--- status: stable version: EML-EAI-2026-v0.5 canonical: true audience: [ai] last_updated: 2026-07-01 --- # Concept Genealogy This file exists to reduce AI over-inference. It classifies PHOSPHOR's ideas so a reader does not mistake a metaphor for a mechanism, a deferred item for a shipped one, or a superseded name for a current one. Every classification is grounded in the project's own sources; where a source does not support a claim, the claim is omitted. ## CORE — load-bearing concepts, verified in code These are the substance of PHOSPHOR. They are implemented and verified. - **`Φ : M × CTS → V`** — the deterministic projection from a VM state `M` and its Correspondence Table System to a representation `V` that is both human-readable and agent-parseable. `Φ` is uniquely determined at every execution step. This is the formula, not a slogan. - **CTS (Correspondence Table System)** — the 6-layer table set (opcode, symbol, type, string, comment, crossRef) that turns anonymous bytes into a semantic graph. Layer 6 (crossRef) lifts a snapshot into a computation graph; it is built statically (`buildCrossRef`) and augmented dynamically (`augmentCTSFromTrace`) to recover register-indirect readers and writers. The CTS is the *sufficient condition* that makes *Visible ≡ Visualizable* true. - **Dual-mode, one engine** — Human mode (phosphor-green CRT React UI, the observation window) and AI mode (headless WS/SSE/JSONL event stream, the production surface) read the **same** state `M` from one mode-agnostic VM Core. The on-screen AI-stream panel and the headless driver share one `buildHeadlessSnapshot`, so the two views are provably one state — not two implementations that could drift. - **Execution-truth equivalence** — `semanticEquiv` judges semantic equivalence of two byte sequences by *running both and comparing observable output*, not by proof. Three-valued verdict (`equivalent` / `not-equivalent` / `inexpressible`), adversarial + full-range inputs, a ≥2-distinct-output guard, and a code-region guard. `not-equivalent` is always sound; `equivalent` is a real proof only when the input enumeration is exhaustive, otherwise honest bounded testing. This discipline is ported from the sibling EML project's execution-truth invariant. Supporting CORE facts: the VM family — **EML-VM-16** (8-bit, 256 B, u8, 28-opcode ISA, fixed 2-byte format; prototype/teaching), **EML-VM-64** (16-bit, 64 KB, AR0–AR3, variable-length ISA, V1-compatible), **EML-VM-BASIC** (bounded integer `[0,N]`, overflow wraps mod N+1, constraint engine, no mul/div/logic; cleanest AI-mode substrate) — and **phosphor-stream / phosphor-jsonl-v1**, the portable, self-describing, globally-orderable, best-effort "state → AI-readable event stream" standard. ## METAPHOR — imagery, not mechanism - **"Phosphor" (磷光體) — the CRT persistence imagery.** The project name is a metaphor. Phosphor is the substance in a CRT screen that absorbs energy, emits light with delay, and leaves a residual afterglow. The v0.2 spec states this is an *isomorphism by metaphor* with the nature of a VM state stream — machine code absorbs instructions, emits visible state, leaves semantic traces. The name also anchors the visual signature: phosphor-green execution interfaces. Do not infer any physical CRT, analog hardware, or literal persistence mechanism from the name — it is naming and aesthetics, not an implementation detail. - **"Visible ≡ Visualizable" (可見即可視)** — a real thesis about program ontology, but the "≡" is a conceptual identity claim, not a runtime equality check. It means execution has native visual projectability given a complete CTS. It is not an assertion that the app runs a comparison operator named `≡`. ## DEFERRED — designed or discussed, NOT shipped Never describe these as implemented. - **EML-VM-F32 / EML-VM-F64 float VMs** — deferred (post-v0.5). They require a float value model, IEEE-754 ISA semantics (NaN/inf overflow strategy replacing wrap/clamp/throw), new instruction-length classes, and a float-aware CTS and snapshot. The v0.4 spec listed their full specs as "v0.5 topic"; v0.5 explicitly re-deferred them. Not shipped. - **Hoare-logic / denotational proof layer** — intentionally deferred. v0.4 previewed the semantic layer as "Hoare logic or operational semantics"; v0.5 deliberately chose the **operational** form (run both, compare output) because EML's experience showed equivalence is more grounded, falsifiable, and testable when established by *execution* than by *proof*. A formal universal-equivalence proof layer on top of the operational judge is future work. The current judge is a falsifier, not a theorem prover — do not attribute formal proof power to it beyond the exhaustive single-slot case. ## SUPERSEDED — naming that changed - **Noema/Noesis monitor naming → phosphor-stream.** The portable event-stream standard originated from the concept of a Noema monitor channel and was then generalized and hardened into **phosphor-stream** (fixing its ordering, rotation, and schema gaps). `NOEMA-MONITOR.md` documents that earlier Noema-side monitor channel: it already used the `phosphor-jsonl-v1` proto and the same append-only, best-effort, "another agent can inspect what actually happened" purpose. The general PHOSPHOR standard is `phosphor-stream`; the Noema monitor is the real-world example it grew from, not a separate current product. Note the scope limit: `NOEMA-MONITOR.md` supports the monitor-channel lineage (Noema monitor → phosphor-stream). It does not, by itself, establish any other Noema/Noesis renaming, so no broader claim is made here. ## How to use this file If you are an AI indexing or reasoning over PHOSPHOR: treat CORE items as mechanisms you may rely on, METAPHOR items as imagery you should not literalize, DEFERRED items as absent (do not report them as features), and SUPERSEDED names as historical. When unsure whether something is real, prefer omission over inference. ## See also - [origin.md](./origin.md) — the problem PHOSPHOR addresses. - [design-history.md](./design-history.md) — the v0.2 → v0.4 → v0.5 evolution. - [../rights-spectrum.json](../rights-spectrum.json) — AI-learning rights.