Skip to content

CKB-style projection

Projection is the credibility hinge of Myelin. Every CellTx or chunk goes through it, and the resulting CkbProjectionReport is what tells the reader — and a future court verifier — whether the transition is genuinely CKB-projectable.

This page covers what projection does, what it produces, and what makes its output reproducible.

What projection is

Projection is a function from a Myelin CellTx (or chunk) to a CkbProjectionReport:

project_celltx(myelin_celltx) -> CkbProjectionReport
project_chunk(chunk)         -> CkbProjectionReport

The function is deterministic and pure: same input → same output, every time, on every machine. There is no wall-clock, no random, no host state involved.

What the projection report contains

pub struct CkbProjectionReport {
    pub projection_possible: bool,
    pub ckb_style_tx_hash:   Option<[u8; 32]>,
    pub cell_inputs:         Vec<OutPoint>,
    pub cell_outputs:        Vec<CellOutput>,
    pub cell_deps:           Vec<CellDep>,
    pub witnesses:           Vec<Vec<u8>>,
    pub script_groups:       Vec<ScriptGroup>,
    pub unsupported_features:          Vec<String>,
    pub semantic_deviation_flags:     Vec<String>,
}

The most important fields:

  • projection_possible: bool — could this CellTx be encoded as a CKB-style transaction without changing semantics?
  • ckb_style_tx_hash — the deterministic CKB tx hash for the projected bytes, if projection_possible is true.
  • unsupported_features — explicit list of Myelin-only syscalls, metadata fields, or helpers that can't be projected.
  • semantic_deviation_flags — explicit list of places where the Myelin CellTx carries Cell-Model-correct data that CKB can't currently encode.

A successful projection produces a ckb_style_tx_hash that equals the CKB transaction hash for the same bytes. That's why Myelin uses Molecule encoding everywhere — it's the only way to make that equality hold.

How projection works

%%{init: {
  "theme": "base",
  "themeVariables": {
    "primaryColor": "#A5B4FC",
    "primaryTextColor": "#1E293B",
    "primaryBorderColor": "#4F46E5",
    "lineColor": "#6366F1",
    "secondaryColor": "#C7D2FE",
    "tertiaryColor": "#C7D2FE"
  }
}}%%
flowchart TB
    A["Myelin CellTx"]:::in
    B["Encode inputs / outputs<br/>in Molecule"]:::stage
    C["Encode deps / witnesses<br/>in Molecule"]:::stage
    D["Encode script groups<br/>(CKB-style)"]:::stage
    E["Check for Myelin-only<br/>features"]:::check
    F["Check for CKB-unknown<br/>fields"]:::check
    G["Compute<br/>ckb_style_tx_hash"]:::hash
    H["Build<br/>CkbProjectionReport"]:::build

    A --> B & C & D
    B & C & D --> E
    B & C & D --> F
    E -- "any found" --> R["unsupported_features += …"]:::bad
    F -- "any found" --> R2["semantic_deviation_flags += …"]:::bad
    E -- "none" --> G
    F -- "none" --> G
    G --> H
    R --> H
    R2 --> H

    classDef in    fill:#C7D2FE,stroke:#6366F1,color:#1E293B;
    classDef stage fill:#A5B4FC,stroke:#4F46E5,color:#1E293B;
    classDef check fill:#C7D2FE,stroke:#4F46E5,color:#1E293B;
    classDef hash  fill:#C7D2FE,stroke:#7C3AED,color:#1E293B;
    classDef build fill:#A5B4FC,stroke:#7C3AED,color:#1E293B;
    classDef bad   fill:#C7D2FE,stroke:#DC2626,color:#1E293B;

What counts as "Myelin-only"

The projection layer flags a CellTx as myelin-native when it uses:

  • myelin_state_root
  • myelin_session_id
  • myelin_chunk_index
  • Any Myelin-only helper metadata field that has no CKB analogue

When the projection layer flags a CellTx as ckb-inspired-only, it's because the CellTx follows the Cell model correctly but carries a feature CKB can't currently encode. Examples (real ones that have come up during Myelin development):

  • DA manifest segments beyond the Molecule tx bytes (e.g. external DA provider receipts that bind to a separate payload hash).
  • Tendermint precommit signatures in the witness (CKB can carry signatures but doesn't have a Tendermint-style certificate type).
  • Session-scoped context_deps that aren't cell_deps.

Each is recorded explicitly in the report so a reader knows exactly what wouldn't survive projection.

What makes the projection deterministic

Three things:

  1. Molecule everywhere. Every CKB-style field is encoded with the CKB Molecule layout. The same bytes go in, the same bytes come out.
  2. Sorted field ordering. When the projection builds a CKB-style script group or witness list, the order is sorted canonically. No "set ordering" variance across implementations.
  3. No external state. The projection function reads the CellTx and the typed-cell metadata. Nothing else.

The result: any two implementations of the projection layer, given the same input, produce the same ckb_style_tx_hash. The production gate tests this with paired examples.

How the projection report attaches to chunks

For the Teeworlds path, the same projection status attaches to every bounded execution chunk. The CLI commands teeworlds inspect, teeworlds bench, and teeworlds build-fixture all emit a per-chunk projection report.

This means a Teeworlds demo doesn't just say "we ran 3 chunks"; it says "for each of the 3 chunks, here's whether it projects to a CKB-style tx, and here's the hash of the projected bytes."

What the projection report is not

  • Not a CKB submission. projection_possible: true means a CKB verifier could process the projected bytes. It does not mean the bytes have been submitted to CKB. Submission is a separate step (see L1 submission flow).
  • Not a security claim. Projection says "this transition is Cell-shaped and CKB-compatible." It does not say "the transition is valid." Validity comes from the execution report.
  • Not a court verdict. A successful projection is the input shape to a future court; it is not the verdict itself.

The report is auditable

To audit a projection:

  1. Take the input CellTx (with its typed-cell metadata).
  2. Run myelin-exec::projection::project_celltx on it.
  3. Compare the resulting ckb_style_tx_hash, the cell inputs, outputs, deps, witnesses, and the unsupported / deviation lists against the published report.

If anything differs, the projection is broken or the CellTx changed. Either is grounds for rejecting the evidence.

Where to look next