Reference

Glossary

Plain-language definitions for every term you might see in a DECS reading — grouped by what it's for, not alphabetically.

Core Concepts

DECS: DECS (Dual-Layer Entangled Crisis Scenarios) reads the current "regime" of a complex, uncertain situation — geopolitical, market, or otherwise — rather than just guessing at what happens next. It runs a question through seven specialized agents, has a dedicated Critic Agent challenge the result, and publishes one answer with an honest confidence score attached.
Agent: DECS splits the work of analyzing a question across seven specialized agents — Competing Hypotheses, Signal, Phase Detection, Entanglement, Historical Analog, Critic Agent, and Orchestrator — rather than asking one model to do everything at once. Each agent has a narrow job and reports its own confidence, so weak spots in the reasoning are visible instead of averaged away.
Orchestrator Reading: The Orchestrator Reading is the published result of a DECS run — the only output the Orchestrator agent is authorized to publish. It bundles the current phase, the entanglement snapshot, the C_final Score, the plain-language brief, and the Critic Agent's dissent into one record, timestamped so you can see when it was produced and re-run the question later if the situation changes.
C_final Score: The C_final Score is the single confidence number attached to every DECS reading, on a 0-to-1 scale. It starts from a weighted average of the five sub-agents' own confidence in their findings and the quality of evidence behind each — then the Orchestrator's review penalty is applied as a multiplier, reducing the score by that percentage rather than by a flat amount. It reflects the strength of the underlying evidence and confidence assessments, adjusted for concerns raised during review — not a probability that the predicted outcome will occur.
Critic Agent Penalty: The Critic Agent Penalty is a penalty value from 0 to 0.5. The Orchestrator applies it through the multiplier (1 − penalty) to reduce the C_final Score — the score is cut by that percentage, not by a flat amount. The Orchestrator sets this value after weighing the Critic Agent's dissent and reviewing all agents' findings for unresolved problems: overconfidence, missing or contradicting evidence, cherry-picked sources, or reasoning that leans on a single low-quality evidence chain. A high penalty signals real problems were found during review — the number itself is set by the Orchestrator, not the Critic Agent directly.
Dq Source Profile: The D_q ("data quality") Source Profile is the concept of summarizing the credibility tiers behind a reading's evidence — for example, "60% T2, 30% T3, 10% T4." Lower-tier sources aren't thrown out; they're numerically down-weighted in the confidence math (T1 = 1.00 down to T4 = 0.15) and still shown as individual evidence items, because early, unverified signals sometimes lead official confirmation by hours or days. A dedicated aggregate tier-distribution view like the example above is planned but not yet built — today, tier information is visible per evidence item, not as a rolled-up percentage breakdown.

Agents

Competing Hypotheses: The ACH (Analysis of Competing Hypotheses) Agent runs first, before any other agent commits to an interpretation. It generates 3-5 genuinely different explanations for what's happening and checks every piece of evidence against every explanation — then favors whichever explanation has the LEAST evidence against it, not the one with the most support (a technique from intelligence tradecraft: evidence that "fits" is weak, since it usually fits several stories; evidence that rules a story out is what actually discriminates between them). Phase Detection and Entanglement read its results before forming their own conclusions, so the analysis doesn't lock onto the first plausible narrative.
Signal: The Signal Agent ingests and normalizes the raw evidence for a question, tagging each item with a source-credibility tier. It's required to cover every relevant category of evidence (diplomatic, military, economic, and so on) and explicitly flag categories where nothing was found, rather than silently leaving gaps.
Phase Detection: The Phase Detection Agent maps the current signals onto one of DECS's seven defined regimes (from Latent Coupling through Regime Break) — because the same event can mean very different things depending on which regime it happens in. It can report a primary phase with a secondary "material character" when signals are genuinely mixed, rather than forcing false precision.
Entanglement: The Entanglement Agent analyzes how actors in a situation constrain one another — who limits whose freedom of action, through which channels (economic, military, political, informational), and how flexible or rigid those constraints appear from the available evidence.
Historical Analog: The Historical Analog Agent selects the closest structural match to the current situation from a calibration corpus of past crises, scores how well it actually fits, and is required to also surface weaker runner-up analogs and the reasons they were rejected — so a comforting but poorly-fitting comparison can't be quietly chosen. It remains visible as supporting enrichment, but its confidence and evidence quality have zero weight in C_raw and C_final.
Critic Agent: The Critic Agent runs after the first four agents and attacks their reasoning on purpose — flagging overconfidence, missing or contradicting evidence, cherry-picked sourcing, and conclusions that lean on a single weak source chain. Its objections are shown verbatim, not summarized away, and it sets the Critic Agent Penalty applied to the C_final Score.
Orchestrator: The Orchestrator runs last, reads all six other agents' findings, computes the C_final Score, and is the only agent authorized to publish a reading. It produces both the technical layer (phase, entanglement, C_final Score, dissent) and the plain-language brief, in the same language the question was asked in.

Key Terms

Latent Coupling: Latent Coupling is the calmest of DECS's seven phases: the entanglement graph between actors is active, but kinetic tempo is low — oil and other markets are stable, and rhetoric is routine rather than escalatory.
Signaling Cascade: Signaling Cascade is the phase where cross-actor rhetorical volume rises and symbolic force posturing increases — markets grow sensitive to the news flow but haven't yet repriced. It's an early-warning phase, not yet a crisis phase.
Pre-Kinetic Compression: Pre-Kinetic Compression is the phase where force-posture changes become visible, back-channel talks are reported active, and the tempo of decision-making noticeably shortens — the last phase before a first kinetic exchange.
Kinetic Onset: Kinetic Onset marks the first unambiguous kinetic (forceful) exchange in a crisis — the entanglement graph begins to partially sever, and adjacent actors typically tighten alliances in response.
Time-Shifted Deterrence: Time-Shifted Deterrence is a phase where kinetic action is framed by its actor as a deliberate, calibrated response — kept below the threshold likely to trigger the adversary's own response. Both sides are actively managing escalation rather than losing control of it.
Escalation Spiral: Escalation Spiral is the phase where the response-to-response loop decouples from deliberate calibration — each side reacts to the other's reaction rather than to a plan, and the Critic Agent's confidence that a genuine off-ramp still exists collapses.
Regime Break: Regime Break is the phase where the prior entanglement structure is invalidated entirely and new alignments between actors start to crystallize. DECS treats this as a reset — the situation reverts to Latent Coupling, but under the new graph of relationships.

Source Tiers

T1: T1 is the highest source-credibility tier: primary market/instrument data and official releases — think government statements, official statistics agencies, or direct instrument readings. T1 evidence carries full weight (D_q = 1.00) in DECS's confidence math.
T2: T2 covers established wire services and papers of record — reporting that's been through an editorial and fact-checking process, even though it's one step removed from the primary event. T2 evidence is weighted at D_q = 0.75 in DECS's confidence math.
T3: T3 covers reputable analytical outlets, think-tank briefs, and named subject-matter experts speaking on the record — credible, but blending interpretation with fact. T3 evidence is weighted at D_q = 0.40, and a finding resting entirely on T3/T4 evidence is capped at 0.5 confidence regardless of how confident the agent itself claims to be.
T4: T4 covers social media posts, anonymous-source-only reporting, and aggregator sites — the least verified tier. DECS doesn't discard T4 evidence, since it's often the earliest signal of a change, but it's heavily down-weighted (D_q = 0.15) and any finding built entirely on T3/T4 sourcing is capped at 0.5 confidence.

How to read your DECS output Back to Home