Connectome
Connectome
The connectome is the system's measured wiring map: every node and every characterized edge the system currently believes in — each edge carrying a fitted response, a confidence, and a noise floor. It is the formal name for the map the engine keeps.
The map is not inferred from observation. It is measured by driven pressure: one parameter is pushed, the responses elsewhere are read, and coupling that answers becomes an edge with a curve. The live system is the model — the connectome is its measured part.
Partial by Design
Curves exist only where probed. The connectome is an honest map of what has been measured, not a guessed model of everything — an absent edge means not probed here or below the noise floor, and the map does not pretend to know which.
Two Forms
| Form | Where | Nature |
|---|---|---|
| Live connectome | /connectome |
The living partial map — every node and characterized edge currently believed in |
| Artifact | /connectome/artifact |
The certified export, persisted at last build — complete at build time, the citable object |
The distinction is deliberate: the live map grows with every probe; the artifact freezes a build of it for citation and comparison.
┌───────────────────────────────────────────────────────────────┐
│ Building and Asking │
│ │
│ probe ──▶ measure ──▶ live connectome ──▶ artifact │
│ (push) (read) (grows, partial) (frozen, citable)│
│ │
│ predict / impact / causes ──▶ answers, never touches │
│ the running system │
└───────────────────────────────────────────────────────────────┘
What an Edge Carries
Per edge, the connectome holds response curves: probe levels with measured shifts and honest error bars (/connectome/curves). A curve is the coupling's measured shape — not a coefficient asserted by a model, and not more precise than the noise floor allows.
Two Kinds of Ground
The map holds two measured scopes and never blends them.
Response curves — cross-effects across a parameter's whole range, measured in the default climate (Characterization).
Ambience entries — local ground. The ambience is where the system stands: the resting value of every parameter, everywhere, at one moment. Every measurement is taken in an ambience, and because parameters interfere, the same move can produce different responses in different ambiences — a number without its ambience is a number of unknown validity. An entry is one parameter's own effect on its own reading, measured in a small window around its current resting value (every other parameter frozen), with the ambience as part of the entry's name: points (level, shift, bar) like any curve, an address like no curve had.
┌────────────────────────────────────────────────────────────────┐
│ Two kinds of measured ground │
│ │
│ curve: 0 ──────────── resting ──────────────────────── 100 │
│ (whole range, default climate) │
│ │
│ entry: ┌─ window ─┐ │
│ 0 ──────────────────┤ ▫ ▫ ▫ ▫ ├────────────────────── 100 │
│ └──────────┘ │
│ (around the resting value, │
│ at one named ambience) │
└────────────────────────────────────────────────────────────────┘
The window is an interval because wishes ask interval questions — a band of readings can only be judged reachable over a span of parameter positions. Entries age (the world drifts even when nothing moves) and are re-walked by the engine's own schedule; a wish may commission that earlier, never replace it. The scopes answer different questions and compose with nothing: a curve says what a parameter does in general; an entry says what it does from where the system stands, precisely, with bars.
Asking the Connectome
Reads never touch the system — the map answers; the running system is not disturbed.
| Question | Operation | Answer |
|---|---|---|
| If this node moves by X, what does its direct neighbor do? | predict |
One-hop shift, linearized |
| What happens at the far end of a measured path? | predict/multihop |
Composed cascade shift along measured edges |
| What has one systemtender's probing moved? | impact/{id} |
Measured impact across the map |
| What upstream causes feed one systemtender's patch? | causes/{id} |
Measured upstream causes |
Refusal instead of extrapolation: an ask outside the measured range is refused — the map says so, it does not guess. Composition along measured paths is validated at its boundary, not assumed: 95/97 referee points within propagated 2σ bars, with a composition horizon of ≈2 nonlinear hops at σ=0.02 (Publications).
How It Connects
Interference detection finds candidate coupling; characterization walks measure the curves and the ambience entries; the connectome accumulates all of it. Systemtenders produce it by probing under guardrails — and wishes are served on it: a hold keeps a declared outcome at map coordinates; a commissioned walk searches a region of them.
Surface
The connectome is readable through the REST API's /connectome family (relayed to the causal service), through the connectome_* MCP tools, and in the OpenAPI connectomes group — see API and MCP Interface.
See Also
- Interference Detection — Finds the candidate edges
- Characterization — The walks that measure the curves and entries
- Systemtender — Produces the map by probing
- Guardrails — Every probe is bounded
- Wish — Holds and walks, declared on the map