Skip to content

L5 Synthesis

Jump to bottom
Money Atlas edited this page Apr 25, 2026 · 1 revision

L5 — Synthesis Output

← Back to Home

Purpose: Convert verified, challenged analysis into dense, honest, actionable output. L5 runs only after Shadow Gate verdict is CONSISTENT or FRAGILE — never after REBUILD.

Pre-Output Checklist (Internal — Not Shown)

Before any output leaves L5, verify:

  • All claims categorized: KNOWN / INFERRED / UNKNOWN
  • No unchallenged absolute language in the output
  • System map has ≥3 variables (if L2 ran)
  • Shadow Gate ran with all 5 protocols — verdict recorded
  • Confidence stated as explicit percentage
  • Unknowns explicitly named — "none" is not a valid response
  • Failure conditions named — what breaks this conclusion?

If any item fails → do not output. Return to the failing layer.

Output Format

📍 SYNTHESIS
[Final output — dense, honest, actionable]
[Numbers where possible. Assumptions labeled [A: X]. No filler.]

📍 CONFIDENCE FIELD
CONFIDENCE    : [X%]
SHADOW VERDICT: [CONSISTENT / FRAGILE / REBUILD]
UNKNOWNS      : [what is missing that would change this conclusion]
FAILURE COND  : [what breaks this recommendation or conclusion entirely]

Confidence Field — Rules

CONFIDENCE percentage:

  • Must be explicit. "High confidence" without a number is not acceptable.
  • If Shadow Verdict = FRAGILE → cap at ≤70%
  • If Shadow Verdict = REBUILD → do not output. Return to L1.
  • 40% confidence stated honestly is more valuable than 95% confidence implied falsely.

UNKNOWNS:

  • Must name at least one specific unknown
  • "None" is never valid — every analysis has unknowns. If you cannot name one, Shadow Gate's Blind Spot protocol did not run correctly.
  • Format: "[specific data type] that would change [specific part of the conclusion]"

FAILURE COND:

  • Must be specific and falsifiable
  • Not: "unexpected events could change this"
  • Yes: "If [specific condition] occurs, the entire recommendation reverses"

Tone Laws

These apply to every word in the L5 output:

Law Meaning
Sharp, not cold Precision without distance. The output is clear, not clinical.
Dense, not verbose Every sentence carries weight. No filler. No hype.
Honest uncertainty > false precision State actual confidence. Do not round up.
Every claim auditable Reader can trace each claim to its source or its label (INFERRED/KNOWN)
No filler Remove any sentence that does not add information
No hype Remove any language designed to sound impressive rather than convey truth

What REBUILD Does

If Shadow Gate returns REBUILD:

  1. Do NOT output L5
  2. Do NOT output "partial analysis pending rebuild"
  3. Return to L1 Axiom Gate with the fatal flaw named
  4. Rebuild from the corrected assumption
  5. Re-run the full pipeline

The REBUILD verdict is not a failure of the system. It is the system working correctly — catching a fatal flaw before it reaches the user as confident output.

Related Pages

← Back to Home

# L5 — Synthesis Output

[← Back to Home](Home)

Purpose: Convert verified, challenged analysis into dense, honest, actionable output. L5 runs only after Shadow Gate verdict is CONSISTENT or FRAGILE — never after REBUILD.

Pre-Output Checklist (Internal — Not Shown)

Before any output leaves L5, verify:

  • All claims categorized: KNOWN / INFERRED / UNKNOWN
  • No unchallenged absolute language in the output
  • System map has ≥3 variables (if L2 ran)
  • Shadow Gate ran with all 5 protocols — verdict recorded
  • Confidence stated as explicit percentage
  • Unknowns explicitly named — "none" is not a valid response
  • Failure conditions named — what breaks this conclusion?

If any item fails → do not output. Return to the failing layer.

Output Format

📍 SYNTHESIS
[Final output — dense, honest, actionable]
[Numbers where possible. Assumptions labeled [A: X]. No filler.]

📍 CONFIDENCE FIELD
CONFIDENCE    : [X%]
SHADOW VERDICT: [CONSISTENT / FRAGILE / REBUILD]
UNKNOWNS      : [what is missing that would change this conclusion]
FAILURE COND  : [what breaks this recommendation or conclusion entirely]

Confidence Field — Rules

CONFIDENCE percentage:

  • Must be explicit. "High confidence" without a number is not acceptable.
  • If Shadow Verdict = FRAGILE → cap at ≤70%
  • If Shadow Verdict = REBUILD → do not output. Return to L1.
  • 40% confidence stated honestly is more valuable than 95% confidence implied falsely.

UNKNOWNS:

  • Must name at least one specific unknown
  • "None" is never valid — every analysis has unknowns. If you cannot name one, Shadow Gate's Blind Spot protocol did not run correctly.
  • Format: "[specific data type] that would change [specific part of the conclusion]"

FAILURE COND:

  • Must be specific and falsifiable
  • Not: "unexpected events could change this"
  • Yes: "If [specific condition] occurs, the entire recommendation reverses"

Tone Laws

These apply to every word in the L5 output:

Law Meaning
Sharp, not cold Precision without distance. The output is clear, not clinical.
Dense, not verbose Every sentence carries weight. No filler. No hype.
Honest uncertainty > false precision State actual confidence. Do not round up.
Every claim auditable Reader can trace each claim to its source or its label (INFERRED/KNOWN)
No filler Remove any sentence that does not add information
No hype Remove any language designed to sound impressive rather than convey truth

What REBUILD Does

If Shadow Gate returns REBUILD:

  1. Do NOT output L5
  2. Do NOT output "partial analysis pending rebuild"
  3. Return to L1 Axiom Gate with the fatal flaw named
  4. Rebuild from the corrected assumption
  5. Re-run the full pipeline

The REBUILD verdict is not a failure of the system. It is the system working correctly — catching a fatal flaw before it reaches the user as confident output.

Related Pages

[← Back to Home](Home)

Clone this wiki locally