graph TD
    IN([Task Schema v2.0]) --> GATE{Status Gate
Circuit Breaker}

    GATE -- "backlog / todo" --> BYPASS[CIRCUIT OPEN
Weight 0 - Monitors OFF]
    GATE -- "in_progress" --> ACTIVE[CIRCUIT CLOSED
Weight 1 - All Monitors ON]
    GATE -- "paused / done" --> PARTIAL[PARTIAL STATE
Ghost RUN - Timer FROZEN]

    ACTIVE --> M1[01 Ghost Filter
Forensic Value Audit]
    ACTIVE --> M2[02 Efficiency Monitor
Budget Audit]
    ACTIVE --> M3[03 Heartbeat Monitor
Stagnation Audit]
    ACTIVE --> M4[04 Scarcity Check
Statistical Viability]
    PARTIAL --> M1
    PARTIAL --> M3

    M1 --> G1{Commits = 0
AND Hours > 0?}
    G1 -- "Ghost Flag" --> P1[GHOST PENALTY
-40 pts]
    G1 -- "Ratio > 5.0" --> P1
    G1 -. "Valid Output" .-> CE

    M2 --> G2{Ratio > 2.5?
Time Spent / Est. Hours}
    G2 -- "Zombie Flag" --> P2[ZOMBIE PENALTY
-20 pts]
    G2 -. "Normal Ratio" .-> CE

    M3 --> G3{Index > 3.0?
Activity / Threshold x Weight}
    G3 -- "Abandoned Flag" --> P3[STAGNATION PENALTY
-30 pts]
    G3 -. "Active Pulse" .-> CE

    M4 --> G4{Tasks < 3
OR Commits < 5?}
    G4 -- "Insufficient Data" --> P4[SCARCITY PENALTY
-50 pts]
    G4 -. "Sufficient Data" .-> CE

    BYPASS --> CE{Confidence Engine
GCS = 100 minus Penalties}
    P1 --> CE
    P2 --> CE
    P3 --> CE
    P4 --> CE

    CE --> T_HIGH[GCS 90-100 - HIGH
Standard Summary]
    CE --> T_MED[GCS 50-79 - MEDIUM
Conditional Summary]
    CE --> T_LOW[GCS below 50 - LOW
HALLUCINATION BRAKE]

    T_LOW --> SUPP[Suppress Projections
Completion Date - Costs - Velocity]
    T_LOW --> DISC[Mandatory Disclaimer
Data Confidence: LOW]

    classDef default    fill:#14110f,stroke:#7e7f83,stroke-width:1px,color:#f3f3f4;
    classDef decision   fill:#34312d,stroke:#d9c5b2,stroke-width:2px,color:#d9c5b2;
    classDef monitor    fill:#131e2a,stroke:#4477aa,stroke-width:1px,color:#6699cc;
    classDef penalty    fill:#2a1410,stroke:#ff5f56,stroke-width:1px,color:#ff5f56;
    classDef brakeNode  fill:#3a1010,stroke:#ff5f56,stroke-width:2px,color:#ff5f56;
    classDef warnNode   fill:#2a2010,stroke:#ffbd2e,stroke-width:1px,color:#ffbd2e;
    classDef passNode   fill:#18281e,stroke:#27c93f,stroke-width:1px,color:#27c93f;
    classDef bypassNode fill:#161616,stroke:#7e7f83,stroke-width:1px,stroke-dasharray: 5 4,color:#8b949e;

    class GATE,G1,G2,G3,G4,CE decision;
    class M1,M2,M3,M4 monitor;
    class P1,P2,P3,P4 penalty;
    class T_LOW,SUPP brakeNode;
    class T_MED,DISC warnNode;
    class T_HIGH passNode;
    class BYPASS bypassNode;
    class ACTIVE,PARTIAL,IN default;

AI Summary Logic Workflow

v2.0 Target Architecture

This architecture represents a "Physics-Based" Control Plane to actively manage human data entropy and prevent AI hallucinations caused by performative reporting.

Status Physics Matrix

Status	Weight	Timer	Ghost	Efficiency	Heartbeat
backlog	0	Off	SKIP	SKIP	SKIP
todo	0	Off	SKIP	SKIP	SKIP
in_progress	1	ON	RUN	RUN	RUN
paused	0	Off	RUN	SKIP	FREEZE
done	0	Off	RUN	Finalize	FREEZE

The Circuit Breaker

We manage circuit states via a State Pattern. When a task is in_progress, the circuit closes—temporal weight becomes 1, and auditing clocks engage.

Entropy Monitors (Mathematical Primitives)

Active tasks are subjected to three deterministic audits. Every anomaly subtracts from the base 100% confidence score.

1. The Ghost Filter (Forensic Auditor for Value)

This workflow acts like a forensic accountant auditing for value, rather than just volume. It asks: "Does the type of work delivered justify the time billed?" It divides Logged Effort (Hours) by Semantic Value (Sum of Commit Weights). Instead of counting lines of code, which can be gamed, it weighs the intent of the work. This prevents "Commit Spam" by correctly valuing low-line-count architectural work while flagging situations where high hours are logged against trivial activities.

$$Ghost\ Ratio = \frac{Logged\ Hours}{\sum(Commit\ Weight)}$$

Penalty: -40 pts if Logged Hours > 0 AND Commit Weight == 0.

2. The Efficiency Monitor (Budget Auditor)

This formula acts as a budget auditor, asking: "Is this task burning hours faster than the plan allowed?" It compares the Actual Cost (Time Spent) against the Approved Budget (Estimated Duration) set by the manager. If no budget was set, the system uses a safety fallback based on Task Scope. A high ratio signals Inefficiency (scope creep, roadblocks, or skill gaps)—distinct from fabrication.

$$Efficiency\ Ratio = \frac{Time\ Spent}{Estimated\ Hours}$$

Penalty: -20 pts if Ratio > 2.5.

3. The Heartbeat Monitor (Hospital Heart Monitor)

This workflow functions like a hospital heart monitor; it ignores the "size" of the patient (Scope) and focuses entirely on whether they have a pulse. It calculates how long a task has been silent relative to the team's standard "heartbeat" (e.g., 24 hours), then multiplies that silence by an "Urgency Factor" (Status Weight). If a high-priority task has no activity for several days, the index spikes, identifying tasks that are technically "Active" but practically dead.

$$Index = \left( \frac{Last\ Activity}{24h\ Threshold} \right) \times Status\ Weight$$

Penalty: -30 pts if Index > 3.0.

4. The Scarcity Check (Statistical Viability)

A new variable in v2.0. If Total Active Tasks < 3 OR Total Commits < 5, the dataset is statistically insignificant. You cannot summarise a project that has not started. Deduction: -50 pts.

Confidence Score Logic (The Hallucination Brake)

The system's "Reliability Index" is designed to prevent the AI from sounding authoritative when the underlying data is corrupted. It synthesizes all four penalty signals into a single trust metric (0–100%).

$$GCS = 100 - \bigl((P)_{ghost} + (P)_{stagnation} + (P)_{efficiency} + (P)_{scarcity}\bigr)$$

Three-Tier Output

GCS Range	Tier	Behaviour
90 – 100	HIGH	Standard Summary. Declarative statements permitted.
50 – 79	MEDIUM	Conditional Summary. Hedging verbs required. Forbidden: "Certain", "Definite", "Guaranteed".
Below 50	LOW	Hallucination Brake triggered. Mandatory disclaimer prepended. Projections suppressed.

Suppression Rules (GCS < 50)

When the brake triggers, the system is forbidden from displaying: Estimated Completion Date, Projected Costs, or Velocity Trends. The summary must begin with the exact string:

⚠ Data Confidence: LOW. Signals are conflicting or insufficient. The following summary reflects available logs but cannot be verified against physical output.