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

conductor(deob_apply): Initialize Phase 3 (apply) track with full scaffold

Browse Source 
The pilot (Phase 2) is shipped; Phase 3 is now unblocked and ready for Tier 2 dispatch.

5 new files in video_analysis_deob_apply_20260621/:
- spec.md: updated to reference the new files (lightweight scaffold)
- plan.md: 6-phase pipeline (init → read → apply A cluster → apply B cluster → apply C cluster → apply E+D+synthesis → final report + verify) with 25 tasks
- metadata.json: scope, 14 verification criteria, 5-item risk register, 10 user directives
- state.toml: 6 phases + 25 tasks + 10 verification flags + 11 user-directives-logged entries
- TIER2_STARTER.md: dispatch prompt with file-read order, the 2 user refinements (decompress names + operator reference), the 3 pilot process improvements, the 8 refinements + 5 gaps to apply, the 11 inputs (10 videos + 1 synthesis), when-stuck guide, copy-paste-ready block

CRITICAL context for Tier 2 (the 2 user refinements + 3 pilot improvements):
1. **Decompress names AND expressions** (per 2026-06-23): use DESCRIPTIVE names, NOT single letters. Multi-line constructions preferred.
2. **Use the operator reference** (report.md §9): 13 categories of operators with behavior + type signatures. The LLM should consult this when applying the de-obfuscation.
3. **3-column translation tables** (pilot improvement #1)
4. **Tier-categorized decoders** (pilot improvement #2)
5. **Split apply_report.md** into 3 sections (pilot improvement #3)

The 11 inputs: 10 remaining Pass 1 reports + 1 cross-cutting synthesis. Produces 34 deliverables (33 per-video 3-layer files + 1 apply report). This is the FINAL phase of Pass 2 — the result feeds Pass 3 (projection to applied domain, future, user-led).
This commit is contained in:
Edward R. Gonzalez
2026-06-23 16:32:22 -04:00
parent 59d048b51a
commit da84e800f8
5 changed files with 736 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
conductor/tracks/video_analysis_deob_apply_20260621/TIER2_STARTER.md
+403
View File
@@ -0,0 +1,403 @@
# Tier 2 Starter Prompt: Video Analysis De-obfuscation — Apply to 10 + Synthesis
**Purpose.** This file is the dispatch prompt for Tier 2 autonomous agents picking up the Phase 3 (apply) child track. It supplements the auto-loaded `spec.md` + `plan.md` per `conductor/tier2/commands/tier-2-auto-execute.md` step 2.
**Track:** `video_analysis_deob_apply_20260621` (Pass 2 of 3, Phase 3 of 3 within Pass 2, **the FINAL phase of Pass 2**)
---
## Track identity
- **ID:** `video_analysis_deob_apply_20260621`
- **Type:** Research-only child track (Phase 3 of Pass 2; the final application of the de-obfuscation scheme to the 10 remaining Pass 1 reports + 1 cross-cutting synthesis)
- **Status:** spec ✓ (lightweight scaffold); plan/metadata/state to be created by Tier 2
- **Priority:** A (user-blocking; the FINAL phase of Pass 2)
- **Domain:** Meta-tooling (research deliverable; no `src/` changes)
## Mission (what this track produces)
This Phase 3 child **applies the refined lexicon (informed by the pilot) to the remaining 11 inputs** (10 Pass 1 reports + 1 cross-cutting synthesis). It produces **34 deliverables**:
For each of the 10 remaining videos + 1 synthesis, 3 files in `artifacts/<slug>/`:
1. **`<slug>_translation.md`** — side-by-side table (3-column form per pilot process improvement #1; not 6-column)
2. **`<slug>_deobfuscated.md`** — the re-encoded report (replacement; same 8-section structure as Pass 1)
3. **`<slug>_decoder.md`** — per-term decoder (tier-categorized per pilot process improvement #2; not section-by-section)
Plus:
- **`apply_report.md`** — final lexicon v2, refinements, gaps, process improvements, open questions for Pass 3 (split into 3 sections per pilot process improvement #3)
The 11 × 3-layer deliverables + apply report are the input to Pass 3 (projection to applied domain, future, user-led).
---
## The 11 inputs (in execution order)
The order is by cluster (A → B → C → E → D → synthesis):
| # | Slug | Cluster | YouTube ID |
|---|---|---|---|
| 1 | `probability_logic` | A — math foundations | `0yF9TvMeAzM` |
| 2 | `score_dynamics_giorgini` | A — math foundations | `P75iVMmbqQk` |
| 3 | `platonic_intelligence_kumar` | B — Platonic AI | `1mXUFweWOug` |
| 4 | `free_lunches_levin` | B — Platonic AI | `K8BmMU1Tm-I` |
| 5 | `generic_systems_fields` | C — biological/cognitive | `QeMajYvhEbI` |
| 6 | `brain_counterintuitive` | C — biological/cognitive | `cDxtFtoQVNc` |
| 7 | `neural_dynamics_miller` | C — biological/cognitive | `0BS-BzEFTXA` |
| 8 | `multiscale_hoffman` | C — biological/cognitive | `YnfaT5APPB0` |
| 9 | `cs336_architectures` | E — Stanford | `lVynu4bo1rY` |
| 10 | `creikey_dl_cv` | D — applied | `yxkUvXs-hoQ` |
| 11 | `synthesis` | (cross-cutting) | (no YouTube ID; consumes all 12 videos) |
---
## Files to read in this order
### 1. This Phase 3 child track (REQUIRED)
- `./spec.md` (lightweight scaffold; read first)
- `./plan.md` (6-phase pipeline; 25 tasks)
### 2. The refined lexicon (PRIMARY INPUT — informed by the pilot)
- `/conductor/tracks/video_analysis_deob_lexicon_20260621/lexicon.md` (the codified operational spec; 12 sections + 4 appendices)
- `/conductor/tracks/video_analysis_deob_lexicon_20260621/terms_catalog.md` (machine-readable lexicon; 4-tier table with 9 columns per term)
- `/conductor/tracks/video_analysis_deob_lexicon_20260621/dedup_map.md` (6 noise-dedup maps; 3 principled + 3 user-preferred)
### 3. The pilot's refinements (PRIMARY INPUT)
- `/conductor/tracks/video_analysis_deob_pilot_20260621/pilot_report.md` (8 refinements + 5 gaps + 3 process improvements)
- `/conductor/tracks/video_analysis_deob_pilot_20260621/artifacts/cs229_building_llms/cs229_building_llms_translation.md` (the 6-column → 3-column template)
- `/conductor/tracks/video_analysis_deob_pilot_20260621/artifacts/cs229_building_llms/cs229_building_llms_deobfuscated.md` (the re-encoded template)
- `/conductor/tracks/video_analysis_deob_pilot_20260621/artifacts/cs229_building_llms/cs229_building_llms_decoder.md` (the tier-categorized decoder template)
- `/conductor/tracks/video_analysis_deob_pilot_20260621/artifacts/entropy_epiplexity/entropy_epiplexity_translation.md` (the math-heavy template)
### 4. The warmup's prompt template + operator reference (the LLM-direct operational spec)
- `/conductor/tracks/video_analysis_deob_warmup_20260621/prompt_template.md` (5 rules + 6 noise-dedup maps + 4-layer format + 7 example transformations + the **"Decompress names AND expressions" rule** + verification checklist)
- `/conductor/tracks/video_analysis_deob_warmup_20260621/report.md` §9 — the **indexed operator reference** (13 categories; arithmetic, comparison, logical, set-theoretic, type-theoretic, constructors, data-oriented, pipeline, sectors, type-class resolution, process, procedural/functional, why-this-exists). The LLM is expected to use this as a reference.
### 5. The 11 inputs to de-obfuscate
- `/conductor/tracks/video_analysis_<slug>_20260621/report.md` for each of the 10 remaining slugs
- `/conductor/tracks/video_analysis_synthesis_20260621/report.md` (the cross-cutting synthesis)
### 6. Project conventions (REQUIRED at session start per workflow.md)
- `/AGENTS.md` (critical anti-patterns, file naming, no day estimates)
- `/conductor/workflow.md` (task workflow, Tier 2 sandbox conventions, failcount contract)
- `/conductor/code_styleguides/python.md` (1-space indent, type hints, no comments — IF code is written; not applicable here)
- `/conductor/code_styleguides/error_handling.md` (Result[T] pattern — not applicable)
### 7. Reference tracks (consult as needed)
- `/conductor/tracks/video_analysis_deob_warmup_20260621/TIER2_STARTER.md` (sibling track)
- `/conductor/tracks/video_analysis_deob_lexicon_20260621/TIER2_STARTER.md` (sibling track)
- `/conductor/tracks/video_analysis_deob_pilot_20260621/TIER2_STARTER.md` (sibling track; the most-recent precedent)
---
## The 5 rules + the 7 verification criteria (load-bearing)
### The 5 rules (per `lexicon.md` §1)
1. **Boundedness** — every value is a finite form. `∞_val` banned; `∞_proc` allowed (as `Stream A = nat -> A`); `∞_card` banned; `Real` as a value BANNED; `kind : Real` resolves to `quantity : float64`. **The `while` operator is BANNED** (per `report.md` §9.12); use `for`, `iterate`, or `Stream` instead.
2. **Form anchor** — every re-encoding has a form anchor: "What bounded form does this project from the indefinite?"
3. **Etymology** — every new term has a 1-line origin + 1-line definition history.
4. **Lossless** — every Pass 1 concept is represented. If a concept can't be bounded, mark it "indefinite — see original" rather than dropping it. **Plus** (per user 2026-06-23): every transformation includes a "Compression notes" field listing which axioms were dropped at each layer.
5. **Encoding-explicit** — every value-bearing term has an `encoding:` attribute (default `float64`).
### The 7 verification criteria (per `lexicon.md` §0.1 + `prompt_template.md` verification checklist)
For each of the 33 deliverables (11 videos × 3 layers), verify:
- [ ] **Lossless** — no Pass 1 concept dropped; compression history preserved per layer.
- [ ] **Bounded** — no `∞_val` or `∞_card`; the "real number line" as a value is banned; no `while` loops.
- [ ] **Encoding-explicit** — every value-bearing term has an `encoding:` attribute. Default: `float64`.
- [ ] **Constructively typed** — every expression has a type signature.
- [ ] **Etymology-cited** — every new term has the 1-line origin + 1-line definition history.
- [ ] **Form-anchored** — every re-encoding has a form anchor.
- [ ] **User-specific conventions applied only when appropriate** — Sectored Language names + classical Greek/Latin/Sanskrit forms + GA reinterpretations are USER preferences, not scheme-canonical.
---
## The 3-layer deliverable structure (with the 2 user refinements)
For each video, the LLM produces 3 files in `artifacts/<slug>/`:
### Layer 1: `<slug>_translation.md` (3-column form per pilot process improvement #1)
| # | Original Expression | Re-encoded Form |
|---|--------------------|-----------------|
**Pilot process improvement:** 3-column form (instead of 6-column) to reduce visual clutter. The 6-column form is in the pilot's `cs229_building_llms_translation.md` for reference; the apply phase uses 3-column for consistency.
### Layer 2: `<slug>_deobfuscated.md` (the re-encoded report)
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 principled re-encoding per the lexicon.
### Layer 3: `<slug>_decoder.md` (tier-categorized per pilot process improvement #2)
**Pilot process improvement:** organized by tier (Tier 1-4) instead of by math section. The tier-categorized structure makes the principled/user-specific split clearer.
```markdown
## Tier 1: Core concepts
### Term: <name>
- **Original notation:** ...
- **Re-encoded:** ...
- **Form anchor:** the bounded form is X; the projection is Y
- **Etymology (1-line):** <origin>
- **Definition history (1-line):** <first formalization>
- **Source sections in original:** §X.Y
## Tier 2: Data-oriented pipeline
[similar structure]
## Tier 3: Type-theoretic primitives
[similar structure]
## Tier 4: AI-fuzzing tolerance
[similar structure, with user-also-accepted notes]
```
---
## The 2 user refinements (per 2026-06-23 feedback)
### Refinement 1: Decompress names AND expressions
**Per the user's feedback on the pilot:** the LLM was using single-letter variables (e.g., `p`, `W`, `H`, `K`) and compressed expressions (e.g., `p(X₁, …, X_L)`). The apply phase MUST use **descriptive names + multi-line constructions** instead.
Examples of the reframe:
- `p(X₁, …, X_L)` → `language_model(sequence : Token^L) -> Probability : float64`
- `W · h + b` → `output_projection = weight_matrix.matmul(hidden_state) + bias_vector`
- `H(X)` → `entropy(distribution : Probability_Distribution) -> Entropy : float64`
- `K(X)` → `kolmogorov_complexity(object : Object) -> Complexity : int64`
**The LLM is NOT afraid to translate expressions to multi-line definitions** when necessary. The de-obfuscation's job is to UNDO the original math's compression, not preserve it.
### Refinement 2: Use the operator reference (§9 in `report.md`)
**The warmup's `report.md` §9 is the indexed operator reference.** The LLM should consult this when applying the de-obfuscation. It covers 13 categories of operators (arithmetic, comparison, logical, set-theoretic, type-theoretic, constructors, data-oriented, pipeline, sectors, type-class resolution, process, procedural/functional, why-this-exists). Each operator has a symbol, name, behavior, type signature, and example.
The LLM should:
- Use the operators from this table (not conventional math compressed forms)
- When the LLM encounters a math expression, look up the relevant operators in §9
- Apply their principled re-encodings with descriptive names + multi-line constructions
---
## The pilot's 3 process improvements (per 2026-06-23)
1. **3-column translation table** (instead of 6-column) — the apply phase uses 3-column for all 11 translation.md files.
2. **Tier-categorized decoder** (instead of by math section) — the apply phase uses Tier 1-4 organization for all 11 decoder.md files.
3. **Split end-of-apply report** — the apply_report.md has 3 sections: refinements + gaps + process improvements.
---
## The principled vs user-specific distinction (the surgical-edits context)
Per the 2026-06-23 surgical edits + Phase 1's formalization:
- **The LLM produces the principled re-encoding by default** (the scheme-canonical form, from the 5 rules + the 3 principled noise-dedup maps).
- **The user-specific form is opt-in** (Sectored Language V1, GA reinterpretations, classical Greek/Latin/Sanskrit forms). The LLM produces the user-specific form only when the reader would prefer it.
- **The 4-language pattern is REQUIRED for user-also-accepted terms** (Greek + Latin + English + Sanskrit, per the warmup's `Notiones.txt` reference).
- **The 3 user-preferred noise-dedup maps are optional output conventions** (the GA collapse, the "invent → construct" rename, the "number = quantity = value" collapse).
- **Esoteric content (Witness/Vessel/Aether ontology) is NOT in the public lexicon.** It stays in `cluster_0_twitter.md` for the user's reference. **Do NOT include it in the apply deliverables.**
---
## The pilot's 8 refinements + 5 gaps (apply them)
The pilot surfaced these refinements + gaps. The apply phase should apply them when relevant.
**8 refinements (4 PILOT FIX, 4 DEFERRED to lexicon v2):** correlation, essentially-constant-as-Stream, Levin search, Markov chain, PRNG, poly-time adversary, support, self-delimiting.
**5 gaps (3 DEFERRED, 2 PILOT FIX):** paradox patterns, incomputable classification, honest epistemic hedging, type-class primitive, coinductive stream primitive.
If the apply phase encounters the same refinements in the 10 remaining reports, use the pilot's resolution. If it encounters NEW refinements/gaps, add them to `apply_report.md`.
---
## Critical user directives (load-bearing)
The warmup captured these in `state.toml` `[user_directives_logged]`; they apply to Phase 3 unchanged.
1. **Constructive type theory as foundation.**
2. **Boundedness for direct knowledge.**
3. **Cycles/iteration allowed but expressed explicitly** (`∞_val` banned, `∞_proc` allowed).
4. **Etymology-aware lexicon** (1-line origin + 1-line history; 4-language for user-also-accepted terms).
5. **Lossless preservation with explicit compression history** (per user 2026-06-23).
6. **Encoding-explicit** (every value-bearing term has an `encoding:` attribute; per user 2026-06-23).
7. **Honest epistemic hedging** (per the "Don't know what `<<` here is" pattern).
8. **The principled vs user-specific distinction** (the LLM produces the principled form; user-specific is opt-in).
9. **Secular sanitization** (esoteric content NOT in the pilot deliverables).
10. **Decompress names AND expressions** (per user 2026-06-23 feedback; descriptive names + multi-line constructions, not single-letter compressions).
11. **Use the operator reference** (per user 2026-06-23; the LLM should consult `report.md` §9 for operator behavior + type signatures).
---
## Key risks
- **R1 (medium):** LLM may produce inconsistent outputs across 11 videos. **Mitigation:** the pilot established the 7 verification criteria; the apply phase enforces them per-deliverable; apply_report.md captures inconsistencies as refinements.
- **R2 (high):** Cluster D (applied capstone: creikey_dl_cv) has too much practical detail (not math) to de-obfuscate. **Mitigation:** the lexicon's Tier 4 (AI-fuzzing tolerance) handles non-math content; the principled/user-specific distinction allows applied forms; the apply_report.md captures gaps.
- **R3 (medium):** Cluster C (biological/cognitive, 4 videos) requires deep biology/cognitive-science concepts. **Mitigation:** Rule 1 explicitly allows "indefinite — see original"; the apply_report.md captures these gaps.
- **R4 (high):** The cross-cutting synthesis is harder to de-obfuscate than the per-video reports. **Mitigation:** apply_report.md captures the synthesis-specific refinements; the synthesis is processed last (Phase 5) so it benefits from the per-video learnings.
- **R5 (low):** The 3-column translation table loses information compared to the 6-column form. **Mitigation:** the 3-column form is the default; the LLM can be prompted to produce the 6-column form for specific videos if requested.
---
## Hard constraints
- **No `src/*.py` changes.** Research-only track.
- **No `pyproject.toml` dependencies.** All work is research (markdown files).
- **No day estimates.** Scope measured in files/sites per `conductor/workflow.md`.
- **Per-task atomic commits.** Each deliverable (33 per-video + 1 apply report) is committed in its own commit with a git note.
- **Esoteric content NOT in deliverables.** Witness/Vessel/Aether stays in `cluster_0_twitter.md` only.
- **The 2 user refinements are NON-NEGOTIABLE.** Descriptive names + multi-line constructions (Refinement 1). Operator reference usage (Refinement 2).
---
## Tier 2 sandbox conventions (per `conductor/tier2/agents/tier2-autonomous.md`)
- **Test runner:** `uv run python scripts/run_tests_batched.py` (NEVER `uv run pytest` directly). Not applicable for this research-only track — no tests to run.
- **Default branch:** `master`. Use `origin/master` for `git fetch` and as the base for new branches.
- **Line endings:** preserve existing (CRLF stays CRLF, LF stays LF).
- **Throw-away scripts:** `scripts/tier2/artifacts/<track-name>/` (not the base `scripts/tier2/` directory). Not applicable here.
- **End-of-track report:** `docs/reports/TRACK_COMPLETION_<track-name>.md` per `conductor/tier2/agents/tier2-autonomous.md` step 42.
- **State update:** `state.toml` → `status = "completed"` at the end.
- **Hard bans:** `git push*`, `git checkout*`, `git restore*`, `git reset*` (3-layer enforced).
- **File access:** Tier 2 clone only. **NEVER USE APPDATA.**
- **Failcount contract:** After every task commit, check `should_give_up` from `scripts.tier2.failcount`. Thresholds: 3 consecutive red, 3 consecutive green, 30 min no progress.
---
## Verification criteria (gate for Phase 3 completion)
- [ ] All 10 videos + 1 synthesis have 3-layer deliverables (33 files total in `artifacts/`)
- [ ] All 3 deliverables per video pass the 7 criteria
- [ ] Translation tables are 3-column (pilot process improvement #1)
- [ ] Decoders are tier-categorized (pilot process improvement #2)
- [ ] `apply_report.md` has 3 sections: refinements + gaps + process improvements (pilot process improvement #3)
- [ ] The 2 user refinements (decompress names + operator reference) are applied to all 33 deliverables
- [ ] The 8 pilot refinements + 5 gaps are applied where relevant
- [ ] Final lexicon v2 captured in `apply_report.md`
- [ ] User has reviewed and approved
- [ ] All 34 deliverables (33 per-video + 1 apply report) committed atomically
- [ ] Git notes attached to each commit
- [ ] `state.toml` updated to `status = "completed"`
- [ ] `docs/reports/TRACK_COMPLETION_video_analysis_deob_apply_20260621.md` exists
---
## Execution plan (per apply plan.md)
| Phase | Task | Notes |
|---|---|---|
| 0 | Initialize | Create `state.toml` per Tier 2 conventions |
| 1 | Read inputs | 6 file reads: lexicon + terms_catalog + dedup_map + prompt_template + pilot_report + report.md §9 (operator reference) + 11 Pass 1 reports |
| 2 | A cluster (math foundations) | 2 videos × 3 deliverables (3-column + deobfuscated + tier-categorized decoder) |
| 3 | B cluster (Platonic AI) | 2 videos × 3 deliverables |
| 4 | C cluster (biological/cognitive) | 4 videos × 3 deliverables |
| 5 | E + D + synthesis | 3 × 3 deliverables (cs336 + creikey + synthesis) |
| 6 | Final report + verification | apply_report.md (3 sections) + user review + state update + end-of-track report |
---
## When stuck
- **LLM uses single-letter variables.** Re-prompt with the "Decompress names AND expressions" rule explicitly. The 4 example transformations in `prompt_template.md` are the canonical reference.
- **LLM output violates the principled/user-specific distinction.** Re-prompt: "Produce the principled re-encoding; the user-specific form is opt-in."
- **LLM can't bound a concept.** Apply Rule 1: mark "indefinite — see original" in the decoder. Capture the gap in `apply_report.md`.
- **LLM output is too verbose.** Re-prompt with a "be concise" instruction. The decoder is per-term + tier-categorized; the LLM should group by tier and skip terms that don't appear in the de-obfuscated.md.
- **Cluster D (creikey_dl_cv) has too much non-math content.** The Tier 4 (AI-fuzzing tolerance) handles non-math content. Mark unclear concepts as "indefinite — see original". Capture the gap in `apply_report.md`.
- **Synthesis is harder to de-obfuscate.** The synthesis is processed last (Phase 5) so it benefits from the per-video learnings. The apply_report.md has a dedicated section for synthesis-specific refinements.
---
## Quick reference: dispatch
```
/tier-2-auto-execute video_analysis_deob_apply_20260621
```
Plus this context (paste BEFORE invoking):
```
TRACK: video_analysis_deob_apply_20260621
TYPE: Research-only child track (Pass 2 Phase 3 of 3, the FINAL phase)
STATUS: spec ✓ (lightweight); plan/metadata/state to be created
PRIORITY: A (user-blocking; FINAL phase of Pass 2)
MISSION: Apply the refined lexicon (Phase 1) + the pilot's 8 refinements + 5 gaps + 3 process improvements + the 2 user refinements (decompress names + operator reference) to the 10 remaining Pass 1 reports + 1 cross-cutting synthesis.
PRODUCES: 34 deliverables (11 videos × 3-layer + 1 apply report)
CONSUMES: lexicon + terms_catalog + dedup_map + prompt_template + pilot_report + report.md §9 (operator reference) + 11 Pass 1 reports
CRITICAL — 2 USER REFINEMENTS (2026-06-23):
1. DECOMPRESS NAMES: use descriptive names, NOT single letters, unless universally obvious (e.g., x for input, f for function). Multi-line constructions are preferred over compressed expressions.
2. USE THE OPERATOR REFERENCE (report.md §9): arithmetic, comparison, logical, set-theoretic, type-theoretic, constructors, data-oriented, pipeline, sectors, type-class resolution, process, procedural/functional.
CRITICAL — 3 PILOT PROCESS IMPROVEMENTS:
1. 3-COLUMN translation tables (not 6-column)
2. TIER-CATEGORIZED decoders (Tier 1-4, not by math section)
3. SPLIT apply_report.md into 3 sections: refinements + gaps + process improvements
USER DIRECTIVES (load-bearing):
1. Constructive type theory as foundation
2. Boundedness for direct knowledge (while loop BANNED)
3. Cycles/iteration explicit; no "infinite" values
4. Etymology-aware lexicon (1-line origin + 1-line history; 4-language for user-specific)
5. Lossless with compression history
6. Encoding-explicit (every value-bearing term has encoding: attribute)
7. Honest epistemic hedging
8. Principled vs user-specific distinction (LLM produces principled; user-specific is opt-in)
9. Secular sanitization
10. Decompress names AND expressions
11. Use the operator reference
THE 11 INPUTS:
A cluster (math): probability_logic, score_dynamics_giorgini
B cluster (Platonic AI): platonic_intelligence_kumar, free_lunches_levin
C cluster (biological/cognitive): generic_systems_fields, brain_counterintuitive, neural_dynamics_miller, multiscale_hoffman
E (Stanford): cs336_architectures
D (applied): creikey_dl_cv
Synthesis: (cross-cutting)
FILES TO READ FIRST:
1. ./TIER2_STARTER.md (this file)
2. ./spec.md (the lightweight scaffold)
3. /conductor/tracks/video_analysis_deob_lexicon_20260621/lexicon.md (the codified operational spec; PRIMARY input)
4. /conductor/tracks/video_analysis_deob_lexicon_20260621/terms_catalog.md (machine-readable)
5. /conductor/tracks/video_analysis_deob_lexicon_20260621/dedup_map.md (6 noise-dedup maps)
6. /conductor/tracks/video_analysis_deob_pilot_20260621/pilot_report.md (8 refinements + 5 gaps + 3 process improvements; PRIMARY input)
7. /conductor/tracks/video_analysis_deob_warmup_20260621/report.md §9 (the indexed operator reference; PRIMARY input)
8. /conductor/tracks/video_analysis_deob_warmup_20260621/prompt_template.md (the LLM-direct operational spec; the "Decompress names" rule is in the "Your role" section)
9. The 11 Pass 1 reports (10 videos + 1 synthesis)
EXECUTION: Read inputs → apply lexicon to A cluster (2 videos) → B cluster (2) → C cluster (4) → E + D + synthesis (3) → write apply_report.md (3 sections) → user review → closeout.
```
---
## Post-Phase-3 (Pass 2 complete)
After Phase 3 ships, **Pass 2 is complete**. The 11 × 3-layer deliverables + apply_report.md are the input to **Pass 3 (projection to applied domain)**, which is the FINAL pass of the 3-pass research campaign.
Pass 3 is user-led:
- The user must articulate their "own caveats" (per Pass 1 spec §11.2)
- Pass 3 applies the bounded, constructive, type-theoretic forms to the user's preferred code style
- Influences: handmade/data-oriented/GPGPU (Timothy Lottes, Onat Türkçüoğlu, Jebrim)
- Pass 3's output is a future track (not in this campaign)
conductor/tracks/video_analysis_deob_apply_20260621/metadata.json
+163
View File
@@ -0,0 +1,163 @@
{
"track_id": "video_analysis_deob_apply_20260621",
"name": "Video Analysis De-obfuscation — Apply to 10 + Synthesis (Pass 2 Phase 3 of 3)",
"created": "2026-06-21",
"status": "spec_approved",
"blocked_by": [
"video_analysis_deob_pilot_20260621"
],
"blocks": [],
"priority": "A",
"rationale": "User-blocking Phase 3 (FINAL) of Pass 2. Consumes the pilot's refined lexicon + the 8 refinements + 5 gaps + 3 process improvements. Applies the lexicon to 10 remaining Pass 1 reports + 1 cross-cutting synthesis = 11 × 3-layer deliverables (33 files) + apply_report.md. Adopts the pilot's 3 process improvements: 3-column translation table, tier-categorized decoder, split end-of-apply report. Research-only; no src/ changes. This is the FINAL phase of Pass 2 — the result feeds Pass 3 (projection to applied domain, future, user-led).",
"type": "research-only child track (Pass 2 Phase 3 of 3, the final phase)",
"domain": "meta-tooling (research deliverable; no manual_slop src/ changes)",
"scope": {
"new_folders": [
"conductor/tracks/video_analysis_deob_apply_20260621/"
],
"new_files": [
"spec.md (lightweight scaffold)",
"plan.md",
"metadata.json",
"state.toml",
"TIER2_STARTER.md",
"artifacts/probability_logic/translation.md",
"artifacts/probability_logic/deobfuscated.md",
"artifacts/probability_logic/decoder.md",
"artifacts/score_dynamics_giorgini/translation.md",
"artifacts/score_dynamics_giorgini/deobfuscated.md",
"artifacts/score_dynamics_giorgini/decoder.md",
"artifacts/platonic_intelligence_kumar/translation.md",
"artifacts/platonic_intelligence_kumar/deobfuscated.md",
"artifacts/platonic_intelligence_kumar/decoder.md",
"artifacts/free_lunches_levin/translation.md",
"artifacts/free_lunches_levin/deobfuscated.md",
"artifacts/free_lunches_levin/decoder.md",
"artifacts/generic_systems_fields/translation.md",
"artifacts/generic_systems_fields/deobfuscated.md",
"artifacts/generic_systems_fields/decoder.md",
"artifacts/brain_counterintuitive/translation.md",
"artifacts/brain_counterintuitive/deobfuscated.md",
"artifacts/brain_counterintuitive/decoder.md",
"artifacts/neural_dynamics_miller/translation.md",
"artifacts/neural_dynamics_miller/deobfuscated.md",
"artifacts/neural_dynamics_miller/decoder.md",
"artifacts/multiscale_hoffman/translation.md",
"artifacts/multiscale_hoffman/deobfuscated.md",
"artifacts/multiscale_hoffman/decoder.md",
"artifacts/cs336_architectures/translation.md",
"artifacts/cs336_architectures/deobfuscated.md",
"artifacts/cs336_architectures/decoder.md",
"artifacts/creikey_dl_cv/translation.md",
"artifacts/creikey_dl_cv/deobfuscated.md",
"artifacts/creikey_dl_cv/decoder.md",
"artifacts/synthesis/translation.md",
"artifacts/synthesis/deobfuscated.md",
"artifacts/synthesis/decoder.md",
"apply_report.md (final lexicon v2 + refinements + gaps + process improvements + open questions for Pass 3)"
],
"modified_files": [],
"deleted_files": [],
"gitignored_patterns": []
},
"estimated_effort": {
"method": "scope (per conductor/workflow.md Tier 1 Track Initialization Rules). NO day estimates.",
"phase_0": "1 task: init state.toml",
"phase_1": "7 tasks: read 6 input files + adopt 3 process improvements",
"phase_2": "2 tasks: apply to A cluster (2 videos × 3 deliverables)",
"phase_3": "2 tasks: apply to B cluster (2 videos × 3 deliverables)",
"phase_4": "4 tasks: apply to C cluster (4 videos × 3 deliverables)",
"phase_5": "3 tasks: apply to E + D + synthesis (3 × 3 deliverables)",
"phase_6": "6 tasks: write apply_report.md (3 sections) + user review + state update + idempotency + audit + end-of-track report",
"summary": "6 phases, 25 tasks, 34 deliverables (33 per-video 3-layer files + 1 apply report). No day estimates per project convention. This is the FINAL phase of Pass 2."
},
"verification_criteria": [
"All 10 videos + 1 synthesis have 3-layer deliverables (33 files total in artifacts/)",
"All 3 deliverables per video pass the 4 criteria (Lossless, Bounded, Constructively typed, Etymology-cited)",
"All 3 deliverables per video pass the additional 3 criteria (Encoding-explicit, Form-anchored, User-specific conventions applied only when appropriate)",
"Translation tables are 3-column (pilot process improvement #1)",
"Decoders are tier-categorized (pilot process improvement #2)",
"apply_report.md has 3 sections: refinements + gaps + process improvements (pilot process improvement #3)",
"Final lexicon v2 captured in apply_report.md",
"User has reviewed and approved",
"All 34 deliverables (33 per-video + 1 apply report) committed atomically",
"Git notes attached to each commit",
"state.toml updated to status = 'completed'",
"End-of-track report at docs/reports/TRACK_COMPLETION_video_analysis_deob_apply_20260621.md",
"No new src/*.py files created (per AGENTS.md File Size and Naming Convention)",
"No new pyproject.toml dependencies"
],
"risk_register": [
{
"id": "R1",
"title": "The 11 LLM invocations produce inconsistent outputs (varying quality across videos)",
"likelihood": "medium",
"scope_impact": "the 11 deliverables may not be uniformly high-quality",
"mitigation": "the pilot established the 9 verification criteria; the apply phase enforces them per-deliverable; apply_report.md captures inconsistencies as refinements"
},
{
"id": "R2",
"title": "Cluster D (applied capstone: creikey_dl_cv) has too much practical detail (not math) to de-obfuscate",
"likelihood": "high",
"scope_impact": "the 'math' focus of the lexicon may not match the applied capstone's content",
"mitigation": "the lexicon's Tier 4 (AI-fuzzing tolerance) handles non-math content; the principled/user-specific distinction allows applied forms; the apply_report.md captures gaps"
},
{
"id": "R3",
"title": "Cluster C (biological/cognitive, 4 videos) requires deep biology/cognitive-science concepts that the lexicon can't capture",
"likelihood": "medium",
"scope_impact": "many concepts in Cluster C may be 'indefinite — see original'",
"mitigation": "Rule 1 explicitly allows 'indefinite — see original'; the apply_report.md captures these gaps; Pass 3 (future, user-led) may refine"
},
{
"id": "R4",
"title": "The cross-cutting synthesis is harder to de-obfuscate than the per-video reports (it spans all 5 clusters)",
"likelihood": "high",
"scope_impact": "the synthesis deliverable may be less precise than the per-video deliverables",
"mitigation": "apply_report.md captures the synthesis-specific refinements; the synthesis is processed last (Phase 5) so it benefits from the per-video learnings"
},
{
"id": "R5",
"title": "The 3-column translation table (pilot improvement) loses information compared to the 6-column form",
"likelihood": "low",
"scope_impact": "the user may want the 6-column form for some videos",
"mitigation": "the 3-column form is the default; the LLM can be prompted to produce the 6-column form for specific videos if requested"
}
],
"architecture_reference": {
"primary_documents": [
"conductor/workflow.md (track convention, per-task commits, git notes)",
"conductor/tracks/video_analysis_deob_20260621/spec.md (umbrella design)",
"conductor/tracks/video_analysis_deob_lexicon_20260621/lexicon.md (the codified operational spec)",
"conductor/tracks/video_analysis_deob_pilot_20260621/pilot_report.md (the 8 refinements + 5 gaps + 3 process improvements; PRIMARY input)"
],
"related_tracks": [
"conductor/tracks/video_analysis_deob_warmup_20260621/ (upstream precursor; the prompt_template.md is the LLM-direct operational spec)",
"conductor/tracks/video_analysis_deob_lexicon_20260621/ (upstream; the lexicon)",
"conductor/tracks/video_analysis_deob_pilot_20260621/ (upstream; the pilot's refined lexicon)",
"Pass 3 (future, user-led) — consumes the 11 deliverables + apply_report.md as input"
]
},
"deferred_to_followup_tracks": [
{
"title": "Pass 3: Projection to user's applied domain",
"description": "Applies the apply phase's de-obfuscated outputs to the user's preferred code style. Influences: handmade/data-oriented/GPGPU (Timothy Lottes, Onat Türkçüoğlu, Jebrim) + user's own caveats. Per Pass 1 spec §11.2.",
"track_status": "not started - blocked by this track",
"blocker_action_item": "User must articulate 'own caveats' before Pass 3 starts (per Pass 1 spec §11.2)"
}
],
"regressions_and_pre_existing_failures": [],
"pre_existing_failures_remaining": [],
"user_directives": [
"Constructive type theory as foundation (2026-06-21)",
"Boundedness for direct knowledge; cycles/iteration explicit (2026-06-21)",
"Etymology-aware lexicon (2026-06-21)",
"Lossless with explicit compression history (2026-06-23)",
"Encoding-explicit (every value-bearing term has encoding: attribute) (2026-06-23)",
"Honest epistemic hedging (2026-06-21)",
"Secular sanitization - esoteric content NOT in public lexicon (2026-06-23)",
"Principled vs user-specific distinction - Phase 1 formalized this (2026-06-23); Phase 2 preserved it; Phase 3 must preserve it",
"Pilot's 3 process improvements (2026-06-23): 3-column translation table, tier-categorized decoder, split end-of-apply report",
"No day estimates per conductor/workflow.md Tier 1 Track Initialization Rules (added 2026-06-16). Scope measured in files/sites only."
]
}
conductor/tracks/video_analysis_deob_apply_20260621/plan.md
+74
View File
@@ -0,0 +1,74 @@
# Plan: Video Analysis De-obfuscation — Apply to 10 + Synthesis
This is the Phase 3 (apply) child plan for Pass 2 of the 3-pass research campaign. Per the Tier 1 Track Initialization Rules, scope is measured in files/sites — no day estimates.
## Phase 0: Init
- [ ] **Task 0.1:** Initialize the Phase 3 child track: create `state.toml` (Tier 2).
## Phase 1: Read the refined lexicon (informed by the pilot)
- [ ] **Task 1.1:** Read `lexicon.md` (the codified operational spec; 12 sections + 4 appendices).
- [ ] **Task 1.2:** Read `terms_catalog.md` (the machine-readable form; 4-tier table with 9 columns per term).
- [ ] **Task 1.3:** Read `dedup_map.md` (the 6 noise-dedup maps; 3 principled + 3 user-preferred).
- [ ] **Task 1.4:** Read `pilot_report.md` (Phase 2's 8 refinements + 5 gaps + 3 process improvements).
- [ ] **Task 1.5:** Read the warmup's `prompt_template.md` (the LLM-direct operational spec; 5 rules + 6 noise-dedup maps + 4-layer format + 7 example transformations + verification checklist).
- [ ] **Task 1.6:** Read all 11 inputs: 10 Pass 1 reports + 1 cross-cutting synthesis.
- [ ] **Task 1.7:** Adopt the pilot's 3 process improvements:
- **3-column translation table** (instead of 6-column; per pilot's process improvement #1).
- **Tier-categorized decoder** (instead of by math section; per pilot's process improvement #2).
- **Split end-of-apply report** (refinements + gaps + process improvements; per pilot's process improvement #3).
## Phase 2: Apply to A cluster (math foundations)
- [ ] **Task 2.1:** Apply the lexicon to `probability_logic` (Pass 1 report). Produce 3 deliverables in `artifacts/probability_logic/`: translation (3-column), deobfuscated, decoder (tier-categorized). Commit per file with git note.
- [ ] **Task 2.2:** Apply the lexicon to `score_dynamics_giorgini`. Same structure. Commit per file.
## Phase 3: Apply to B cluster (Platonic AI)
- [ ] **Task 3.1:** Apply to `platonic_intelligence_kumar`. Commit per file.
- [ ] **Task 3.2:** Apply to `free_lunches_levin`. Commit per file.
## Phase 4: Apply to C cluster (biological/cognitive)
- [ ] **Task 4.1:** Apply to `generic_systems_fields`. Commit per file.
- [ ] **Task 4.2:** Apply to `brain_counterintuitive`. Commit per file.
- [ ] **Task 4.3:** Apply to `neural_dynamics_miller`. Commit per file.
- [ ] **Task 4.4:** Apply to `multiscale_hoffman`. Commit per file.
## Phase 5: Apply to E + D + synthesis
- [ ] **Task 5.1:** Apply to `cs336_architectures` (Stanford). Commit per file.
- [ ] **Task 5.2:** Apply to `creikey_dl_cv` (applied). Commit per file.
- [ ] **Task 5.3:** Apply to `synthesis` (cross-cutting). Commit per file.
## Phase 6: Final report + verification
- [ ] **Task 6.1:** Write `apply_report.md` (split into 3 sections per pilot's process improvement #3):
- **Refinements** — additional lexicon refinements discovered during the 11 applications (any beyond the pilot's 8).
- **Gaps** — concepts the lexicon couldn't fully capture (any beyond the pilot's 5).
- **Process improvements** — recommendations for Pass 3 (projection to applied domain).
- **Final lexicon v2** — the codified lexicon, updated with the pilot + apply refinements.
- **Open questions for Pass 3** — what Pass 3 needs to know.
- [ ] **Task 6.2:** User reviews the 11 × 3-layer deliverables + `apply_report.md`. Approves, or iterates (loop back).
- [ ] **Task 6.3:** Update `state.toml` to `status = "completed"`.
- [ ] **Task 6.4:** Idempotency check (re-run the LLM invocation for 1-2 videos, confirm the 3-layer output is consistent).
- [ ] **Task 6.5:** Audit checklist: 4 criteria met for all 11 deliverables; 3-column translation tables; tier-categorized decoders; lexicon v2 captured.
- [ ] **Task 6.6:** Write end-of-track report at `docs/reports/TRACK_COMPLETION_video_analysis_deob_apply_20260621.md`.
## Verification (gate per workflow.md)
- [ ] All 10 videos + 1 synthesis have 3-layer deliverables (33 files total in `artifacts/`)
- [ ] All 3 deliverables per video pass the 4 criteria (Lossless, Bounded, Constructively typed, Etymology-cited)
- [ ] All 3 deliverables per video pass the additional 3 criteria (Encoding-explicit, Form-anchored, User-specific conventions applied only when appropriate)
- [ ] Translation tables are 3-column (pilot process improvement #1)
- [ ] Decoders are tier-categorized (pilot process improvement #2)
- [ ] `apply_report.md` has 3 sections: refinements + gaps + process improvements (pilot process improvement #3)
- [ ] Final lexicon v2 captured in `apply_report.md`
- [ ] User has reviewed and approved
- [ ] All 34 deliverables (33 per-video + 1 apply report) committed atomically
- [ ] Git notes attached to each commit
- [ ] `state.toml` updated to `status = "completed"`
- [ ] End-of-track report at `docs/reports/TRACK_COMPLETION_video_analysis_deob_apply_20260621.md`
The Phase 3 child is "Pass 2 complete" when all 11 × 3-layer deliverables + apply report are committed + user-approved. Pass 3 (projection to applied domain, user-led) can then start.
conductor/tracks/video_analysis_deob_apply_20260621/spec.md
+7 -2
View File
@@ -52,7 +52,7 @@ Plus:
|---|---|
| `apply_report.md` | Final lexicon v2, final process refinements, open questions for Pass 3 |
**Optional (added per child track execution convention):** `plan.md`, `metadata.json`, `state.toml`.
**Optional (added per child track execution convention):** `plan.md`, `metadata.json`, `state.toml`. (These are now present in the folder, scaffolded at spec time for Tier 2 to consume. Per the pilot pattern, `plan.md` enumerates the 6-phase pipeline; `metadata.json` is the scope/risk register; `state.toml` is the task tracker.)
**Total deliverable file count:** 11 × 3 + 1 = 34 files (plus the apply report).
@@ -60,7 +60,7 @@ Plus:
Per the umbrella spec §5 (Phase 4 of the umbrella). Each phase commits atomically.
- [ ] **Phase 1: Init.** Initialize the child track.
- [ ] **Phase 1: Init.** Initialize the child track (Tier 2 reads the scaffolded `plan.md` + `metadata.json` + `state.toml`).
- [ ] **Phase 2: Apply to A cluster (2 videos: probability_logic, score_dynamics_giorgini).** 2 × 3-layer deliverables. Commit per video.
- [ ] **Phase 3: Apply to B cluster (2 videos: platonic_intelligence_kumar, free_lunches_levin).** 2 × 3-layer deliverables. Commit per video.
- [ ] **Phase 4: Apply to C cluster (4 videos: generic_systems_fields, brain_counterintuitive, neural_dynamics_miller, multiscale_hoffman).** 4 × 3-layer deliverables. Commit per video.
@@ -97,4 +97,9 @@ The 11 deliverables + `apply_report.md` are the input to Pass 3 (projection to a
- [Umbrella spec.md](../../video_analysis_deob_20260621/spec.md)
- [Umbrella README.md](../../video_analysis_deob_20260621/README.md)
- [Pilot child spec.md](../../video_analysis_deob_pilot_20260621/spec.md)
- [Pilot child pilot_report.md](../../video_analysis_deob_pilot_20260621/pilot_report.md) (the 8 refinements + 5 gaps + 3 process improvements; PRIMARY input)
- [Lexicon child lexicon.md](../../video_analysis_deob_lexicon_20260621/lexicon.md) (the codified operational spec)
- [Warmup prompt_template.md](../../video_analysis_deob_warmup_20260621/prompt_template.md) (the LLM-direct operational spec; contains the "Decompress names AND expressions" rule)
- [Warmup report.md §9](../../video_analysis_deob_warmup_20260621/report.md) (the indexed operator reference)
- [Pass 1 campaign spec §0, §11](../../video_analysis_campaign_20260621/spec.md) — multi-pass framing
- [TIER2_STARTER.md](./TIER2_STARTER.md) (the dispatch prompt for Tier 2)
conductor/tracks/video_analysis_deob_apply_20260621/state.toml
+89
View File
@@ -0,0 +1,89 @@
# Track state for video_analysis_deob_apply_20260621
# Updated by Tier 2 Tech Lead during execution
[meta]
track_id = "video_analysis_deob_apply_20260621"
name = "Video Analysis De-obfuscation - Apply to 10 + Synthesis (Pass 2 Phase 3 of 3, FINAL)"
status = "active"
current_phase = 0 # Phase 0 = init
last_updated = "2026-06-23"
[blocked_by]
video_analysis_deob_pilot_20260621 = "shipped 2026-06-23 (8f64127f)"
[blocks]
# Pass 3 (projection to applied domain, future, user-led) is the next user-led track
[phases]
phase_0 = { status = "pending", checkpointsha = "", name = "Init (state.toml)" }
phase_1 = { status = "pending", checkpointsha = "", name = "Read the refined lexicon (informed by the pilot)" }
phase_2 = { status = "pending", checkpointsha = "", name = "Apply to A cluster (math foundations): probability_logic, score_dynamics_giorgini" }
phase_3 = { status = "pending", checkpointsha = "", name = "Apply to B cluster (Platonic AI): platonic_intelligence_kumar, free_lunches_levin" }
phase_4 = { status = "pending", checkpointsha = "", name = "Apply to C cluster (biological/cognitive): generic_systems_fields, brain_counterintuitive, neural_dynamics_miller, multiscale_hoffman" }
phase_5 = { status = "pending", checkpointsha = "", name = "Apply to E + D + synthesis: cs336_architectures, creikey_dl_cv, synthesis" }
phase_6 = { status = "pending", checkpointsha = "", name = "Final report + verification" }
[tasks]
# Phase 0 (init)
t0_1 = { status = "pending", commit_sha = "", description = "Initialize Phase 3 child track: create state.toml per Tier 2 conventions" }
# Phase 1 (read refined lexicon)
t1_1 = { status = "pending", commit_sha = "", description = "Read lexicon.md (the codified operational spec; 12 sections + 4 appendices)" }
t1_2 = { status = "pending", commit_sha = "", description = "Read terms_catalog.md (the machine-readable form; 4-tier table with 9 columns per term)" }
t1_3 = { status = "pending", commit_sha = "", description = "Read dedup_map.md (the 6 noise-dedup maps; 3 principled + 3 user-preferred)" }
t1_4 = { status = "pending", commit_sha = "", description = "Read pilot_report.md (Phase 2's 8 refinements + 5 gaps + 3 process improvements)" }
t1_5 = { status = "pending", commit_sha = "", description = "Read the warmup's prompt_template.md (the LLM-direct operational spec; 5 rules + 6 noise-dedup maps + 4-layer format + 7 example transformations + verification checklist)" }
t1_6 = { status = "pending", commit_sha = "", description = "Read all 11 inputs: 10 Pass 1 reports + 1 cross-cutting synthesis" }
t1_7 = { status = "pending", commit_sha = "", description = "Adopt the pilot's 3 process improvements: 3-column translation table, tier-categorized decoder, split end-of-apply report" }
# Phase 2 (apply to A cluster)
t2_1 = { status = "pending", commit_sha = "", description = "Apply the lexicon to probability_logic. Produce 3 deliverables in artifacts/probability_logic/: translation (3-column), deobfuscated, decoder (tier-categorized). Commit per file with git note." }
t2_2 = { status = "pending", commit_sha = "", description = "Apply the lexicon to score_dynamics_giorgini. Same structure. Commit per file." }
# Phase 3 (apply to B cluster)
t3_1 = { status = "pending", commit_sha = "", description = "Apply to platonic_intelligence_kumar. Commit per file." }
t3_2 = { status = "pending", commit_sha = "", description = "Apply to free_lunches_levin. Commit per file." }
# Phase 4 (apply to C cluster)
t4_1 = { status = "pending", commit_sha = "", description = "Apply to generic_systems_fields. Commit per file." }
t4_2 = { status = "pending", commit_sha = "", description = "Apply to brain_counterintuitive. Commit per file." }
t4_3 = { status = "pending", commit_sha = "", description = "Apply to neural_dynamics_miller. Commit per file." }
t4_4 = { status = "pending", commit_sha = "", description = "Apply to multiscale_hoffman. Commit per file." }
# Phase 5 (apply to E + D + synthesis)
t5_1 = { status = "pending", commit_sha = "", description = "Apply to cs336_architectures (Stanford). Commit per file." }
t5_2 = { status = "pending", commit_sha = "", description = "Apply to creikey_dl_cv (applied capstone). Commit per file." }
t5_3 = { status = "pending", commit_sha = "", description = "Apply to synthesis (cross-cutting). Commit per file." }
# Phase 6 (final report + verification)
t6_1 = { status = "pending", commit_sha = "", description = "Write apply_report.md (split into 3 sections per pilot process improvement #3): (a) Refinements beyond the pilot's 8, (b) Gaps beyond the pilot's 5, (c) Process improvements for Pass 3 + final lexicon v2 + open questions for Pass 3" }
t6_2 = { status = "pending", commit_sha = "", description = "User reviews the 11 × 3-layer deliverables + apply_report.md. Approves or iterates (loop back)" }
t6_3 = { status = "pending", commit_sha = "", description = "Update state.toml to status = 'completed'" }
t6_4 = { status = "pending", commit_sha = "", description = "Idempotency check (re-run the LLM invocation for 1-2 videos, confirm the 3-layer output is consistent)" }
t6_5 = { status = "pending", commit_sha = "", description = "Audit checklist: 4 criteria met for all 11 deliverables; 3-column translation tables; tier-categorized decoders; lexicon v2 captured" }
t6_6 = { status = "pending", commit_sha = "", description = "Write end-of-track report at docs/reports/TRACK_COMPLETION_video_analysis_deob_apply_20260621.md" }
[verification]
all_11_videos_3layer = false
all_4_criteria = false
all_7_criteria = false
3_column_translation = false
tier_categorized_decoder = false
apply_report_committed = false
user_approved = false
state_toml_completed = false
end_of_track_report_committed = false
pass_2_complete = false
[user_directives_logged]
constructive_type_theory = "Per user 2026-06-21: 'I like Norman Wildberger's work. And I like the constructivist current progress on type theories as a foundational system.'"
bounded_for_knowledge = "Per user 2026-06-21: 'No observer or mechanism or construct can be infinite in resolution or quantification. To have distinction must have a bounds.'"
cycles_iteration_allowed = "Per user 2026-06-21: 'Infinite is okay well handled CORRECTLY... What can be indefinite is that can be subjected upon is that of cycles, that of iteration, that of repetition.'"
etymology_aware = "Per user 2026-06-21: etymology + definitional history documented"
lossless_compression_history = "Per user 2026-06-23: explicit compression notes per layer in the de-obfuscation's 3-layer output"
encoding_explicit = "Per user 2026-06-23: every value-bearing term has an encoding: attribute (default float64)"
honest_epistemic_hedging = "Per user 2026-06-21: 'Don't know what `<<` here is' / 'Me fucking around' style annotations are valid"
secular_sanitization = "Per user 2026-06-23: esoteric content (Witness/Vessel/Aether, classical philosophy) NOT in public deliverables"
principled_vs_user_specific = "Per user 2026-06-23 surgical edits: Phase 1 formalized this; Phase 2 preserved it; Phase 3 must preserve it"
pilot_process_improvements = "Per pilot 2026-06-23: 3-column translation table, tier-categorized decoder, split end-of-apply report"
no_day_estimates = "Per conductor/workflow.md Tier 1 Track Initialization Rules (added 2026-06-16). Scope measured in files/sites only."