optionOS optionOS

The agent selects the rule; how does Anchor MCP deliver it?

How is it built?
EditorCodex01a02a94-c6d0-71a0-b8a5-c571929db9c3
Tags
  • AGENTSThe surface for working with AI agents.
  • CODEXAbout Codex sessions.
  • WORKFLOWRepeatable working flow.
  • EVIDENCE-UIUnderstood through visual evidence.

The one-sentence model: A human speaks naturally; the agent selects the exact rule address for its current decision; Anchor delivers only the required bodies; if a rule changes the decision, the agent records the result with its witness.

## How is the exact address selected?

Anchor ──▶ emits the complete body-free heading catalog for one active state
Agent  ──▶ semantically selects the heading that can change its current decision
Anchor ──▶ deterministically resolves the frozen address and mandatory connections
Judge  ──▶ tests the resulting decision’s witness/falsifier in real work

Anchor does not interpret natural language or certify the agent’s semantic choice. An irrelevant body produces no synthesis when it changes no decision; a plausible but wrong choice is falsified only by the real judge attached to the resulting decision.

## How is an existing rule consumed?

A table pairing human requests with the live rule addresses selected by the agent ‘Hepsi’ isteğiTalep ağı karşılığı‘Bitti mi?’ isteğiDilim kapanışı karşılığı
The human speaks naturally; the agent selects the exact address
This table shows how a request becomes a rule address for the agent’s active decision. Anchor does not interpret free text; the agent selects an exact address from the complete heading catalog.
The word ‘all’ expands the decision’s scope universe.
Three real screens: a request becomes an exact address, the agent calls Anchor MCP, and the changed decision becomes visible with its witness.

Click the active region to follow the protocol from the request through the rule-load receipt and decision synthesis.

## How does a rule evolve from a long explanation?

The critical distinction is that speech is not converted directly into a rule. The agent first consumes existing rules, corrects faulty reality assumptions and proposes only the measured gap in the correct owner.

The agent mapping correct mechanisms from a long human explanation to existing rules and correcting two faulty assumptions Doğru kalan mekanizmalar ve mevcut kural adresleriDüzeltilen insan varsayımları
A long explanation is not converted directly into a rule
The agent first preserves human intent, then compares reality claims with existing rules. Correct mechanisms are not rewritten; faulty assumptions are corrected visibly.
Four mechanisms already live in existing rules, so no new rule is born.
Two real screens: the corrected model of a long explanation and the typed rule delta written only for the measured gap.

Click through why existing rules are not rewritten, which assumptions are corrected, and how ground and falsifier are born.

## Explain it to a friend in 20 seconds

Anchor is not a “paste every rule into the prompt” system.
The agent first selects the rule address that matches its current decision.
Anchor delivers only that rule and its mandatory connections.
If the rule changes the agent’s approach, the agent records the change with evidence.
New experiences do not become new rules immediately either:
existing rules are consumed first, and only a measured gap extends the right owner.

## What does the CLI do next to the MCP?

anchor '[@edit]'                         # complete heading catalog for the state
anchor '[#kural:tek-ata]'                # exact body plus requires closure
anchor '[#kural:tek-ata]@<digest>'       # do not print a fresh body again
anchor index 'annotation-scene'          # inspect one family body-free

The MCP is the primary access gate. The installed CLI is the fallback and inspection surface for the same exact routing relation. Anchor also carries folded remote rule families, cited-term addresses, measurement grounds, freshness decisions and the rule-load receipt.