# Track Specification: Video Analysis De-obfuscation Campaign (2026-06-21)

**Status:** Active (spec approved 2026-06-21)
**Initialized:** 2026-06-21
**Owner:** Tier 1 Orchestrator (umbrella spec + synthesis); Tier 2 Tech Lead (per-track execution)
**Priority:** A (user-blocking; Pass 2 of the 3-pass research campaign)
**Type:** Multi-track research campaign (1 warmup + 1 umbrella + 3 phase children = 5 folders total)
**Domain:** Meta-tooling (research deliverable + LLM operational spec; no `src/` changes)

> **Purpose.** This umbrella organizes Pass 2 of the user's 3-pass research campaign: **de-obfuscation** of the Pass 1 video reports via the user's constructive type-theoretic re-encoding DSL. The de-obfuscation reduces standard math notation + verbose DSL/verbiage into a bounded, constructive, type-theoretic form that bridges the conceptual gap and crystallizes the formal language into the reader's mind.

> **Multi-pass context.** Pass 1 produced 12 deep-dive reports (1000-10000 LOC each) + 1 cross-cutting synthesis. Pass 2 takes those and produces a multi-layer de-obfuscated version per video. Pass 3 (future, user-led) projects the de-obfuscated content to the user's applied domain (handmade/data-oriented/GPGPU + own caveats).

> **Companion docs.** The warmup track (`video_analysis_deob_warmup_20260621/`) is the precursor that produces the initial lexicon + LLM prompt template. The 3 phase children (`video_analysis_deob_{lexicon,pilot,apply}_20260621/`) consume the warmup's output and apply it to the Pass 1 reports.

---

## 1. Overview

### 1.1 The user's de-obfuscation philosophy (foundational)

The user curates knowledge unorthodoxy, especially formal math/sciences. Their position:

| Position | Take |
|---|---|
| **Form requires bounds** | "To be known is to project a form." Boundedness is required for direct knowledge. |
| **Indefinite is not directly knowable** | What is unbounded is indefinite; what is indefinite is indiscernible, unobserved, unsubject, unknowable. |
| **Cycles/iteration/repetition are allowed** | Indefinite *operations* on bounded *forms* are expressible. `Stream A = nat -> A` is fine; `∞_val` is not. |
| **The agent is bounded by necessity** | An agent is "envesseled in the soup of the universe," separated from the indefinite to discern. The agent cannot be indefinite. |
| **Standard math notation is "noise"** | Too compressed, error-prone, ASCII-hostile, not programmatic, not verifiable, not visualizable. Lots of synonyms that mean the same thing (Curry-Howard: proofs=programs, types=propositions, etc.). |
| **Constructive type theory is the foundation** | Proofs = programs (Curry-Howard); every value is a bounded form; operations are transformations. |
| **Lexicon is etymology-aware** | Each term's word origin + definitional history is documented. Words are chosen to match modern subjective experience. |
| **Inspiration** | Modern PL design — concatenative (Forth/KYRA/CoSy), data-oriented imperative (Lottes), immediate-mode DAG-building DSLs (O'Donnell's IMGUI). |

### 1.2 What Pass 2 produces

For each of the 12 Pass 1 reports + 1 cross-cutting synthesis, Pass 2 produces a **3-layer de-obfuscated deliverable**:

1. **Translation** (`<slug>_translation.md`) — side-by-side table: original expression ↔ re-encoded form
2. **Replacement** (`<slug>_deobfuscated.md`) — the re-encoded form replaces the original; the report is read as a bounded, constructive, type-theoretic document
3. **Decoder index** (`<slug>_decoder.md`) — per-term decoder: form anchor, etymology, definition history, link to the original section

Plus a per-track **pilot_report.md** or **apply_report.md** capturing lexicon refinements.

### 1.3 The 2-stage Pass 2 flow

```
Stage 1 (Warmup - precursor):                  Stage 2 (Apply - 3 phases):
                                                 ┌─ Phase 1: Lexicon (refine warmup's draft)
User's past notes ──► Warmup report.md +        │
                     prompt_template.md  ───────┤─ Phase 2: Pilot (apply to 2 videos, refine)
                                                 │
                                                 └─ Phase 3: Apply (apply to 10 + synthesis)
```

---

## 2. Current State Audit (as of 2026-06-21)

### 2.1 Already Available (DO NOT re-derive)

| Asset | Location | Use in Pass 2 |
|---|---|---|
| Pass 1 reports (12 + 1 synthesis) | `conductor/tracks/video_analysis_<slug>_20260621/report.md` + `summary.md` | The input to de-obfuscate |
| Pass 1 transcripts + OCR | `conductor/tracks/video_analysis_<slug>_20260621/artifacts/` | Source material for re-encoding context |
| `intent_dsl_survey_20260612` report | `conductor/tracks/intent_dsl_survey_20260612/report_v1.2.md` | Sibling DSL (a tool-verb DSL for AI agents); not the math re-encoding, but shares the philosophy |
| 4-tier vocab + 14-primitive grammar | `intent_dsl_survey_20260612/report_v1.2.md` §3, §4 | Reference for the PL-design vocabulary to use in the de-obfuscation DSL |
| `conductor/code_styleguides/error_handling.md` | Conductor docs | `Result[T]` convention for any new Python tooling |
| `conductor/code_styleguides/python.md` | Conductor docs | 1-space indent, type hints, no comments |
| Reference scripts (bootslop) | `C:\projects\forth\bootslop\*.py` | yt-dlp / cv2 / winsdk OCR — NOT needed for Pass 2 (no video processing) |

### 2.2 Gaps to Fill (this campaign's scope)

| # | Gap | Resolution |
|---|---|---|
| G1 | The user has no codified de-obfuscation DSL | Warmup produces `report.md` + `prompt_template.md` from the user's past samples |
| G2 | The de-obfuscation lexicon is not yet finalized | Phase 1 (lexicon) refines the warmup's draft into a codified spec |
| G3 | No pilot validation of the lexicon | Phase 2 (pilot) applies to 2 videos (1 foundational + 1 math-heavy) and captures refinements |
| G4 | No application to the remaining 10 + synthesis | Phase 3 (apply) applies the refined lexicon to the remaining Pass 1 outputs |
| G5 | No multi-layer deliverable structure | The 3-layer format (translation / replacement / decoder) is the new convention |

---

## 3. Goals

1. **Lexicon derived from the user's exemplars.** The de-obfuscation DSL is not invented from scratch; it is extracted from the user's past de-obfuscation notes via the warmup track. Evidence-based, not imposed.
2. **LLM-direct operational spec.** The de-obfuscation is performed by an LLM following the prompt template. The template is the "code" — the contract between the warmup and the apply phases.
3. **Lossless preservation (carries Pass 1's directive).** No Pass 1 concept is lost. The 3-layer output ensures every standard-math expression is represented (translation), replaced (replacement), and explained (decoder).
4. **Bounded, constructive, type-theoretic.** Every value is a bounded form. Iteration is explicit. "Infinity" is disambiguated lexically: `∞_val` (banned), `∞_proc` (allowed), `∞_card` (banned).
5. **Etymology + definitional history.** Each new term has a 1-line origin note + a 1-line definition history in the decoder.
6. **Multi-pass handoff.** Pass 3 (projection to applied domain) can consume the de-obfuscated outputs as its input. The handoff is clean: Pass 2 produces bounded, constructive forms; Pass 3 can apply them to the user's stylistic preferences.

---

## 4. Functional Requirements

### FR1. Umbrella folder + README

**WHERE:** `conductor/tracks/video_analysis_deob_20260621/`

**WHAT:** This folder contains the umbrella design (this spec) + 4 sibling files (`plan.md`, `metadata.json`, `state.toml`, `README.md`). The README is the index of the 4 sibling tracks (warmup + 3 phases) with their statuses.

### FR2. Warmup track (precursor)

**WHERE:** `conductor/tracks/video_analysis_deob_warmup_20260621/`

**WHAT:** Standalone research-style track. Produces:
- `report.md` — the design philosophy + the curated lexicon (terms + re-encodings) + the 3 noise-dedup maps (Curry-Howard-style collapses)
- `prompt_template.md` — the operational spec; an LLM can be prompted with this directly

**Inputs:** The user provides samples in `samples/` (their past de-obfuscation notes). Format: markdown, txt, or any text the user has.

**Process:** Tier 2 worker surveys the samples for term frequency, structural patterns, "form projection" heuristics, and noise-dedup maps. Produces a report + prompt template following the convention of `intent_dsl_survey_20260612/report_v1.2.md`.

**`blocked_by`:** none (user must provide samples before warmup can start; the user's action item is the FIRST dependency).

**`blocks`:** the 3 phase children (lexicon, pilot, apply) all depend on the warmup's output.

### FR3. Phase 1 — Lexicon refinement

**WHERE:** `conductor/tracks/video_analysis_deob_lexicon_20260621/`

**WHAT:** Consumes the warmup's `report.md` + `prompt_template.md`. Produces a codified `lexicon.md` (the operational spec for the de-obfuscation LLM) + `terms_catalog.md` (the machine-readable lexicon) + `dedup_map.md` (the 3 noise-dedup maps).

The lexicon refinement adds:
- Test cases (5-10 example transformations drawn from the user's samples)
- A "form anchor" requirement: each re-encoding must project from an indefinite to a bounded form
- Cross-references to the warmup's report sections

### FR4. Phase 2 — Pilot on 2 videos

**WHERE:** `conductor/tracks/video_analysis_deob_pilot_20260621/`

**WHAT:** Consumes the codified `lexicon.md` + 2 Pass 1 reports. The 2 pilot videos are:
1. `cs229_building_llms` (foundational ML/LLM coverage — wide scope, good test for "form projection" across many concepts)
2. `entropy_epiplexity` (math-heavy, focused on information-theoretic concepts — good test for "boundedness" + type-theoretic encoding of measure theory)

For each pilot video, produces the 3-layer deliverable in `artifacts/<slug>/`:
- `translation.md` (side-by-side: original ↔ re-encoded)
- `deobfuscated.md` (replacement: re-encoded form replaces the original)
- `decoder.md` (per-term decoder: form anchor, etymology, definition history)

Plus a `pilot_report.md` capturing:
- Lexicon refinements discovered during the pilot
- Concepts that didn't fit the lexicon (gaps)
- Process improvements for Phase 3

### FR5. Phase 3 — Apply to remaining 10 + synthesis

**WHERE:** `conductor/tracks/video_analysis_deob_apply_20260621/`

**WHAT:** Consumes the refined lexicon (from Phase 2) + 10 remaining Pass 1 reports + 1 cross-cutting synthesis. Produces the 3-layer deliverable for each, in `artifacts/<slug>/`.

Plus an `apply_report.md` capturing:
- Final lexicon v2
- Final process refinements
- Open questions for Pass 3

### FR6. Multi-layer deliverable structure (per video)

For each Pass 1 report, the de-obfuscation produces 3 files in `artifacts/<slug>/`:

**`<slug>_translation.md`** — side-by-side translation table:
```markdown
# Translation: <Video Title>

| # | Original Section | Original Expression | Re-encoded Form | Form Anchor |
|---|------------------|---------------------|-----------------|-------------|
| 1 | §2 Key Concepts | `set S = {x | P(x)}` | `kind S = {x : T | proof : P x}` | bounded set comprehension |
| 2 | §4 Transcript | `∀x ∈ ℝ: x² ≥ 0` | `forall x : Real, square x >= 0` | bounded quantification over Reals |
| ... |
```

**`<slug>_deobfuscated.md`** — the re-encoded report (replacement). Same 8-section structure as Pass 1's report, but every standard-math expression is replaced with the constructive type-theoretic form.

**`<slug>_decoder.md`** — per-term decoder:
```markdown
# Decoder: <Video Title>

## Term: Set
- Original notation: `S = {x | P(x)}`
- Re-encoded: `kind S = {x : T | proof : P x}`
- Form anchor: bounded set comprehension (a `kind` is a finite enumerated type)
- Etymology: "set" (Old English "settan" = to set, place); the word evokes "placement"
- Definition history: Cantor (1895) proposed unbounded set theory; the user rejects this in favor of bounded kinds
- Source sections in original: §2.1, §4.3, §5.7

## Term: Forall
- ...
```

### FR7. The de-obfuscation DSL (what the lexicon defines)

The lexicon produced by the warmup + refined by Phase 1 is the **de-obfuscation DSL**. It has:

| Component | Definition |
|---|---|
| **Terms** | The vocabulary: `kind`, `forall`, `exists`, `proof`, `program`, `type`, `bounded`, `stream`, `iterate`, `cycle`, `form`, `anchor`, etc. (the warmup discovers these from the user's samples) |
| **Grammar** | How terms combine. Inherits from constructive type theory: `term := term term | ( term ) | name : type | lambda name . term | ...` |
| **Noise-dedup map** | The 3 collapse maps: proofs=programs (Curry-Howard), types=propositions, sets=kinds, etc. |
| **Boundedness rules** | `∞_val` (banned), `∞_proc` (allowed as `Stream A = nat -> A`), `∞_card` (banned). Every value must be a bounded form. |
| **Form-anchor rule** | Every re-encoding must have a form anchor: "what bounded form does this project from the indefinite?" |
| **Etymology rule** | Every new term has a 1-line origin + 1-line definition history in the decoder. |
| **Verification rule** | The 4 verification criteria per §12: lossless, bounded, constructively typed, etymology-cited. |

### FR8. Dependency graph

```
UMBRELLA (video_analysis_deob_20260621)
│
├── Warmup (video_analysis_deob_warmup_20260621)
│       │
│       ▼ (warmup produces report.md + prompt_template.md)
│
├── Phase 1 (video_analysis_deob_lexicon_20260621) — consumes warmup
│       │
│       ▼ (Phase 1 produces lexicon.md + terms_catalog.md + dedup_map.md)
│
├── Phase 2 (video_analysis_deob_pilot_20260621) — consumes Phase 1
│       │
│       ▼ (Phase 2 produces 2 deobfuscated deliverables + pilot_report.md)
│
└── Phase 3 (video_analysis_deob_apply_20260621) — consumes Phase 2
        │
        ▼ (Phase 3 produces 11 deobfuscated deliverables + apply_report.md)
```

### FR9. Storage & naming

- All 5 new folders under `conductor/tracks/` (matching the user's directive: "just make new files for pass 2 in the same directories")
- The warmup's `samples/` is gitignored (user's past notes are local-only; not committed)
- The 3-layer deliverables are committed (research artifacts, the whole point of Pass 2)
- Reports and decoder files are committed
- Per-phase pilot/apply reports are committed

### FR10. No src/ changes

Pass 2 produces research artifacts (markdown files). It does NOT modify `src/*.py`, add `src/<thing>.py` files, or add new `pyproject.toml` deps. The only code that may be written is for tooling (a possible `scripts/deobfuscate/` namespace IF a Tier 3 worker finds that the prompt template alone is insufficient — this is a judgment call during Phase 1).

---

## 5. Non-Functional Requirements

- **TDD if code is written.** Any new Python tooling in `scripts/deobfuscate/` follows the same conventions as `scripts/video_analysis/` (Result[T], 1-space indent, type hints, no comments, tests in `tests/test_deobfuscate_*.py`).
- **Per-task atomic commits.** Each phase follows `conductor/workflow.md` per-task commit discipline.
- **Git notes.** Each task gets a git note summarizing what was done and why.
- **No day estimates.** Scope measured in files/sites per `conductor/workflow.md` Tier 1 Track Initialization Rules.
- **User-as-source dependency.** The warmup is blocked on the user providing samples. This is a USER action item, not a Tier 2/3 action.
- **Lossless preservation directive (carried from Pass 1).** No Pass 1 concept is lost in the de-obfuscation.

---

## 6. Out of Scope (Explicit)

- **Pass 3 (projection to applied domain).** Future, user-led. The de-obfuscated outputs of Pass 2 are Pass 3's input.
- **The user's "own caveats" (referenced in Pass 1's spec §11.2).** User must articulate these before Pass 3 starts. Out of scope here.
- **The math encoding notation design itself (without the user's exemplars).** Pass 2 is EVIDENCE-BASED — the lexicon is derived from the user's past work, not invented.
- **Interpreter for the de-obfuscation DSL.** Out of scope. The LLM is the executor; no interpreter is built.
- **Modifying `src/*.py` files in manual_slop.** Research-only campaign.
- **Adding `pyproject.toml` dependencies.** All work is research (markdown files).
- **Automated verification of the de-obfuscation's "correctness."** The 4 verification criteria (lossless, bounded, constructively typed, etymology-cited) are checked by Tier 3 + the user, not by automated tooling.

---

## 7. Architecture Reference

This campaign does not modify the manual_slop application architecture. It produces research artifacts. The architecture refs that DO apply:

- **Track convention:** `conductor/workflow.md` "Standard Task Workflow" + "Tier 1 Track Initialization Rules" + per-task commit discipline
- **Code style (if code is written):** `conductor/code_styleguides/python.md` + `conductor/code_styleguides/error_handling.md`
- **Research track precedent:** `conductor/tracks/intent_dsl_survey_20260612/` (research-style report + operational spec)
- **Campaign umbrella precedent:** `conductor/tracks/video_analysis_campaign_20260621/` (1 umbrella + N children at `conductor/tracks/` level)
- **Multi-pass framing (load-bearing):** `conductor/tracks/video_analysis_campaign_20260621/spec.md` §0, §11

---

## 8. Verification Criteria

The campaign is "done" when all of the following are true:

- [ ] Warmup shipped with `report.md` + `prompt_template.md` (and the user has approved the lexicon)
- [ ] Phase 1 (lexicon) shipped with `lexicon.md` + `terms_catalog.md` + `dedup_map.md`
- [ ] Phase 2 (pilot) shipped with 2 deobfuscated deliverables (each 3-layer) + `pilot_report.md` capturing refinements
- [ ] Phase 3 (apply) shipped with 11 deobfuscated deliverables (each 3-layer) + `apply_report.md` capturing final lexicon v2
- [ ] Each deobfuscated deliverable passes the 4 verification criteria:
  1. **Lossless** — every Pass 1 concept is represented in the de-obfuscated form (no dropped content)
  2. **Bounded** — no `∞_val` or `∞_card` in the output; all values are finite forms
  3. **Constructively typed** — every expression has a type; type-checking is mentally executable
  4. **Etymology-cited** — every new term in the deobfuscation has a 1-line origin + 1-line definition history in the decoder
- [ ] Umbrella `state.toml` updated to `status = "completed"`
- [ ] End-of-track report at `docs/reports/TRACK_COMPLETION_video_analysis_deob_20260621.md`
- [ ] All 5 folders move to `conductor/archive/` per the project's archiving convention
- [ ] `conductor/chronology.md` updated with 5 new rows

---

## 9. Risk Register

| ID | Title | Likelihood | Scope impact | Mitigation |
|---|---|---|---|---|
| R1 | User cannot provide samples in time | Medium | Warmup blocked | User can provide partial samples; warmup can use 1-2 examples as a starter |
| R2 | User's samples don't have enough de-obfuscation patterns (e.g., mostly raw notes) | Medium | Warmup produces a thin lexicon | Phase 1 (lexicon) extends the warmup's draft with constructive type theory defaults |
| R3 | Lexicon can't capture a concept in bounded form | Medium | Some concepts remain "indefinite — see original" | Document the gap; don't force a translation |
| R4 | Pilot reveals the lexicon is overfit to the user's style | Low | Refinement needed in Phase 2 | `pilot_report.md` captures gaps; Phase 3 uses the refined lexicon |
| R5 | The 3-layer deliverable format is too verbose for some videos | Low | Adjust per video | Format is a template, not a rigid structure; some sections may be smaller |
| R6 | Tier 2 attempts to invent the lexicon without the user's samples | Low (if user samples present) | Lexicon is invented, not evidence-based | Warmup spec is explicit: "consume user samples FIRST; lexicon is evidence-based" |
| R7 | Pass 3 needs the de-obfuscated outputs but Pass 2 isn't done | High (timeline) | Pass 3 blocked | This campaign's "lossless preservation" ensures Pass 3 has all the input it needs once Pass 2 ships |
| R8 | The user changes their mind about the philosophy mid-campaign | Low | Pilot reveals the shift; lexicon is updated | `pilot_report.md` is the checkpoint for user review |

---

## 10. User Directives (recorded for next agent / future-self)

- **2026-06-21:** "I have a very unorthodox take for how I curate knowledge, especially formal knowledge in the math and sciences." — Pass 2 is curation, not just translation.
- **2026-06-21:** "I like theurgy, I like some aspects of platonic thought... consistent time-invariant shared objective reference to similar patterns." — Platonism: shared reference through subjective lenses.
- **2026-06-21:** "I like Norman Wildberger's work. And I like the constructivist current progress on type theories." — Foundational: constructive type theory + Wildberger's algebraic finitism.
- **2026-06-21:** "I don't like the way indefinites/infinities/infinitesimals are defined or verbally utilized." — Boundedness required for direct knowledge.
- **2026-06-21:** "Infinite is okay well handled CORRECTLY. No observer or mechanism or construct can be infinite in resolution or quantification." — Cycles/iteration are fine; `∞_val` is not.
- **2026-06-21:** "I can provide samples of notes I've done but it will take time and might be best to leave to a 'warmup' track to gather and survey those." — Warmup is the precursor; the lexicon is evidence-based.
- **2026-06-21:** "Multi-layer for sure" (answer to Q1).
- **2026-06-21:** "Report + prompt template" (answer to Q3, Q4).
- **2026-06-21:** "Just make new files for pass 2 in the same directories" + "I like having that umbrella track similar to the campaign track" (answer to Q4) — 5 folders at `conductor/tracks/` level, hybrid umbrella.
- **2026-06-21:** "Without giving examples this is the best I can do to describe where I am." — The ideation has set the philosophy; the exemplars (warmup) provide the concrete terms.

---

## 11. See Also

- `conductor/tracks/video_analysis_campaign_20260621/spec.md` §0 (multi-pass framing) + §11 (Pass 2 handoff contract, now superseded by this spec)
- `conductor/tracks/intent_dsl_survey_20260612/report_v1.2.md` — the sibling DSL; shares the philosophy but is for tool verbs, not math re-encoding
- `conductor/tracks/nagent_review_20260608/` — research-track precedent
- `conductor/tracks/fable_review_20260617/` — research-track precedent
- `conductor/code_styleguides/agent_memory_dimensions.md` — 4 memory dimensions; Pass 2 produces a "knowledge" memory (per-dimension)
- `conductor/code_styleguides/knowledge_artifacts.md` — knowledge harvest pattern; relevant to the de-obfuscation's "durable" nature
- `conductor/workflow.md` "Tier 1 Track Initialization Rules" + "Tier 2 Autonomous Sandbox" — execution conventions
- `conductor/tier2/agents/tier2-autonomous.md` — Tier 2 agent directives (test runner, branch conventions, failcount)
- `conductor/tier2/commands/tier-2-auto-execute.md` — Tier 2 dispatch protocol