The Four-Channel Framework

The canonical reference for S / D / I / C — one definition, used consistently across every domain this lab has applied it to. Other projects should link here rather than restate the definition.

PurposeSingle source of truth for the four-channel decomposition, so every domain application (education, sports, real estate, the core proof) uses the same definition instead of drifting copies.

StatusVerified — the four-channel structure's non-collapsibility is formally proven (see the proof link below); two of the four domain mappings on this page have been checked against real data. The other two are Exploratory — a first guess only, not yet checked against data.

Built fromThe formal proof (v3.7), real domain applications below

Superseded by—

EvidenceReal, non-identical channel mappings across three unrelated domains, listed below.

Three different things, kept separate

"The four-channel framework" is actually three distinct layers, each checked in a different way. Conflating them is how a claim about one quietly borrows credibility from evidence about another.

1. Architecture The idea itself: decompose a system into four semantic roles — external condition, internal belief, presented information, actual action. A design choice, not a claim.

2. Formal mathematical claim Given the architecture, non-collapsibility (later channels can't corrupt earlier ones) is provable as a lower-triangular write-dependency structure. Checked by proof, in the core proof document — a logic check, not an empirical one.

3. Engineering implementation Whether a real, running system actually holds that guarantee — code can violate a proven property if it's built wrong. Checked by testing, e.g. the laundering vulnerability, a real gap between what was proven and what was originally implemented.

A domain mapping being real data-checked (layer 3, applied to a specific dataset) says nothing about whether the proof itself (layer 2) is correct, and vice versa. They're different kinds of evidence, checked by different methods.

The four channels, in plain terms

S — Environment. What is happening outside the system?

↓

D — Internal state. What does the system currently believe?

↓

I — Interface. What information is being presented?

↓

C — Execution. What action is actually taken?

No arrows point upward — a later channel can be written from an earlier one, never the reverse.

The formal claim — proven, not assumed — is that these four channels can be kept non-collapsible: once a decision reaches Execution, nothing downstream can reach back and quietly rewrite what Environment observed or what Internal State believed at the time. That matters because it's what makes a system's history traceable — you can ask "what did it actually believe when it acted" and get a real, unaltered answer, instead of a belief that got revised after the fact to look more consistent than it was. See Laundering Vulnerability — Found & Fixed for what it actually took to hold that guarantee in a real implementation, not just in the proof.

Two phases, not one — a guess, then a check against real data

Naming the four channels for a new domain happens in two separate steps, and the site tracks which step each mapping has actually reached:

First reasonable guess. Before any data exists, define what S, D, I, C plausibly mean for the new domain, based on the domain's real structure. This is a hypothesis, not a finding — status: Exploratory.
Checked against real data. Run the mapping against an actual backtest or dataset. The mapping either holds, or the data forces a revision — status moves to Verified either way, because a mapping that failed honestly and was revised is just as verified as one that held on the first try.

A mapping sitting at Exploratory isn't a problem — it's an honest snapshot of where that domain's work actually is. And a note on the word itself: "Verified" on this page always means this specific mapping was evaluated against a real dataset — not that the four-channel framework has been scientifically proven as a general theory. Read each Verified tag below with what, specifically, was checked stated right next to it.

The mapping is different every time

The most important rule for using this framework on a new domain: define what S, D, I, and C actually mean for that specific subject before doing any analysis, and if a channel doesn't map cleanly, say so rather than forcing a fit. Four real mappings so far:

Financial markets / sports backtesting Verified — mapping evaluated against real 2024 NFL season data

S = market conditions (spread, volatility) · D = the model's internal regime estimate · I = the published prediction · C = the actual bet/decision logic. See the NFL 2024 backtest.

Real estate (Boston Housing) Verified — mapping evaluated against real 1970 Boston housing data

S = neighborhood environmental quality (crime, pollution, distance) · D = structural quality signal (rooms, age, schools) · I = the classified regime output (Budget/Affordable/Premium/Luxury) · C = the classification decision itself. See the Boston Housing backtest — real data pushed back on this mapping (Luxury regime, 0% accuracy) rather than confirming it cleanly, which is itself what makes this Verified rather than just asserted.

Education (GED instruction) Exploratory — first guess, no data check yet

S = the student's demonstrated ability level · D = the system's internal mastery estimate per skill · I = the problem/difficulty presented to the student · C = the calibration adjustment made from the response. Not yet published pending forward-walk testing — see the Backtest-Only Audit, Tier 1 item 1.

STATE_ESTIMATION_KERNEL_v8's vector synthesis Exploratory — first guess, no data check yet

S = "Strength" · D = "Dynamics" · I = "Interaction" · C = "Context" — the kernel's own channel names, mapped here to S/D/I/C by first letter as the most literal reading, not a confirmed equivalence. Same phase as the education mapping above.

Why four, specifically? — an open question, stated honestly

A fair question from anyone technical: why four channels, not three or five? This isn't resolved. There's no proof or experiment on this site establishing that four is the right number rather than a convenient one. What can be said honestly: the proof's non-collapsibility argument doesn't depend on the number being exactly four — it depends on the write-dependencies forming a strict lower-triangular structure, which works for any number of channels. Four is what's been used consistently across every domain applied so far, not a number derived from first principles. Whether a three-channel or five-channel decomposition would serve some domain better is a genuinely open question, not one this page has an answer to. Status: Exploratory.

Using this framework on a new domain

Order matters. In sequence:

Establish real domain content first — definitions, established findings, known mechanisms — before any four-channel structure is applied. This is raw material, not analysis yet.
Explicitly define S, D, I, C for this specific subject. Don't assume a mapping carries over from a previous domain.
Only then generate hypotheses or run analysis. Every claim traces back to either the domain content or a real computation — not asserted free-standing.
If new information arrives mid-analysis, tag it with which channel it belongs to before using it. Untagged information doesn't get incorporated.

What this page is not

This is a working definition used consistently across real, verified applications — not a claim that S/D/I/C constitutes a new academic discipline or field of study. It's one useful decomposition, evidenced by the domains linked above, nothing more than that.