fcf43180bc6" }, { "id": "source-opentelemetry-genai-semconv", "title": "OpenTelemetry Generative AI semantic conventions", "url": "https://opentelemetry.io/docs/specs/semconv/gen-ai/gen-ai-agent-spans/", "source_kind": "official-observability-specification", "resolved_at": "2026-06-30T00:00:00Z", "provenance_class": "lawful_source", "receipt_hash": "sha256:2686765617fd3f26be78a793e8e2923a347bdf325178b7fc672993249ae5c928" } ], "learning_modules": [ { "id": "programming-as-evidence", "level": "intro", "source_receipt_ids": [ "source-neurips-2026-education-track", "source-neurips-2026-reproducibility-track" ], "teaching_object": "A small file-processing tool that extracts claim cards, runs tests, and emits receipts.", "failure_case": "A polished explanation without a source hash or test output remains UNVERIFIABLE.", "crucible_gate": "Assert every lesson output includes source_ref, artifact_hash, failure_case_ref, and verdict." }, { "id": "reasoning-test-time-compute", "level": "builder", "source_receipt_ids": [ "source-deepseek-r1-arxiv" ], "teaching_object": "Compare greedy answer, multiple samples, verifier selection, token cost, and replayed scoring.", "failure_case": "More samples raise cost without improving a hard held-out task.", "crucible_gate": "Measure accuracy, token count, latency, verifier agreement, and failed attempts per task." }, { "id": "agents-tools-mcp-receipts", "level": "builder", "source_receipt_ids": [ "source-ibm-planning-era-language-models", "source-mcp-2025-06-18-spec", "source-opentelemetry-genai-semconv" ], "teaching_object": "A tool-using agent that separates plan state, action admission, execution, trace span, and receipt.", "failure_case": "A completed tool call cannot be joined to the admitted action intent.", "crucible_gate": "Negative fixtures for changed args_hash, missing action_intent_id, stale criterion_ref, and missing verifier result." }, { "id": "coding-agents-repo-work", "level": "research", "source_receipt_ids": [ "source-swe-skills-bench-arxiv", "source-swe-bench-pro-arxiv" ], "teaching_object": "A repo patch lab with issue evidence, Index map, focused tests, code readability check, and review receipt.", "failure_case": "The agent edits boilerplate but misses the hidden invariant or overfits a public test.", "crucible_gate": "Require diff hash, changed-module map, targeted tests, failure typing, and reviewer-readable code evidence." }, { "id": "evaluation-reproducibility", "level": "builder", "source_receipt_ids": [ "source-neurips-2026-reproducibility-track", "source-neurips-2026-evaluations-datasets-track" ], "teaching_object": "A replication sprint that records benchmark card, data split, contamination checks, and negative results.", "failure_case": "A headline benchmark result cannot be reproduced from the provided setup.", "crucible_gate": "Promote MATCH only when artifact, command, environment, source refs, and result digest re-run cleanly." }, { "id": "interpretability-faithfulness", "level": "research", "source_receipt_ids": [ "source-neurips-2026-evaluations-datasets-track" ], "teaching_object": "A small model explanation audit comparing perturbation, attribution, and counterexample behavior.", "failure_case": "The explanation looks plausible but fails under a controlled perturbation.", "crucible_gate": "Record input hash, explanation method, perturbation, observed output shift, and remaining unverifiable claims." }, { "id": "efficient-alternative-compute", "level": "research", "source_receipt_ids": [ "source-deepseek-r1-arxiv" ], "teaching_object": "An accuracy-per-token and latency-per-verdict lab for reasoning strategies and stochastic search loops.", "failure_case": "A method improves raw accuracy but violates cost, latency, or energy budgets.", "crucible_gate": "Attach compute budget, seed, retry count, verifier cost, and promotion threshold to each result." }, { "id": "ai-for-science-measurement", "level": "research", "source_receipt_ids": [ "source-neurips-2026-evaluations-datasets-track", "source-neurips-2026-reproducibility-track" ], "teaching_object": "A scientific claim card that starts from measurement, uncertainty, and source provenance before model output.", "failure_case": "A model analogizes from biology or physics without measured variables, units, or independent checks.", "crucible_gate": "Require units, source ids, measurement method, uncertainty statement, and claim boundary before synthesis." } ], "claim_cards": [ { "id": "claim-youtube-leads-captured", "claim": "The operator-provided YouTube videos and channels are captured as source leads.", "verdict": "MATCH", "evidence_refs": [ "youtube_seed_corpus.inputs" ], "limits": [ "No title, transcript, upload date, or video-specific claim is promoted in this packet." ] }, { "id": "claim-current-learning-forge-sources", "claim": "The Learning Forge frame is backed by current lawful source receipts for education, reproducibility, evaluation, planning, reasoning, coding-agent benchmarks, MCP, and observability.", "verdict": "MATCH", "evidence_refs": [ "current_source_receipts" ], "limits": [ "Source receipts identify current public references; they do not import full text or prove every technical claim inside each source." ] }, { "id": "claim-video-specific-technical-content", "claim": "The supplied videos contain the exact topics inferred in the operator packet.", "verdict": "UNVERIFIABLE", "evidence_refs": [ "youtube_seed_corpus.receipt_state" ], "limits": [ "Requires Gather metadata and transcript receipts before promotion." ] } ], "flagship_roles": { "gather": [ "Capture video metadata, captions, transcript hashes, descriptions, official pages, arXiv metadata, and digest seals.", "Keep unavailable captions, failed metadata, and blocked sources as explicit receipts instead of silent gaps." ], "index": [ "Map concepts into a context graph spanning programming, models, reasoning, agents, benchmarks, interpretability, systems, and science.", "Produce budgeted context envelopes for each lab without raw transcript inflation." ], "forum": [ "Route each module to education, research, systems, agent, science, or verification lanes.", "Humanize lab prose without adding facts or promoting unsupported video claims." ], "crucible": [ "Split concept summaries from falsifiable claims and assign MATCH, DRIFT, or UNVERIFIABLE.", "Build negative corpora for benchmark overfit, missing source hash, stale evidence, unjoinable action, and unverifiable transcript cases." ], "telos": [ "Expose the learning forge through CLI and MCP as a reusable host-neutral packet.", "Join later labs, receipts, context packs, and loop-ledger entries into one operator-visible research state." ] }, "next_actions": [ { "owner": "gather", "action": "capture_youtube_metadata", "input": ".telos/gather/youtube-research-seed/videos.jsonl", "expected_receipt": "metadata MATCH or explicit fetch_failed" }, { "owner": "index", "action": "build_concept_graph", "input": "learning_modules", "expected_receipt": "context graph with retained source refs and topic edges" }, { "owner": "forum", "action": "route_first_five_labs", "input": "learning_modules[0:5]", "expected_receipt": "one lane and escalation state per lab" }, { "owner": "crucible", "action": "build_negative_corpus", "input": "claim_cards", "expected_receipt": "negative fixtures for source, transcript, benchmark, and action-join failures" }, { "owner": "telos", "action": "publish_learning_forge_v0", "input": "this packet plus first runnable toy lab", "expected_receipt": "host-neutral CLI/MCP packet and README operator path" } ] }