artifacts/incoming
Finance Semantic System — Starter Guide (v0.1)
artifacts/incoming/finance_semantic_system_starter_guide_v_0.mdRendered from markdown source. Open raw source on GitHub.
Finance Semantic System — Starter Guide (v0.1)
Status
Introductory guide for practitioners.
---
1. Why this exists
Finance already looks structured:
- ledgers
- reports
- reconciliations
But the language inside finance is not machine-reliable.
Example:
EBITDA increased by 12%
Questions a machine cannot safely answer:
- Which EBITDA?
- What exclusions?
- Which policy version?
- Which entity?
Humans resolve this implicitly. Machines guess.
That gap is why AI adoption in finance lags.
---
2. Why machine-readable meaning matters
Most AI systems today operate on:
patterns in language
But finance requires:
binding between language and reality
Without machine-readable meaning:
- AI hallucinations look plausible
- auditability breaks
- numbers lose trust
With machine-readable meaning:
- AI becomes deterministic
- outputs are explainable
- audits become traceable
- systems interoperate cleanly
---
3. The core problem
Finance terms are often:
- reused
- overloaded
- context-dependent
Example:
"Revenue" may differ by:
- GAAP vs management
- entity vs consolidated
- time recognition
This creates a hidden failure mode:
same word, different reality
---
4. The solution: explicit semantic grounding
Instead of assuming meaning, we bind meaning explicitly.
Example:
{Revenue@GAAP}
This means:
- this is not just a word
- it is a reference to a defined object
---
5. The three-layer system
The system has three core components:
FSGF — Grounding Framework
Defines what a term means
FSP — Parser
Detects and structures semantic references
SRE — Resolution Engine
Keeps meaning valid at runtime
---
6. Mental model
- FSGF = dictionary + rules
- FSP = reader
- SRE = interpreter + safety system
---
7. Notation (how you write it)
Grounded term
{EBITDA}
Means: resolve this term
With context
{Revenue@GAAP}
With version or policy
{EBITDA@Board#V3}
---
Ambiguous term
[EBITDA]
Meaning unclear or intentionally unbound
Proposed term
<Adjusted EBITDA>
Draft definition
Narrative term
(core profitability)
Not authoritative
---
8. What happens under the hood
Step 1 — Parsing
Input:
{EBITDA@Board}
FSP converts it into:
- term = EBITDA
- context = Board
- status = grounded
---
Step 2 — Resolution
SRE:
- finds canonical definition
- checks policy
- checks freshness
- checks lineage
---
Step 3 — Decision
SRE returns:
- proceed
- warn
- pause
- reject
---
Step 4 — Execution / AI use
AI receives:
- definition
- exclusions
- allowed actions
So it does not guess meaning.
---
9. Example: without system
EBITDA increased 12%
Problems:
- ambiguous
- not auditable
- not reproducible
---
10. Example: with system
{EBITDA@Board#V3} increased 12%
Now:
- definition is known
- exclusions are known
- lineage is traceable
- AI can reason safely
---
11. Real-world usage examples
Board report
{Revenue@GAAP} grew 8% while {EBITDA@Board} declined due to restructuring costs.
Analysis
Compare {FreeCashFlow@Mgmt} to [Free Cash Flow] reported last quarter.
Drafting
<Normalized EBITDA> may exclude one-time litigation costs.
Narrative explanation
(core profitability) improved despite lower revenue growth.
---
12. Why this changes AI adoption
Without grounding:
- AI guesses meaning
- finance rejects it
With grounding:
- AI operates on defined objects
- outputs are auditable
- trust increases
Result:
AI becomes safer than manual interpretation
---
13. Why this feels “expensive”
Because the system must:
- track context
- detect drift
- maintain meaning
- revalidate continuously
But this cost replaces:
- audit failures
- misinterpretation
- semantic drift
---
14. What you actually get
- consistent terminology
- cross-system alignment
- AI-safe workflows
- audit-ready reasoning
- reduced ambiguity
---
15. Minimal adoption path
Start small:
- Define 20–50 key terms
- Use
{}in one report (e.g. board deck) - Map those terms to definitions
- Let AI reference them
---
16. Key rules
- Do not assume meaning
- Use
{}for anything important - Use
[]when unsure - Never upgrade ambiguity silently
---
17. Final takeaway
This system does one thing:
It turns financial language into something machines can understand without guessing.
And that is the missing step that allows AI to safely operate in finance.
---
Anchor note
This guide is intentionally simple. It introduces the system without requiring full technical depth. It should evolve alongside FSGF, FSP, and SRE as the system matures.