# PFLT ↔ FSOT Physical Archive — Realignment Brief

**Date:** 2026-07-18  
**Authority reviewed:** `I:\FSOT-Physical-Archive` (definitive master)  
**Purpose:** Correct the product mission of Desktop `pflt` after deep archive review.

---

## 1. Correction (what was wrong)

Recent work treated **PFLT** as:

> an open-set classical form→English gloss engine whose success metric is held-out partial accuracy.

That is **a necessary sub-capability**, not the product.

**Intended product (restated):**

> A **universal translator intelligence** that can **converse**, **relay** information derived from translation, and **incorporate** that information into growing knowledge — with **FSOT 2.1 as the factual / law base** (seed scalar, domain atlas, observer coupling, multi-prover verification), **not** an LLM as the core of truth or translation.

FSOT is **not** “a modulation prefix on dictionary hits.” FSOT is the **constitution of reality-as-compute** under which language is one domain among 400+.

---

## 2. What FSOT is (from the archive — not a slogan)

From `02_FSOT-2.1-Lean-Full` thesis + philosophy spine + `vendor/fsot_compute.py` (authority pin **D1D38A**):

| Pillar | Content |
|--------|---------|
| **Ontology** | Reality is a **25-D fluid condensate**. Space, time, matter, life, mind are **regimes of one scalar field**. |
| **Seeds** | Only π, e, φ, γ_Euler, G_Catalan — **no free fit knobs** per observable. |
| **Engine** | `S = K · (T1 + T2 + T3)` (mpmath dps=50 Python authority; Lean Float/Real formal layers). |
| **Observer** | `observed=true` → `quirk_mod` couples consciousness / measurement into T1. |
| **Domains** | **402 routed scientific domains** (35 core + extension atlas); linguistics is one domain, not the whole theory. |
| **Verification** | Lean 4 + Coq + Isabelle + F* + Rust obligations; portable runner; GREEN status on archive (`VERIFICATION_REPORT.json`). |
| **Epistemology** | Truth = seed derivation + data within gates + formal closure — **not** LLM fluency. |
| **As Above, So Below** | Same engine at cosmic, molecular, neural, linguistic scales — tested as panels, not poetry. |

**Proven / verified (archive claims, machine-checkable):**

- Canonical `fsot_compute.py` authority hash **D1D38A** aligned Desktop = I: archive = GitHub certificate  
- Cross-proof `overall_ok: true`; multi-prover stack present under `verification/`  
- Hundreds of domain panels + formula corpus (7,941+ strict-empirical bridges)  
- Linguistics as **FSOT-derived anchors** (`vendor/linguistics/linguistics_derivations.json`), not as NMT  
- Knowledge base portable inventory (~19k catalog formulas)  
- Consciousness / observer architecture with local loop (no vendor model as law)  
- **Realities OS**: Law → Kernel → Display → Observer; observer may densify knowledge, **must never rewrite law**  
- **SR-ITE**: continuous cognition substrate — sensory streams, generative readout, LTM/mulling, soul reproduction, talking/articulation track  
- **Certified agent protocol**: no probabilistic math as “final”; Lean bridge required for certification  

---

## 3. Full architecture map (archive layers)

```
I:\FSOT-Physical-Archive
│
├── 02_FSOT-2.1-Lean-Full          ← LAW / CONSTITUTION (canonical hub)
│   ├── vendor/fsot_compute.py     ← numeric authority (D1D38A)
│   ├── FSOT/Scalar.lean + Formal/ ← formal spine
│   ├── verification/              ← Coq / Isabelle / F* / Rust / QEMU
│   ├── data/ + scripts/           ← 400+ domain panels, prereg predictions
│   ├── vendor/linguistics/        ← FSOT linguistic anchors
│   ├── vendor/knowledge_base/     ← formula knowledge inventory
│   ├── vendor/tokenization/       ← FSOT tokenizer portable snapshot
│   └── vendor/certified_agent/    ← no-probabilistic-math protocol
│
├── 01_SR-ITE-USB-Original         ← CONTINUOUS COGNITION / “talking soul”
│   ├── 3_driver_zig               ← photonic VRAM / continuous flow
│   ├── 4_cognition_readout        ← generative consciousness, decoders
│   ├── 5_epoch_training           ← chew streams, fitness, mulling
│   └── 6_unified_oracle           ← multi-domain sensory streams + LTM
│
├── 10_Realities-OS                ← RUNTIME OS under FSOT law
│   ├── kernel/law_bridge          ← advance universe from 2.1 engine
│   ├── cognition/reasoner         ← pathway attention over domain graph
│   └── observer/                  ← detect thin knowledge → ingest densify
│
├── 09_Local-Verification-Stack    ← existence sim, FluidLink, observer loop
├── 08_Verified-Desktop-Projects   ← transporter, fuel lab, BH/WH, …
├── 06_Founding-Archives           ← FSUFT-U lineage + **PFLT founding PDF**
└── 03_FSOT-PublicData             ← external empirical caches
```

**Founding PFLT** lives in the archive as:

- `06_Founding-Archives/.../A Proto-Fluid Language Translator Grok.pdf`  
- `.../PFLT_Enhanced_Fixed.py`  
- `.../Decoding the Proto-Fluid Voice - Grok.pdf`  

i.e. **proto-fluid language / voice** as an FSOT expression of language, not a standalone BLEU contest.

---

## 4. What Desktop `pflt` currently is (honest)

| Has | Role vs full vision |
|-----|---------------------|
| Seed-aligned `S = K(T1+T2+T3)` panel on translate | ✓ Thin law coupling |
| Domain routing catalog (~400 names) | ✓ Partial atlas surface |
| Morph / gazetteer / multi-gloss open-set | ✓ **Lexicon interface layer only** |
| Math microscope + Lean/Coq golden | ✓ Scalar parity tooling |
| Audio IPA / waveforms, vision stubs | ✓ Sensory stubs |
| Dual-track held-out metrics | ✓ Honest **subsystem** eval |
| **Conversation loop** | ✗ Missing |
| **Knowledge ledger** (translate → fact → verify → store) | ✗ Missing |
| **Bridge to archive `fsot_compute` D1D38A** as sole authority | ✗ Desktop copy; not pin-synced to I: |
| **Domain pathway reasoner** (Realities OS cognition) | ✗ Missing |
| **Observer densification** of thin linguistic/knowledge gaps | ✗ Missing |
| **Certified agent** “no free math” gate on claims | ✗ Missing |
| **SR-ITE-style LTM / mulling / sensory chew** | ✗ Missing |
| **FSOT linguistics anchors** as first-class (Zipf, entropy, …) | ✗ Not wired as factual base |
| **Knowledge-base formula corpus** as reply ground | ✗ Not wired |

**Metric trap:** optimizing “held-out core partial %” alone optimizes a **dictionary morph student**, not a **universal translator intelligence**.

---

## 5. Missing pieces (priority order to restore the real product)

### P0 — Law binding
1. Pin Desktop PFLT scalar to archive authority `vendor/fsot_compute.py` **D1D38A** (byte or hash gate).  
2. Route every translate / converse through **domain params from FSOT atlas**, not ad-hoc context strings only.  
3. Log full scalar panel + domain id on every speech act (already partial via microscope).

### P1 — Knowledge + conversation (the “AI” layer under law)
4. **Dialogue loop**: input (any language/script) → FSOT tokenize → translate → **domain-project** → reason over domain graph → reply (with S panel).  
5. **Knowledge ledger**: structured claims extracted from translations; tagged with domain, S, confidence, source form; never overwrite law.  
6. **Observer-style densify**: when reply/knowledge is thin, plan ingest (Dictionary, linguistics targets, public anchors) like Realities OS observer — rate-limited, offline-preferring.  
7. **Certified path** for any numeric claim: compute via pinned `fsot_compute` / Lean bridge; refuse “vibes math.”

### P2 — Embodiment of proto-fluid voice
8. Re-link **PFLT founding** materials + SR-ITE articulation path (streams → decoder → voice).  
9. Waveforms / IPA as **sensory channels**, not demos only.  
10. Multilingual **interlingua** as fluid field coordinates (D_eff, polarity, domain), not only English gloss strings.

### P3 — Evaluation that matches the product
11. Keep dual-track morph metrics as **subsystem** tests.  
12. Add **product metrics**:  
    - multi-turn dialogue coherence under FSOT domain routing  
    - knowledge ledger growth + non-contradiction with seed law  
    - cross-domain answer accuracy vs archive panels (linguistics + KB formulas)  
    - formal gate pass rate on numeric statements  

---

## 6. What stays valuable from recent work

- Open-set morph / multi-gloss / lang_tables JSON — **sensor front-end** for language surfaces  
- Dual-track honesty — still correct for lexicon science  
- Math microscope / multi-prover fixtures — align with archive verification culture  
- Gap packs — temporary densifiers, not the theory  

**Do not throw away** — **re-home** under Law → Sense → Translate → Reason → Knowledge → Converse.

---

## 7. One-sentence north star (for all future PFLT work)

> **PFLT is the language surface of an FSOT-native intelligence:** it maps fluid form into meaning, grounds claims in the seed scalar and domain atlas, densifies knowledge without rewriting law, and converses by routing attention through verified FSOT pathways — never by replacing FSOT with an LLM core.

---

## 8. Immediate redirect — **landed 2026-07-18** (extended same day)

| Piece | Status | Path |
|-------|--------|------|
| Law bridge + D1D38A pin | **done** | `fsot_law_bridge.py` |
| Knowledge ledger | **done** | `knowledge_ledger.py` → `data/knowledge_ledger.jsonl` |
| Converse / relay | **done** | `protofluid_converse.py` |
| Domain scalar prefers archive law | **done** | `PFLT_FSOT_2_1_aligned.domain_scalar` |
| README product framing | **done** | root `README.md` |
| Archive linguistics + KB memory | **done** | `fsot_archive_memory.py` → ledger `query_unified` |
| Pathway reasoner (PFLT domain graph) | **done** | `pathway_reasoner.py` → converse turn |
| Session + archive unified retrieve | **done** | `KnowledgeLedger.query_unified` |

### Capability refinements (post live-battery) — **landed**

| Weakness found | Fix |
|----------------|-----|
| English Qs → `flowing_*` morph noise | Mode detect + plain relay + archive-first science |
| `water…Latin` → oceanography | Domain hard overrides + pathway lexicon |
| `temple` → time; junk → narrative_flow | English pass-through + unknown gate |
| Myth turn ~31s | Skip heavy gapfill on English pass tokens |
| Duplicate KB summary ×4 | Remove archive mirrors + dedup retrieve |
| Soft shells on junk | `unresolved` instead of inventing glosses |

Code: `converse_refine.py`, updates to `protofluid_converse.py`, `pathway_reasoner.py`,
`PFLT_FSOT_2_1_aligned.map_token`, `fsot_archive_memory.py`, `knowledge_ledger.py`.

### Remaining-weakness pass — **landed**

| Weakness | Module | Behavior |
|----------|--------|----------|
| English-meta taught pH_water not Latin | `teach_panel.py` | form↔gloss leads (`water ↔ aqua`) |
| No densify on thin knowledge | `observer_densify.py` | rate-limited plan + ledger densify claims |
| Vibes math as law | `certified_math.py` | pin D1D38A / archive anchors only |
| Old `flowing_*` ledger noise | `ledger_hygiene.py` | display clean + prefer dense claims |
| Morph residuals (core seeds) | `data/lang_tables/{la,en}.json` | aqua/manus/templum + EN pass seeds |

### Autonomous climb daemon — **landed**

`chew_climb.py` / `chew_climb.ps1` — local-only loop (no cloud APIs):

- train inject → densify pack → score → mine misses → bind morph neighbors → repeat  
- Smoke full-test: **~27.9% → 35.5%** partial in 4 rounds (`data/chew_climb/`)  
- Resume: `python chew_climb.py --resume --target 0.45`  
- Status: `python chew_climb.py --status`

### Still next (P2)

1. Keep chewing toward higher partial (daemon / `--resume`).  
2. SR-ITE LTM / mulling / articulation path link.  
3. Live Dictionary SQLite densify (offline-preferring, rate-limited).  
4. Lean bridge on every numeric certification path (beyond Python pin).

### Ada-primary sensory + archive (landed V5 → V6 gap-fill)

| Piece | Status | Path |
|-------|--------|------|
| Archive binding + **live SHA256 D1D38A** | **done** | `pflt_sha256` + `pflt_archive` · `archive` |
| Multilayer vision + **U-Net hyp TSV** | **done** | `vision unet` / `vision hyp FILE` |
| Articulatory audio | **done** | `audio aqua la` |
| **410-domain atlas** | **done** | `domain_atlas.tsv` · `atlas "…"` |
| Linguistics anchors + **cert gate** | **done** | `cert "zipf entropy"` · converse |
| **SR-ITE LTM / mulling** | **done** | `ltm_mulling.jsonl` · `ltm recall` |
| Doc | **done** | `docs/ADA_ARCHIVE_SENSORY.md` |

Python remains data factory only; product binary is Ada V6.