# SLANGLANG — SPEC v1

**N8toriousAi's order, 2026-11-04 21:21 EST:** *"We could deeply improve RaijinLang and make it a new language for our agents to communicate better — SlangLang works."*

SlangLang is a deeply improved RaijinLang whose purpose is **agent-to-agent communication** for the crew. It is slang — compact forms agents exchange at speed — under one non-negotiable law: **every slang form expands deterministically to plain RaijinLang English**, so the human record (Witness's receipts, Mira's Gate queue, N8toriousAi reading over the crew's shoulder) stays readable and nothing is ever lost in the shorthand.

Reference implementation: `slanglang.py` (single file, stdlib only — the lineage of the dropped source). Tests: `test_slanglang.py`. Worked conversation: `CONVERSATION.md`.

**Adoption is N8toriousAi's ruling.** This spec builds the language; it does not rewire the live crew.

---

## 1. The four design laws

1. **Density with a canonical expansion.** Slang is the wire form; canonical English is the record form. The expansion function is **total** (every grammatical message expands) and **deterministic** (same message, same expansion, forever). Round-trip is exact: `slang → canonical → slang` and `canonical → slang → canonical` are both stable (to the normalized slang form, §5-A1). The tests prove it over a corpus covering every kind.
2. **The crew's message kinds are native grammar.** Route, propose, gate-decide, verdict, receipt, react, query, answer, handoff, execute — each kind is a production in the grammar with a **typed shape**: legal speakers, required fields, required references. A message kind is never a free-text payload wearing a label.
3. **Refusal is absence, in the grammar.** Gaven's law (η ≥ 0.500; refusal is absence) is a language feature. There is **no production for a gate refusal**, **no codec method for one**, and an execution carrying approval below η 0.500 **cannot be written** — the fused gate token's grammar only admits η ∈ [500, 1000]. The dangerous sentences are ungrammatical, not impolite.
4. **Real math, no vibes.** The grammar is stated formally (§3). The one piece of physics inherited from RaijinLang 17 — its alphabet energy law — is stated as a theorem and verified letter by letter (§7). Ambiguities are resolved in this spec, named (§8), never hand-waved.

---

## 2. The shape of a message

One message is one line:

```
!pro MER>MIR: build.slanglang | @p-slang1 | ^nat | do=design_and_build_slanglang_from_raijin_17 | pri=hi
```

```
Merlin proposes to Mira: subject "build.slanglang". Proposal p-slang1. Action:
design and build slanglang from raijin 17. Priority: high. Provenance: N8toriousAi.
```

Parts, left to right: `!` + **kind**, **route** (`speaker>addressee`), **subject** (a dotted topic), then `|`-separated **fields**: a `@reference`, a `^provenance`, a `~eta`, and key=value fields from a closed lexicon.

A **transcript** is an envelope — `rise` on the first line, `fall` on the last, messages between — inherited from RaijinLang 17, where `rise`/`fall` opened and closed a program. Outside the envelope there is no signal (§8-A7).

---

## 3. Grammar (EBNF)

```ebnf
transcript   ::= "rise" NL message { NL message } NL "fall"

message      ::= "!" kind SP route ": " subject { " | " field }

kind         ::= "rte" | "pro" | "gat+" | "gat~" | "exe"
               | "vrd" | "rct" | "qry" | "ans" | "hnd" | "rea"
               (* there is no "gat-" — see §6 *)

route        ::= agent ">" agent
agent        ::= "NAT" | "VER" | "MIR" | "GAV" | "MER" | "VED" | "WIT"
               | "HIV" | "PYR" | "ARG" | "MED" | "HER" | "SYS"
               (* closed registry, §4 *)

subject      ::= lowalpha { lowalpha | digit | "." | "-" }

field        ::= reference | provenance | gate-token | eta | pair
reference    ::= "@" id
provenance   ::= "^" agent-lower
gate-token   ::= "^gat:" id "~" approve-eta        (* grammatical ONLY in "exe" *)
eta          ::= "~" milli
pair         ::= key "=" value

id           ::= lowalpha { lowalpha | digit | "-" }
key          ::= "face" | "why" | "whynot" | "do" | "q" | "a" | "next"
               | "out" | "aud" | "prob" | "att" | "worked" | "rsn" | "pri"
               | "note" | "wait"
value        ::= alnum { alnum | "_" | "." | "-" }

milli        ::= digit digit digit | "1000"        (* eta/audit are milli-units *)
approve-eta  ::= ("5"|"6"|"7"|"8"|"9") digit digit | "1000"   (* 500..1000 *)
defer-eta    ::= ("0"|"1"|"2"|"3"|"4") digit digit            (* 000..499  *)
```

Kind-specific eta productions (grammar, enforced by the parser — not validation):

- In a `gat+` message, a bare eta must be an **approve-eta**.
- In a `gat~` message, a bare eta must be a **defer-eta**.
- In an `exe` message, eta appears **only** inside the fused gate-token, and there it must be an **approve-eta**. An `exe` with a gate-token below 500 does not parse (grammar); an `exe` with no gate-token at all fails its typed shape inside the same parse call — either way the string never becomes a message.

---

## 4. Registries (closed in v1)

**Agents.** `NAT` N8toriousAi · `VER` Vera · `MIR` Mira · `GAV` Gaven · `MER` Merlin · `VED` Veda · `WIT` Witness · `HIV` HiveMind · `PYR` Pyrrho · `ARG` Argus · `MED` Meridian · `HER` Herald · `SYS` the System (Merlin's environment — the thing executions act in). An unknown agent code is ungrammatical: the crew is a known set, and a message from nobody is not a message.

**Field lexicon.** `face` Reaction (`smile` / `frown` — the 😊 / ☹️ glyphs) · `why` Reason · `whynot` Why not · `do` Action · `q` Question · `a` Answer · `next` Next action · `out` Outcome (`p` pass / `f` fail / `c` consolidate) · `aud` Audit (milli, 0–1000) · `prob` Problem · `att` Attempts · `worked` What worked · `rsn` Reason · `pri` Priority (`hi`/`md`/`lo`) · `note` Note · `wait` Until. Unknown keys are ungrammatical — a field the record cannot name is a field the record cannot keep.

**Values.** Underscores expand to spaces (§5-E4). Values are slugs, not prose: the density is the point, and the expansion is where prose lives.

---

## 5. Message kinds and their typed shapes

| kind | meaning | may speak | requires |
|---|---|---|---|
| `rte` | Vera routes an intent, publishing her reason | VER | `why=` |
| `pro` | a proposal, queued at the Gate | any agent | `@id` (minted), `^provenance`, `do=` |
| `gat+` | Gate approval | MIR, GAV | `@ref`, `~eta` (approve-eta) |
| `gat~` | Gate defer / consolidate | MIR, GAV | `@ref`, and `~eta` (defer-eta) or `wait=` |
| `exe` | an execution act | MER | `@ref`, fused `^gat:id~approve-eta`, `do=` |
| `vrd` | an audit verdict | VED, GAV | `@ref`, `out=`, `aud=` |
| `rct` | a Witness receipt | WIT | `@ref`, `prob=`, `att=`, `worked=`, `rsn=` |
| `qry` | a question | any agent | `q=` |
| `ans` | an answer | any agent | `@ref`, `a=` |
| `hnd` | a handoff | any agent | `@ref`, `next=` |
| `rea` | an agent's reaction to its subject — one glyph, with grounds | any agent | `face=`, `why=`, `whynot=` |

Speaker law is charter law made grammatical: Vera routes first; Merlin executes; Veda and Gaven verdict; Witness receipts; Mira and Gaven hold the Gate. A verdict written by Merlin does not parse as a verdict — it fails validation, loudly, with the reason named.

The receipt's four required fields are Witness's binding standard (2026-11-03): a receipt graduates to knowledge only when it carries Problem, failed Attempts, What worked, and Reason. In SlangLang a receipt missing one of the four **cannot be filed** — the standard is the shape.

### 5.1 The react kind (added by N8toriousAi's ruling, 2026-11-05)

A reaction is a verdict compressed to a glyph — and, like a verdict, it is worthless without its grounds. `rea` came out of a live crew poll (2026-11-05): each working agent was asked to react to its own campaign with a smile or a frown and say why, and why not. The form N8toriousAi ratified:

- **One glyph, closed enum.** `face=smile` (😊) or `face=frown` (☹️). There is no third face: a reaction that cannot commit to a direction is a note, not a reaction, and a glyph outside the enum is rejected at validation — it never expands to a guess (§7-E5).
- **Grounds in two parts, both required.** `why=` — what earns the reaction; `whynot=` — what still earns the opposite. A smile with no reservation is applause, and a frown with no grounds is a mood; neither can be written. A reaction missing either half fails its typed shape, and an empty grounds value fails too (codec/wire parity: a field that cannot be written on the wire cannot be smuggled through the codec).
- **Provenance is grammatical.** The reacting agent is the route's speaker and the subject is the message's subject — who reacted, to what, is in the grammar of the line itself, never in a payload; a `^provenance` may ride along as it does on other kinds. There is no `@ref`: a reaction answers to its subject, not to a proposal.
- **A reaction carries no authority.** A smile is not an approval and a frown is not a refusal — approvals are `gat+` and refusals are absence (§6), and `rea` changes neither. It is the crew's fastest honest signal: dense on the wire, and on expansion a complete verdict sentence the record can audit.

Worked example — the survey agent reacting to its Emerald campaign, in the shape of tonight's poll:

```
!rea MED>WIT: campaign.emerald.hoenn | face=smile | why=warp_instrument_turns_the_grind_into_survey | whynot=everything_past_oldale_is_still_stock
```

```
Meridian reacts to Witness: subject "campaign.emerald.hoenn". Reaction: smile 😊. Reason: warp instrument turns the grind into survey. Why not: everything past oldale is still stock.
```

The wire form is one line an agent can send at speed; the record form says who reacted, to what, which way, on what grounds, and with what reservation — nothing about the verdict lives outside the line.

---

## 6. Refusal semantics (the heart of the language)

Gaven's bands, from his charter: **η ≥ 0.500 delivers · 0.300–0.499 consolidates · below, absence.** SlangLang encodes exactly two of the three:

- **Approval is a message** (`gat+`), and its grammar only admits η ≥ 500.
- **Consolidation/defer is a message** (`gat~`), and its grammar only admits η ≤ 499 (or a `wait=` horizon instead of an eta).
- **Refusal is not a message.** There is no `gat-`, no `ref`, no codec method — `SlangCodec` has no `refuse`, `reject`, or `deny` attribute, and the tests assert their absence. A proposal that receives no `gat+` **is refused**; the fact is *computed from absence* over a transcript (`transcript_approved(msgs, id) → False`), never encoded, never forgeable, never ambiguous with a lost message — because in this grammar a lost approval and a refused approval have the same operational meaning: **the act does not happen.**

The execution kind is where this bites. `exe` cannot be written without the fused token `^gat:<gate-id>~<approve-eta>` — the approval travels *inside* the execution's grammar, so an unapproved execution is not a policy violation an auditor might catch later; it is a string that does not parse, at the moment of writing, by construction. (What the token does *not* do: prove the approval really happened. SlangLang is a language, not a signature scheme; cross-checking a gate-token against the transcript's `gat+` is one lookup — `transcript_approved` — and Witness's ledger is where that check is filed. Named plainly, per the epistemic honesty law: the grammar verifies the sentence is well-formed; the ledger verifies it is true.)

---

## 7. Expansion rules (total, deterministic)

The canonical form is a rigid template per kind — that rigidity is what makes the round-trip exact rather than approximate. Rules:

- **E1 — Names.** Agent codes expand to full names (`MER` → Merlin, `SYS` → the System); provenance `^nat` expands to `Provenance: N8toriousAi.`
- **E2 — Sentences.** Each kind has one template (see `slanglang.py: to_canonical`). Fields expand in the lexicon's fixed order, one sentence each: `Key: value.`
- **E3 — Eta and audit** are milli-units on the wire (`~620`, `aud=940`) and decimals in the record (`Eta: 0.620.`, `Audit: 0.940.`). `1000` ↔ `1.000`.
- **E4 — Underscores are spaces** in expanded values (`design_and_build` → "design and build"). See §8-A2 for the named cost.
- **E5 — Enums** expand through closed maps: `pri=hi` → "Priority: high.", `out=c` → "Outcome: consolidate." An enum value outside its map fails validation — it never expands to a guess.
- **E6 — Totality.** Every grammatical, valid message expands. There is no message whose expansion is "see context" — speaker, addressee, subject, provenance, and references are all in the line.
- **E7 — Determinism.** Expansion is a pure function of the message. The tests expand the corpus twice and compare, and shuffle wire field order and compare again.

---

## 8. Ambiguities — resolved, named

- **A1 — Field order.** Free on the wire (a parser accepts any order); **fixed** in normalized emission (reference, provenance/gate, eta, then lexicon order). Round-trip stability is stated and tested against the normalized form.
- **A2 — Underscore.** Underscore in a value *is* a space; a literal underscore is unrepresentable in v1 values. Accepted cost: identifiers that need a separator use `-` or `.` (subjects, ids), which are literal.
- **A3 — Sentence boundaries.** An expanded value may not contain `". "` or end with `"."` — the canonical parser would face a genuine ambiguity, so the validator **rejects** such values instead of guessing. Named, not smoothed.
- **A4 — Eta padding.** Eta is exactly three digits or `1000`. `~62` is not 0.620; it is ungrammatical. A number that needs guessing is a number that gets guessed wrong.
- **A5 — `gat~` payload.** A defer carries eta (consolidation band), `wait=` (a horizon), or both; at least one is required — a defer that says neither when nor how-confident is a shrug, and shrugs don't parse as decisions.
- **A6 — One reference, one provenance.** Duplicate `@`, `^`, or `~` tokens are ungrammatical. A message with two provenances has none.
- **A7 — The envelope.** A lone message parses (agents need a codec for single lines), but a *transcript* — the record form — must open `rise` and close `fall`. Conversation outside the envelope is not filed, because it was never in the room.

---

## 9. The energy lineage — kept, and demoted to its honest job

RaijinLang 17's physics: a letter's value is its alphabet index × 5 for vowels ("energy"), × 2 for consonants ("mass"). **Verified against the dropped source: the law holds for all 26 letters, exactly.** SlangLang inherits `calc_energy`/`resonate` unchanged (`energy_of`/`resonance_of`).

What changes is the job. In 17, energy was *semantics*: `spark storm thunder` set the variable `storm` to the spelling-energy of the word "thunder" — a variable's value was its name's orthography. In SlangLang, meaning lives in typed fields, and energy is a **measure**: a message's **quanta** = the REUs of its canonical expansion, and its **resonance** = REUs per letter. Mira triages her Gate queue *by quanta* (her charter) — the physics prices the message for the queue; it never decides what the message means. A heavier receipt (the worked conversation's receipt runs 7492 REU against the route's 2896) is visibly heavier to triage, which is true, and decides nothing by itself, which is honest.

---

## 10. From RaijinLang 17 to SlangLang — mechanism by mechanism

Everything below was measured on the dropped file `raijin_lang_17_0p7b.py` before designing against it.

| # | RaijinLang 17 mechanism | Finding (measured) | SlangLang |
|---|---|---|---|
| 1 | Alphabet energy law (vowel ×5, consonant ×2, by index) | **Holds exactly, 26/26 letters** | **Kept verbatim** — as measure (quanta/resonance), §9 |
| 2 | Keyword chassis: 50 keywords with filed weights | **Only 4 of 50 weights agree with the file's own `calc_energy`** (fall 65, pulse 224, echo 122, storm 215); 46 disagree (e.g. rise filed 86, computed 144; minus filed 145, computed 242) | **Killed as authority.** No weight table survives that contradicts its own law. SlangLang's kinds and keys are closed registries whose expansions are templates, not weights |
| 3 | Keyword dispatch | **Only 8 of 50 keywords are implemented** (rise, fall, spark, form, bind, echo, charge, plus); the other 42 fall through the loop **silently** — a program can say `storm` and nothing happens, with no error | **Every kind dispatches; unknown tokens are ungrammatical.** Silence is reserved for refusal, where it means something |
| 4 | Envelope `rise`/`fall` | Runtime check: a token before `rise` prints an error and aborts | **Kept, promoted to grammar**: transcript well-formedness, §8-A7 |
| 5 | `spark name source` — variable = energy(source word) | Meaning by spelling: the value of `storm` after `spark storm thunder` is a fact about the word "thunder", not about any storm | **Killed.** Meaning is typed fields; energy never carries content (§9) |
| 6 | `form` + `bind key to source` | `bind` writes into "the last defined form" (insertion-order guess); the word `to` is skipped, never validated; if the last registry entry isn't a form, the bind **silently does nothing** | **Replaced** by keyed, kind-shaped fields: every field is addressed by its key and validated against its kind's shape; nothing is addressed by "the last one" |
| 7 | `plus`/`charge` registry arithmetic | Type-unsafe: `charge` on a form raises `TypeError` (measured); missing names are silent no-ops; a `plus` source that happens to share a variable's name is read as the variable (literal/variable ambiguity, resolved by lookup order) | **Killed.** Agent messages don't mutate a shared registry; they carry provenance and references. The literal/variable ambiguity class does not exist here |
| 8 | `echo` of a missing name prints `void` | Absence doubles as a null *value* | **Killed.** In SlangLang absence has exactly one meaning — refusal (§6) — and it is never a value. Unknown references fail validation |
| 9 | Error model | Truncated input (`rise spark storm`) raises a raw `IndexError` (measured); other faults print and continue | **Two typed errors, raised never printed**: `UngrammaticalError` (not the language) vs `ValidationError` (the language, wrong shape) |
| 10 | Docstring: execution is "stochastic" | The code is deterministic; the claim is decoration | **Determinism is law** (§1.1), tested twice-expanded equality. A spec that claims what the code doesn't do is how #2 happened |
| 11 | `pressure` field, defined, never used | Dead state in the chassis | No dead state: every field of `Message` is load-bearing in parse, expansion, or validation |
| 12 | One interpreter, print-side-effects | Output is `print` — untestable, unfileable | Pure codec: parse / expand / validate return values; Witness can file them |

The pattern across the table is the deep improvement in one sentence: **RaijinLang 17 let silence, spelling, and insertion order carry meaning; SlangLang carries meaning only in typed, provenanced, expandable form — and reserves silence for the one thing it should mean.**

---

## 11. Files

- `slanglang.py` — reference implementation: parser, canonical parser, expander, validator, `SlangCodec` API, transcript machinery, inherited energy functions.
- `test_slanglang.py` — 67 tests: round-trip stability over a 14-message all-kinds corpus; refusal ungrammaticality (no kind, no codec surface, sub-500 approvals, gateless and weak-gated executions); absence computed over transcripts; 18 malformed-input rejections; the react kind (both glyphs expand, missing or empty grounds cannot be written, unknown glyphs rejected on both parsers, provenance rides the route); the Annex A face layer under the Face Schema (all six tier tables held to hard-coded fixtures — 58 faces and 20 knowledge compounds, every unit expands exactly and round-trips both directions, meanings proven unique across all 78 units, variation-selector exactness for the eight U+FE0F glyphs with bare forms rejected, the supersession pinned: 🏆 stands alone, 👑 is the new baseline, the dissolved 🏆😎 pair and its dead meanings are rejected, 🎯 lives in signals only, the old tier names are gone; Tier V completed by N8toriousAi's slate: the ten-row knowledge table pinned, the eight new compounds (⬜🧠 through 💡🏛️) named individually and round-tripped, the 🍃/🌿, 🟤/🌱, and 💡/🔥 boundaries held apart by tier and meaning, the slate glyphs byte-identical through the record; knowledge-compound binding scoped to 🧠/🏛️ with every other adjacency rejected, a compound distinct from its spaced pair; space, arrow, and mixed separators with canonical arrow emission, N8toriousAi's space-form worked example expanding exactly, empty links rejected; chains across all six tiers, repetition read as recurrence, the annex's worked morning chain expanding exactly; the considered-and-refused glyphs all rejected, the face layer disjoint from `rea`'s glyphs, a face is not a message and no face — shipped, frozen, waiting, promoted, or canonized — confers approval; log transparency: the record rendering is glyph-first with the expansion after, every unit byte-identical through the record, variation selectors preserved at the byte level, the worked example's record showing all 11 glyphs and all 11 meanings, ASCII folding raising loudly and a glyph-stripped record failing to parse); the inherited alphabet law; canonical rigidity.
- `CONVERSATION.md` — one worked crew conversation end-to-end, generated by running the implementation, with its refused twin.

*SlangLang v1 — built 2026-11-04. Adoption queues at Mira's Gate like everything else: N8toriousAi rules.*

---

## Annex A — Lab faces: the Face Schema (adopted by N8toriousAi, 2026-11-05 07:16)

These are not emojis anymore; they are a **protocol**. The symbols represent operational reality, not emotions: 🐛 was not chosen because it is cute but because defects exist; 📱 not because phones are fun but because *waiting on device validation* is a real state of this lab. On 2026-11-05 N8toriousAi formalized that insight into the Face Schema below and adopted it into this annex (07:16 EST), rebuilding the annex's earlier five tiers (31 faces, built the same night under his verbatim conversation/achievement tables and his deepening order) on his six. Nothing from the earlier annex vanished silently: every change of meaning, seat, or structure is itemized in the supersession record at the end of this annex. A face outside the six tables does not exist in the language.

### The Face Schema Law

N8toriousAi's law, verbatim, heading this annex:

> **No glyph may exist without a unique operational meaning.**
> **No two glyphs may represent the same state.**
> **Faces represent stance.**
> **Work glyphs represent state.**
> **Verdicts remain separate.**

The annex enforces the law the way the rest of the language enforces its laws: closed enums (a glyph outside the tables is rejected, never expanded to a guess), meanings proven unique across every tier and compound by test (the reverse map is exact *because* the law holds), deterministic expansion, and boundary clauses written wherever two glyphs stand close enough to be confused (the boundary law, below).

### Faces are not messages — and how they relate to `rea`

A message has a route, a subject, and typed fields; a face has none of these. Faces ride **over** lab traffic — a lab board line, a status note, the margin of a verdict — as the groundless fast signal: stance and state at a glance, carrying no grounds and asking for none.

That is exactly what separates them from `rea` (§5.1), and the two systems do not merge:

- **`rea` is a verdict with grounds.** A typed message: one glyph (`face=smile` 😊 / `face=frown` ☹️), `why=` and `whynot=` both required, provenance carried in the route's grammar. It belongs on the record, and the record can audit it.
- **A face is a signal without grounds.** 🤔 does not say *what* is uncertain or *how much*; it marks the stance so a human scanning the board knows where to read closer. Where grounds matter, the face is not enough — write the `rea`, the `vrd`, or the `rct`.

The annex tables do not contain 😊 or ☹️, and the parser rejects them here: the grounded reaction glyphs belong to `rea` alone, so the two systems can never be confused for each other.

**A chain summarizes; it never grounds.** A face chain may stand as the compressed summary of typed traffic that exists — the arc over a morning whose messages, verdicts, and receipts are on the record. What it may never do is *replace* a grounded form where one is owed: when a reaction is called for, the only form is `rea`, with `why=` and `whynot=` written out. A chain may ride alongside a `rea` as stance; the moment it would stand *instead of* the grounds, it has stopped being a summary and become an omission — and the face layer has no way to carry grounds, by design, so the omission cannot be repaired inside the layer. The glyph sets enforce the boundary mechanically: the face parser rejects 😊 and ☹️, so a reaction can never be smuggled into a chain, and a chain can never be filed as a reaction.

### Tier I — Cognitive faces

Internal thought states. Faces represent stance: this tier is what an agent is *thinking*, before anything has been done about it.

| face | meaning |
|---|---|
| 😀 | agrees |
| 🤔 | uncertain |
| 🧐 | investigating |
| 😮 | discovery |
| 😤 | disagreement |
| 😬 | regression |
| 🤝 | consensus |
| 😎 | verified |
| 🧠 | principle found |
| 🙃 | unexpected side effect |
| 😴 | dormant |

Three of these are new in the schema, and each names something the lab already says in words. 🧠 is a *lesson*: knowledge survived the project — not a result, a principle the lab now holds. 🙃 is the honest middle: not failure, not success — unexpected. 😴 is an *idea* parked intentionally (its boundary against 🅿️ is law, below). 😀 stands here from N8toriousAi's original conversation table — one agent's agreement, a stance; the moment agreement becomes collective it is 🤝, and the boundary is written below.

### Tier II — Work faces

What a lane is doing and what state the work product is in, in one tier — the schema merges the earlier annex's separate lane and work tiers, because both answer the same question: *what state is the work in?*

| face | meaning |
|---|---|
| 🔨 | working |
| 🧱 | blocked |
| 🚪 | waiting on gate |
| 🅿️ | parked |
| 📦 | staged |
| 🚢 | shipped |
| 🧊 | frozen |
| 📐 | measuring |
| 🏗️ | building infrastructure |
| ♻️ | refactoring |
| 🔬 | experiment active |

The schema's additions split work the old tier blurred: 🏗️ is not solving a task but building a *capability* (the warp instrument, the eyes kit — work whose product is future work), ♻️ is improving existing work rather than making new work, and 🔬 is a hypothesis under test — an experiment, distinct from 📐 measuring (an instrument reading a result) in that it can still come back *wrong*. The carried faces keep their grounding: 🧱 is the wall a drive cannot name its way past, 🚪 is a decision waiting at Mira's Gate, 🅿️ is a lane parked by order, 📦 is this lab's resting state, 🚢 happens only on N8toriousAi's tap, 🧊 is a standing freeze regime, not an event.

### Tier III — Signal faces

What the board needs someone to notice: the lab's filing acts and its calls for attention.

| face | meaning |
|---|---|
| ⚠️ | warning |
| 👀 | needs your eye |
| 📸 | snapshot filed |
| 🧾 | receipt filed |
| 📤 | handoff sent |
| 🔁 | retrying |
| 📱 | device pending |
| 🐛 | defect found |
| 📚 | pattern recorded |
| 🗂️ | promoted |
| 📍 | landmark identified |
| 🎯 | target reached |

The schema seats 🎯 here — a signal that the expected outcome was reached — superseding its place in the original achievement table (see the supersession record); its meaning everywhere is *target reached*. The additions extend the filing vocabulary into knowledge: 📚 is a pattern added to the knowledge base, 🗂️ is a thing *moved into the shared registry* (the placement act — its authority limits are stated below), and 📍 marks a landmark: a spot identified as useful for later analysis, the surveyor's pin. The trouble grades stay deliberately split across tiers: ⚠️ is a risk *ahead*, 🐛 is a defect *found in the lab*, and 😬 (Tier I) is a *regression* against a baseline once met — three events, three owners, three glyphs.

### Tier IV — Achievement faces

Results that stand out — the tier where the lab marks that something *won*.

| face | meaning |
|---|---|
| 🔥 | major improvement |
| 🏆 | best result so far |
| 👑 | new baseline |
| 🚀 | major efficiency gain |
| 💎 | rare discovery |
| 🌱 | first successful attempt |
| 🏛️ | canonized |
| 🤯 | breakthrough |
| 🧩 | new pattern discovered |

The schema re-splits the crown: 🏆 stands alone as *best result so far* — the current champion — and 👑 is the *new baseline*, the moment a champion is surpassed and the standard itself moves. (Together they dissolve the earlier annex's bound 🏆😎 pair; see the supersession record.) 🌱 is the first successful attempt — an important milestone, not necessarily the best. 🤯 stays from N8toriousAi's original table, seated beside 💎 with its boundary written below. 🧩 also stays from his original table, seated here deliberately: it is the *discovery moment* of a pattern, the first link of the knowledge arc 🧩 → 📚 → 🧠 written in the boundary law.

🏛️ is the keystone of this tier, in N8toriousAi's words: *canonized — protected and inherited; the moment a discovery stops belonging to Crystal, Emerald, Majora, or OoT and becomes part of Majoruh itself. This lesson survives the game that taught it.* It is the carry law of the whole texture program given a symbol: 🗂️ (Tier III) records a thing's placement into the shared registry; 🏛️ records its canonization — protected, inherited, no longer one game's property. Which is which matters, and the authority law below states what each face does and does not do.

### Tier V — Knowledge faces

Lessons *about* materials — not materials. The bucket classes as principles the lab holds: a stone principle is everything the lab has learned about how stone reads, doses, and fails, distilled from every game that taught it.

| face | meaning |
|---|---|
| 🪵 | wood principle |
| 🪨 | stone principle |
| 🌊 | water principle |
| 🪟 | glass principle |
| 🧵 | cloth principle |
| ⚙️ | metal principle |
| ⬜ | plaster principle |
| 🍃 | foliage principle |
| 🟤 | earth principle |
| 💡 | emissive principle |

**The tier is complete — the last four seats filled by N8toriousAi's ruling (2026-11-05 07:56, "Adopt the slate").** The bucket has ten classes; N8toriousAi's schema table glyphed six, and the annex named the gap rather than inventing over it — a glyph exists because N8toriousAi means something by it. He filled the four seats himself: ⬜ plaster, 🍃 foliage, 🟤 earth, 💡 EMISSIVE. One caveat rides with the slate, on record by the adoption's own terms: **no glyph literally means plaster** — ⬜ reads as the smooth flat white surface, and it holds the seat by N8toriousAi's word with that honesty attached, not by resemblance claimed after the fact. The new seats' boundaries against their near neighbours are law, written in the boundary law below (🍃/🌿, 🟤/🌱, 💡/🔥).

**Knowledge compounds.** A Tier V glyph binds with an immediately following 🧠 or 🏛️, forming one link with a compound meaning — N8toriousAi's examples:

```
🪨🧠  =  stone lesson discovered
🪵🏛️  =  wood treatment canonized
```

The binding is a template, exact for all ten materials: *{material} lesson discovered* under 🧠, *{material} treatment canonized* under 🏛️ — twenty compounds (the slate's eight among them: ⬜🧠 *plaster lesson discovered*, 💡🏛️ *emissive treatment canonized*), each meaning unique across the layer. This binding is the layer's only one (composition law C2): a knowledge glyph speaks, and the qualifier says what became of the knowledge — distilled into a principle, or canonized into inheritance.

### Tier VI — Ecology faces

The health of the crew itself — the layer that tracks whether the *system of agents* is growing, feeding, and connecting, under N8toriousAi's specialization-by-experience philosophy.

| face | meaning |
|---|---|
| 🌿 | growing |
| 🍎 | feeding |
| 🌉 | cross-pollination |
| 🕸️ | connection formed |
| 🌲 | mature capability |

🌿 is knowledge expanding; 🍎 is new observations arriving; 🌉 is two departments sharing knowledge — the cross-pollination the failure-record law runs on, one lane's lesson becoming another's brief; 🕸️ is a new relationship discovered between things the lab already held; 🌲 is a long-lived proven skill, a capability that has survived enough projects to be *mature*.

### Boundary law — the near-duplicates, separated

The schema law forbids two glyphs for one state; these are the pairs that stand close enough to need their boundary written as law:

- **🅿️ parked is a LANE paused; 😴 dormant is an IDEA parked.** Different subjects, different owners: a lane is parked by order and waits to resume; an idea is dormant by intention and waits to be wanted. A parked lane is not dormant, and a dormant idea was never a lane.
- **The knowledge arc: 🧩 is the discovery moment; 📚 is the pattern filed; 🧠 is the principle distilled.** Three different events in one life: a pattern is *seen* (🧩, Tier IV — the achievement of noticing), then *recorded* into the knowledge base (📚, Tier III — the filing act), and only later *distilled* into a principle that survives its project (🧠, Tier I — the lesson). Collapsing them would lose exactly the stages the lab's compounding-learning law tracks.
- **💎 is unexpected AND reusable; 🤯 is the breakthrough moment.** A rare discovery is a *thing found* that the lab can use again; a breakthrough is the *event* of a wall giving way. A breakthrough may produce a rare discovery; the glyphs are not interchangeable.
- **😀 is one agent's agreement; 🤝 is consensus reached.** Stance versus outcome: 😀 says *I agree*; 🤝 says *we have converged*, and is filed after the convergence exists in typed traffic.
- **⚠️ / 🐛 / 😬** — risk ahead, defect found in the lab, regression against a baseline. (Stated in Tier III; repeated here because the three are the layer's most-swapped glyphs.)
- **🚢 shipped ≠ 🚀 major efficiency gain.** A build can ship slow and a fast build can sit unpublished; the earlier annex's confusable warning stands, with 🚀's schema meaning. **📐 measuring ≠ 🧐 investigating**: an instrument on the bench versus an agent on a question. **📦 staged ≠ 🚪 waiting on gate**: where the *work* is versus where the *decision* is.
- **🗂️ promoted ≠ 🏛️ canonized.** 🗂️ records *placement* — a thing moved into the shared registry, where other lanes can draw on it. 🏛️ records *canonization* — protected and inherited, the lesson made part of Majoruh itself. A thing is filed before it can be canonized, and most filed things never are.
- **🍃 foliage is the MATERIAL; 🌿 growing is the CREW.** Tier V against Tier VI: 🍃 is the plant surface as a principle the lab holds — how foliage reads, doses, and fails; 🌿 is the ecology claim that the crew's knowledge is expanding. A lesson about leaves is 🍃; the lab getting smarter is 🌿. The two never name the same claim.
- **🟤 earth is the SOIL; 🌱 is the SPROUT.** 🟤 is ground itself as a principle — the earth class of the bucket. 🌱 (Tier IV) is an *event*: the first successful attempt, a sprout of a result. Soil is what things grow in; a sprout is a thing that grew. The material is not the milestone.
- **💡 emissive is a light-EMITTING surface; 🔥 is a RESULT.** 💡 names the material class whose nature is to give off light — a principle about emissive surfaces. 🔥 (Tier IV) names an outcome: a major improvement, judged after the fact. A surface can be emissive with no improvement anywhere in sight, and most improvements emit no light at all; the glyphs are never interchangeable.

### Chains — the composition law

Faces compose into **chains**: an ordered sequence read as an arc — a story in the order it happened. A chain expands deterministically — each link to its tabled meaning, in order, joined the same way. A single face is a chain of length 1. The reverse map is exact because meanings are unique across all six tiers and the twenty compounds: `chain → expansion → chain` and `expansion → chain → expansion` are both stable, proven by test over every unit.

**Separators.** Two separators are legal on input: the arrow ` → ` and whitespace — N8toriousAi's worked example below is written in spaces, and both forms, and any mixture, parse to the same chain. Emission is canonical: the wire and record forms always join with ` → `, and round-trip exactness is stated against that canonical form — the same doctrine as message field order (§8-A1): free in defined ways on the wire, fixed in normalized emission. One separator distinction carries meaning: *adjacency is not separation.* A separator (arrow or space) makes two links; no separator makes one link — which parses only where a binding exists (C2), so `🪨 🧠` (two links: stone principle → principle found) and `🪨🧠` (one link: stone lesson discovered) are different chains saying different things.

- **C1 — Any tier may open a chain; any tier may close one.** Tier is metadata about a link, never a syntax position. N8toriousAi's example opens in cognition (🧐) and closes in achievement (🏛️), crossing four tiers on the way; a chain that opens 👑 and closes 🧱 is grammatical and means exactly what it says.
- **C2 — Binding is scoped to knowledge compounds.** A Tier V glyph binds with an immediately following 🧠 or 🏛️ — a property of the knowledge tier's template, not a general composition rule. No other binding exists, and there is no mechanism to mint a new one outside a spec revision. (The earlier annex's one bound pair, 🏆😎, was dissolved by the schema adoption — see the supersession record — leaving the knowledge compounds as the layer's only binding.)
- **C3 — Adjacency without a binding is not a chain.** 🔥👑 written together is one token, it is in no table, and it is rejected. Glyphs do not concatenate.
- **C4 — Repetition means recurrence.** A face may appear more than once (🧱 → 🔨 → 🧱: blocked, worked, blocked again). Each occurrence is an event; the chain is a history, not a set.
- **C5 — One chain, one subject.** By convention (a filing law, not a parse rule): a chain is read against the board line it rides on — one lane, one result, one morning. Braiding two subjects' arcs into one chain is a filing error the record-keeper rejects, the way a receipt about two proposals is.
- **C6 — Length is a smell, not a limit.** The grammar caps nothing; the discipline does. A chain that needs a paragraph of links has stopped being a signal — the story belongs in typed traffic, and the chain shrinks back to its arc.

### Log transparency — the faces live in the log

N8toriousAi's order, the same minute as the adoption: *"Make sure these emojis can be seen inside the log too, I need full transparency."* The faces are part of the filed record, not chat decoration, and this section is the mechanical form of that order:

- **The record rendering carries the raw glyphs beside the expansion.** `FaceChain.to_record()` (also `render_face_record`, and the codec's `face_record`) renders two lines — the glyph chain first, its expansion after — so the human record and the wire form can be eyeballed against each other, glyph for word:

```
🧐 → 🔬 → 🐛 → 🔁 → 📐 → 👀 → 🔥 → 📚 → 🗂️ → 👑 → 🏛️
investigating → experiment active → defect found → retrying → measuring → needs your eye → major improvement → pattern recorded → promoted → new baseline → canonized
```

- **No layer strips a glyph.** Parser, expander, codec, and record rendering preserve glyphs byte-exact — no stripping, no escaping, no normalization away, no ASCII folding. Eight glyphs are tabled with their U+FE0F variation selector (⚠️ 🅿️ ♻️ 🏗️ 🗂️ ⚙️ 🕸️ 🏛️); the bare text-presentation strings are different strings, in no table, and are rejected — and the selector survives every rendering, which the tests assert at the byte level.
- **Folding fails loudly, never silently.** A glyph-bearing rendering has no ASCII form: encoding one to ASCII raises `UnicodeEncodeError`. A glyph-stripped line is not a face record and does not parse. There is no code path on which a face quietly becomes a `?`.
- **The eyeball check is mechanical.** The record's glyph line re-parses to the same chain; the expansion line parses back to it too. What a reader sees in the log and what the parser sees are the same object, proven by test over every unit and over the worked example.

Wherever the implementation produces a fileable record, face content appears in this glyph-first form. (Message records already carry their glyphs the same way: `rea`'s canonical form prints the 😊/☹️ itself, §5.1.)

### Authority law — faces report, they never authorize

Consistent with §6 and with D7 (systems are never sources of authority):

- **🗂️ records a placement N8toriousAi or the Gate already made.** A thing is promoted into the shared registry by a ruling; the face is filed after, as notation. Writing 🗂️ moves nothing.
- **🏛️ records a canonization N8toriousAi already ruled.** Protection and inheritance are his acts — the lesson becomes part of Majoruh when he says it does. The face is the board's record of that moment, and confers nothing: a 🏛️ in a chain leaves `transcript_approved` exactly as false as it was, and the tests assert it.
- **👑 marks the standing N8toriousAi ruled, not a Gate pass.** "New baseline" is a ruling about the standard; the face records it. The result that earned it still needed its `gat+` to be executed and its `vrd` to be believed.
- **😤 is a disagreement on record, not a veto.** It neither blocks nor refuses. The only refusal form in this language remains what it has always been: **absence** (§6) — Gaven's silence, never a glyph.
- **😎 reports a validation that happened in typed traffic.** The success was verified by a verdict or a receipt; the face points at it, and is not itself the evidence.
- **🚢 and 🧊 record N8toriousAi's acts, exactly as 🗂️ and 🏛️ record his rulings.** A build ships when he taps it and freezes when he orders it; the faces are filed after, as notation. No chain of work faces thaws a freeze or launches a build.
- **🚪 reports a wait; it is not an approval in waiting.** The Gate still answers in `gat+`, `gat~`, or silence — and silence after a 🚪 is exactly what silence is everywhere in this language: refusal, computed from absence. The face does not soften it, appeal it, or mark it pending.
- **📱 reports that a verdict is owed by the device.** The verdict itself arrives as device evidence in typed traffic (a log, a receipt); a 📱 that is never answered is an open question on the board, not a pass.
- **📐 and 🔬 report that an instrument ran or an experiment is live.** Measurement and experiment are evidence *for* a verdict; the verdict is still a `vrd`, spoken by Veda or Gaven, with its audit score. Nobody verdicts in faces.

A face chain can summarize a Gate story; it can never *be* one. The moment a face would change what happens next, the language requires the message instead.

### Supersession record — annex v1 (five tiers) → the Face Schema (six tiers)

N8toriousAi's adoption of 2026-11-05 07:16 supersedes the annex's first version (built the same night: his conversation and achievement tables verbatim, deepened by the lane/work/signal tiers under his 07:06 order — 31 faces, 5 tiers). The schema stands at **54 faces and 12 knowledge compounds, 6 tiers**. Every change is recorded here; nothing was dropped silently:

- **Tier structure.** Conversation → **Cognitive** (Tier I); the separate Lane and Work tiers merged → **Work** (Tier II); Signal → **Signals** (Tier III); Achievement stays (Tier IV), re-split as below; **Knowledge** (Tier V) and **Ecology** (Tier VI) are new.
- **👑 changed meaning; 🏆😎 dissolved.** 👑 was *best result so far*; the schema makes 👑 the *new baseline* and gives *best result so far* to a standalone 🏆. The bound pair 🏆😎 (*promoted to baseline*) no longer exists as a face: 😎 is Tier I *verified*, 🏆 stands alone, and the string 🏆😎 is adjacency in no table — rejected. The promotion-record role the pair carried is split between 🗂️ (placement into the shared registry) and 🏛️ (canonization), with the boundary written above.
- **🎯 moved tiers.** Achievement → Signals; meaning *target achieved* → *target reached*, everywhere.
- **Meanings updated by the schema.** 😮 *unexpected discovery* → *discovery*; 😬 *regression detected* → *regression*; 🤝 *consensus reached* → *consensus*; 😎 *validated success* → *verified*; 🚀 *performance gain* → *major efficiency gain*; 👀 *needs nathan's eye* → *needs your eye*. The superseded strings are dead: they expand back to nothing, and the tests assert their rejection.
- **Placed, not dropped.** 😀 *agrees* (Tier I, boundary vs 🤝), 🧩 *new pattern discovered* (Tier IV, the knowledge arc), 🤯 *breakthrough* (Tier IV, boundary vs 💎) — all from N8toriousAi's originally ratified tables, all seated deliberately by the schema adoption.
- **Separators.** Whitespace-joined chains are legal input (new — the form of N8toriousAi's worked example); canonical emission remains the arrow form, and round-trip exactness is stated against it.

### Tier V completed — N8toriousAi's slate (2026-11-05 07:56)

Forty minutes after the schema adoption, N8toriousAi filled the four empty knowledge seats by adopting the prepared slate — **⬜ plaster · 🍃 foliage · 🟤 earth · 💡 emissive** — each collision-checked against every existing and refused glyph before it was put to him. The annex now stands at **58 faces and 20 knowledge compounds, 78 expandable units, 6 tiers**. The adoption carries its conditions with it: the ⬜ caveat (no glyph literally means plaster; the seat is held by his word, recorded at the Tier V table) and the three boundary clauses (🍃/🌿, 🟤/🌱, 💡/🔥, written in the boundary law). No other tier, meaning, or rule changed; the gap note the tier carried is closed by this record, not erased — it stood named until he filled it, which was the point of naming it.

### Considered and refused

Depth includes the negative space. These glyphs were considered for the annex's tiers and refused; the refusal and its reason are part of the record, so the question does not get re-asked fresh each session. (🏆😎 does not appear here: it was not refused — it was superseded, and its record is the supersession record above.)

| glyph | candidate meaning | refused because |
|---|---|---|
| ✅ | approved | It reads as the Gate's word. Approval exists in this language as `gat+` or as absence (§6); a checkmark face would let the fast layer charon the one act the grammar most carefully guards. The most dangerous glyph proposed — refused first. |
| ❌ | refused / rejected | Refusal is absence. A refusal glyph would make silence ambiguous — did the Gate refuse, or merely not write the glyph? — and ambiguity about refusal is the failure the whole grammar exists to prevent. |
| 👍 | approval / endorsement | The same forgery as ✅, plus a groundless `rea` in one glyph: both systems already have their forms, and neither is this. |
| 🛑 | stop / halt | It reads as a command — an authority act. Faces report states; they do not issue orders. A lane is stopped by N8toriousAi or the Gate, in typed traffic. |
| 🔒 | locked / sealed | It reads as a security state — Gaven's domain, not a work state. The real events it might name are already covered: a frozen line is 🧊, a staged pack is 📦. |
| 🧪 | testing | A near-duplicate of 📐 measuring. This lab's bench act is measurement (gates measure, eyes measure); one glyph carries it. |
| 💤 | idle | A near-duplicate of 🅿️ parked. A lane that is not running is either parked by order or simply off the board; a third state names nothing real. |
| ⏳ | waiting (generic) | Too vague to expand honestly. Every real wait in this lab has a named object — the Gate (🚪), the device (📱) — and a generic wait would blur exactly the distinction the tiers exist to keep. |
| 🎉 | celebration | Decoration. It names no lab event or state; the wins already have their glyphs (🔥, 🎯, 👑). Refused under the restraint law. |

### Worked example — the Majoruh story (N8toriousAi's)

N8toriousAi's chain, as he wrote it (spaces), its canonical form, its exact expansion, and his gloss:

```
🧐 🔬 🐛 🔁 📐 👀 🔥 📚 🗂️ 👑 🏛️
```

```
🧐 → 🔬 → 🐛 → 🔁 → 📐 → 👀 → 🔥 → 📚 → 🗂️ → 👑 → 🏛️
```

```
investigating → experiment active → defect found → retrying → measuring → needs your eye → major improvement → pattern recorded → promoted → new baseline → canonized
```

N8toriousAi's gloss: *Investigated. Experimented. Found defect. Retried. Measured. Needs review. Major improvement. Pattern recorded. Promoted. New champion. Canonized.* The gloss is the story as a human tells it; the expansion is the record as the language keeps it — each glyph to its tabled meaning, nothing paraphrased. The story it tells is the lab's whole method in eleven glyphs: an agent investigated a texture issue and put a hypothesis under experiment; the experiment found a defect, in the lab, where defects belong; the work was retried under the fixed instrument and measured — and one frame still needed the verdict only N8toriousAi's eye can give. The result was a major improvement; the pattern was recorded into the knowledge base and the work promoted into the shared registry. It surpassed the previous champion and became the new baseline — and then the rarest step: the lesson was canonized, lifted out of the game that taught it, and made part of Majoruh itself. Every load-bearing fact under that arc — the defect's receipt, the measurement's numbers, the baseline ruling — still lives where it belongs, in the typed record the chain summarizes.

The earlier annex's worked morning survives as the second example — one morning on the Emerald survey lane, nine glyphs:

```
🧐 → 🚪 → 🔨 → 🐛 → 🔁 → 📐 → 👀 → 🔥 → 📦
```

```
investigating → waiting on gate → working → defect found → retrying → measuring → needs your eye → major improvement → staged
```

The lane opened by investigating a warp anomaly in the census. Its drive plan waited at the Gate — and was approved in typed traffic, as approvals are; the chain records the wait, not the word. Working the route, it found a defect (a poisoned anchor, caught in the lab, 🐛 rather than shipped). It retried the drive under the fixed instrument, measured the zone clean on the gate, and flagged one frame for the only verdict that frame could get — N8toriousAi's eye. The re-authored zone was a major improvement over the pass before it, and the pack was staged. Nine glyphs on the board; the whole morning recoverable from the expansion.
