manual_slop/conductor/tracks/video_analysis_synthesis_20260621/spec.md

Track: Video Analysis Campaign — Synthesis (2026-06-21)

Status: Not started (umbrella published 2026-06-21) Type: Research-only synthesis track (Pass 1 of 3) Owner: Tier 1 Orchestrator (synthesis spec + report); Tier 2 Tech Lead (execution) Priority: A (delivers the user's requested "summary of each video" + "summary report of key takeaways") Domain: Meta-tooling (cross-cutting research synthesis; no src/ changes)

Parent: This synthesis track is blocked_by all 12 child tracks of the video_analysis_campaign_20260621 umbrella. See umbrella spec.md for the full campaign design and multi-pass context.

Multi-pass note: The synthesis report.md is intermediate input to Pass 2 (de-obfuscation). Per the campaign's lossless-preservation directive (umbrella spec §0), the synthesis must preserve detail — Pass 2 will compress, not this pass.

---

1. Inputs

This track consumes the outputs of all 12 child tracks:

#	Slug	Cluster	Source
1	cs229_building_llms	E	video_analysis_cs229_building_llms_20260621/
2	probability_logic	A	video_analysis_probability_logic_20260621/
3	entropy_epiplexity	A	video_analysis_entropy_epiplexity_20260621/
4	score_dynamics_giorgini	A	video_analysis_score_dynamics_giorgini_20260621/
5	platonic_intelligence_kumar	B	video_analysis_platonic_intelligence_kumar_20260621/
6	free_lunches_levin	B	video_analysis_free_lunches_levin_20260621/
7	generic_systems_fields	C	video_analysis_generic_systems_fields_20260621/
8	brain_counterintuitive	C	video_analysis_brain_counterintuitive_20260621/
9	neural_dynamics_miller	C	video_analysis_neural_dynamics_miller_20260621/
10	multiscale_hoffman	C	video_analysis_multiscale_hoffman_20260621/
11	cs336_architectures	E	video_analysis_cs336_architectures_20260621/
12	creikey_dl_cv	D	video_analysis_creikey_dl_cv_20260621/

Per-child inputs consumed:

• report.md (the 1000-10000 LOC deep-dive)
• summary.md (the 200-400 word quick summary)

The per-child transcript.json, frames/, ocr.md, and extraction_meta.json are NOT consumed here — they feed Pass 2 directly.

---

2. Deliverables

Artifact	Path	Description
Per-video roll-up	per_video_summary.md	One paragraph (150-250 words) per video — the "summary of each video" the user requested. Ordered by execution order (matches umbrella §6).
Synthesis report	report.md	The "summary report of key takeaways" — 6 sections per umbrella §FR7.

---

3. Synthesis report structure (6 sections)

Per umbrella spec §FR7.

# Video Analysis Campaign — Synthesis

## 1. Theme Matrix (across clusters A/B/C/D/E)
## 2. Cross-Video Concept Map
## 3. 5-10 High-Level Takeaways
## 4. Mathematical Prerequisite Graph
## 5. Open Research Questions
## 6. Recommended Next-Watch List

Section detail:

§1 Theme Matrix — a 2D table with rows = clusters (A/B/C/D/E) and columns = themes (e.g., foundations, representations, training, applications, biological inspiration, ethics). Each cell: which videos address this theme.

§2 Cross-Video Concept Map — for each major concept that appeared in 2+ videos, list: (a) which videos introduced it, (b) which videos built on it, (c) which videos referenced it. Format: per-concept subsection with a list of video slugs + brief role description.

§3 5-10 High-Level Takeaways — bullet list of the most important cross-cutting insights the user should walk away with. Each takeaway: 2-5 sentences with references to the videos that support it.

§4 Mathematical Prerequisite Graph — a directed graph showing which mathematical concepts are needed to understand which. E.g., "to understand CS336 Lecture 3, you need: linear algebra (CS229), probability (Probability = Extension of Logic), score-based dynamics (Giorgini)." Format: text-based DAG or ASCII graph.

§5 Open Research Questions — questions raised by the videos that the field doesn't have consensus answers to. The user mentioned this is "largely impenetrable" to them and associates — these questions are the campaign's open frontier.

§6 Recommended Next-Watch List — based on what the user liked in this batch, suggest related videos/authors/topics to investigate next. Source from cross-references in the per-video reports + the user's stated stylistic preferences.

---

4. Pipeline (per umbrella spec §FR7)

• Phase 1: Ingest. Read all 12 child report.md files + summary.md files. Build an in-memory index.
• Phase 2: Per-video roll-up. Generate per_video_summary.md by either lifting each child's summary.md (preferred) or writing a 150-250 word summary if the child's is too short.
• Phase 3: Synthesis report. Generate report.md per the 6-section structure above. Pass 1 of 3 = detailed; Pass 2 will compress.
• Phase 4: Verification. Cross-check that every video in the campaign has a roll-up entry. Cross-check that every theme in §1 is sourced from at least one video. Cross-check that every takeaway in §3 has at least one supporting video reference.

---

5. Lossless preservation directive

Per umbrella spec §0: this synthesis is intermediate input to Pass 2 (de-obfuscation). DO NOT over-summarize. The §3 takeaways should be 2-5 sentences each (not 1 sentence). The §4 math prerequisite graph should reference specific videos, not just "foundational math." The §5 open research questions should include the user's own context (what's impenetrable to them) and not just generic AI debates.

If the synthesis report is less than 1000 LOC, it is too short. Target: 1000-5000 LOC for the synthesis report (less than per-video because the heavy lifting is in the per-video reports).

---

6. Verification (per umbrella spec §12)

• All 12 children shipped (read all their report.md + summary.md)
• per_video_summary.md has 12 entries (one per video), each 150-250 words
• report.md has all 6 sections populated
• report.md is 1000-5000 LOC (detailed enough for Pass 2 to compress)
• Every §3 takeaway references at least one video
• Every §1 theme cell references at least one video
• §6 next-watch list references at least 3 sources

---

7. Out of scope (per umbrella spec §10)

• De-obfuscation (Pass 2 — future track, user must first rediscover encoding notation)
• Projection to applied domain (Pass 3 — future track, user must first articulate "own caveats")
• Modifying any src/*.py files in manual_slop
• Building a video knowledge base (separate dimension per conductor/code_styleguides/knowledge_artifacts.md)

---

8. See also

• Umbrella spec.md — full campaign design + multi-pass context
• Umbrella plan.md — campaign-level plan
• Umbrella README.md — child index
• All 12 child spec.md files (linked in §1 above)