ed/manual_slop
Private
Public Access
0
0
Fork 0
Code Activity

conductor(deob_phases): Initialize 3 phase child spec scaffolds

Browse Source 
Each child spec is lightweight (~120 lines): references the umbrella, gives the deliverable structure, specifies the inputs/outputs, and the 5-phase pipeline.

Phase 1 (lexicon): refines the warmup's draft into a codified operational spec (lexicon.md + terms_catalog.md + dedup_map.md)
Phase 2 (pilot): applies the lexicon to 2 Pass 1 reports (cs229_building_llms + entropy_epiplexity), captures refinements in pilot_report.md
Phase 3 (apply): applies the refined lexicon to 10 remaining Pass 1 reports + 1 cross-cutting synthesis, final apply_report.md

3-layer deliverable per video: translation (side-by-side) + replacement (re-encoded) + decoder (per-term etymology + form anchor + definition history)
4 verification criteria: lossless, bounded, constructively typed, etymology-cited
This commit is contained in:
ed
2026-06-23 00:08:23 -04:00
parent f830798822
commit 256af96bf3
3 changed files with 308 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
conductor/tracks/video_analysis_deob_pilot_20260621/spec.md
+104
View File
@@ -0,0 +1,104 @@
# Track: Video Analysis De-obfuscation — Pilot on 2 Videos (2026-06-21)
**Status:** Not started (umbrella + warmup published 2026-06-21)
**Type:** Research-only child track (Pass 2 of 3, Phase 2 of 3 within Pass 2)
**Owner:** Tier 2 Tech Lead (execution)
**Priority:** A (user-blocking)
> **Parent:** This is child #2 of the [video_analysis_deob_20260621](../../video_analysis_deob_20260621/) umbrella. See [umbrella spec.md](../../video_analysis_deob_20260621/spec.md) for the full design.
> **Upstream consumer of:** [video_analysis_deob_lexicon_20260621](../../video_analysis_deob_lexicon_20260621/) (the refined `lexicon.md` + `terms_catalog.md` + `dedup_map.md`).
> **Downstream consumer:** [video_analysis_deob_apply_20260621](../../video_analysis_deob_apply_20260621/) (consumes the refined lexicon from the pilot's `pilot_report.md`).
---
## 1. Inputs
| Input | Source |
|---|---|
| Refined `lexicon.md` | [video_analysis_deob_lexicon_20260621/lexicon.md](../../video_analysis_deob_lexicon_20260621/lexicon.md) |
| Refined `terms_catalog.md` | [video_analysis_deob_lexicon_20260621/terms_catalog.md](../../video_analysis_deob_lexicon_20260621/terms_catalog.md) |
| Refined `dedup_map.md` | [video_analysis_deob_lexicon_20260621/dedup_map.md](../../video_analysis_deob_lexicon_20260621/dedup_map.md) |
| Warmup `prompt_template.md` | [video_analysis_deob_warmup_20260621/prompt_template.md](../../video_analysis_deob_warmup_20260621/prompt_template.md) |
| Pass 1 report #1 | [video_analysis_cs229_building_llms_20260621/report.md](../../video_analysis_cs229_building_llms_20260621/report.md) |
| Pass 1 report #2 | [video_analysis_entropy_epiplexity_20260621/report.md](../../video_analysis_entropy_epiplexity_20260621/report.md) |
## 2. Pilot videos (why these 2)
1. **`cs229_building_llms`** — Foundational ML/LLM coverage. Wide scope, tests "form projection" across many concepts (sets, functions, optimization, learning). Good test for the lexicon's general applicability.
2. **`entropy_epiplexity`** — Math-heavy, focused on information-theoretic concepts (entropy, mutual information, epiplexity). Good test for "boundedness" + type-theoretic encoding of measure theory (a common bottleneck in de-obfuscation efforts).
These 2 videos test the lexicon on different shapes: one broad-and-shallow, one narrow-and-deep. The pilot's refinements will inform the apply phase.
## 3. Deliverables
For each of the 2 pilot videos, 3 files in `artifacts/<slug>/`:
| Artifact | Description |
|---|---|
| `<slug>_translation.md` | Side-by-side table: original expression ↔ re-encoded form, with form anchor per row |
| `<slug>_deobfuscated.md` | The re-encoded report (replacement; same 8-section structure as Pass 1, but every standard-math expression is replaced with the constructive type-theoretic form) |
| `<slug>_decoder.md` | Per-term decoder: form anchor, etymology, definition history, link to original section |
Plus:
| Artifact | Description |
|---|---|
| `pilot_report.md` | Lexicon refinements discovered, concepts that didn't fit (gaps), process improvements for the apply phase |
**Optional (added per child track execution convention):** `plan.md`, `metadata.json`, `state.toml`.
## 4. Pipeline (5 phases)
Per the umbrella spec §5 (Phase 3 of the umbrella). Each phase commits atomically.
- [ ] **Phase 1: Init.** Initialize the child track.
- [ ] **Phase 2: Apply to `cs229_building_llms`.** Tier 3 worker (LLM invoked with the prompt template + the report). Produces 3-layer deliverable in `artifacts/cs229_building_llms/`.
- [ ] **Phase 3: Apply to `entropy_epiplexity`.** Same as above. Produces 3-layer deliverable in `artifacts/entropy_epiplexity/`.
- [ ] **Phase 4: Refine.** Write `pilot_report.md` capturing: lexicon refinements, concepts that didn't fit (gaps), process improvements.
- [ ] **Phase 5: User review + verification.** User reviews the 2 deliverables. Approves before the apply child can start. Idempotency + audit + end-of-track report.
## 5. Report structure (per pilot video)
For `<slug>_deobfuscated.md`, the structure is the same 8-section structure as Pass 1's report:
1. TL;DR
2. Key Concepts
3. Frame Analysis
4. Transcript Highlights
5. Mathematical / Theoretical Content (re-encoded)
6. Connections to Other Videos in Campaign
7. Open Questions / Follow-up
8. References
But every standard-math expression in §5 (and anywhere else) is replaced with the constructive type-theoretic form per the lexicon.
For `<slug>_decoder.md`, the per-term format:
```markdown
## Term: <name>
- Original notation: ...
- Re-encoded: ...
- Form anchor: ...
- Etymology: ...
- Definition history: ...
- Source sections in original: ...
```
## 6. Connections
- **Upstream:** consumes the refined lexicon from the lexicon child
- **Downstream:** the apply child consumes the pilot's `pilot_report.md` (refinements) + the (unchanged) lexicon
## 7. Verification (4 criteria per deliverable)
- [ ] **Lossless** — every Pass 1 concept is represented in the de-obfuscated form
- [ ] **Bounded** — no `∞_val` or `∞_card`; all values are finite forms
- [ ] **Constructively typed** — every expression has a type
- [ ] **Etymology-cited** — every new term has a 1-line origin + 1-line definition history in the decoder
## 8. See also
- [Umbrella spec.md](../../video_analysis_deob_20260621/spec.md)
- [Umbrella README.md](../../video_analysis_deob_20260621/README.md)
- [Lexicon child spec.md](../../video_analysis_deob_lexicon_20260621/spec.md)
- Pass 1 reports: `cs229_building_llms` + `entropy_epiplexity`