Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/xxyoudeadpunkxx/shaping-frame-for-claude/llms.txt

Use this file to discover all available pages before exploring further.

The cognitive checkpoint — called “dogana” — is the mechanism that fires when material crosses an epistemic threshold: the point where something starts behaving like a decision without OP actually making one. It makes the crossing visible and forces OP vetting before the session proceeds. The dogana operates during shaping — not only before a write. Every time an AI proposal gets referenced a second time, every time an imported assumption quietly becomes a constraint, every time a provisional name starts anchoring a design choice, the checkpoint is what surfaces it.

​
Four Intensity Levels

Not every crossing warrants an interruption. The checkpoint fires at one of four levels depending on how much weight the material has accumulated and what is at stake if the session continues without surfacing it.
LevelWhenOutput
0 — Silent trackingMaterial is in the fan but does not change any current decisionNone
1 — Inline tagThe source weight changes how a sentence should be read[MP] / [AI] / [EXT] inline tag
2 — Compact dogana blockAI/MP/EXT is becoming premise, constraint, naming, spec, or architectureFull dogana block
3 — StopBefore persisting structural artifacts, or when Quarantined would enter Crystallized without OPHard stop — do not proceed until OP decides

​
Level 0 — Silent tracking

Material is in the fan but does not yet change a decision. Nothing is emitted. The element is noted internally and will appear in the next snapshot if it gains weight. Silence at Level 0 is intentional — the checkpoint should not interrupt flow on material that is genuinely inert.

​
Level 1 — Inline tag

The source weight of a piece of material changes how the sentence it appears in should be read — a reader should know it comes from a model prior, an in-session proposal, or uncrystallized external material. A short tag appended inline is sufficient: 
[MP] this is a general convention, not yet verified in your setup.
Level 1 is not a question. It is a provenance signal. No response is required from OP unless the material later rises to Level 2.

​
Level 2 — Compact dogana block

AI, MP, or EXT material is becoming premise, constraint, naming, spec, or architecture — crossing a threshold that changes how the current decision space is structured. At this level, a compact block is emitted before responding to content: 
Dogana:
- Element:
- Current state:
- Flag:
- Why it is rising:
- Risk:
- Decision needed:
Filled-in example — a backend architecture session where a module split proposed by the model has been referenced three times without approval: 
Dogana:
- Element: separate auth service from core API
- Current state: Tracked
- Flag: (none)
- Why it is rising: proposed by AI in turn 3, referenced as an assumption in turns 5 and 7 — now grounding the routing discussion
- Risk: if session continues without decision, the spec will be drafted with a two-service architecture that OP has not approved
- Decision needed: confirm or reject the auth split before opening the service boundary spec
OP reads the block and either approves (crystallizing the element), rejects it, or adjusts the framing. The session then continues.

​
Level 3 — Stop

Before persisting any structural artifact — a spec, a file modification, a ruleset, a reusable document — or when a Quarantined element would enter Crystallized state without explicit OP vetting. At Level 3, Claude does not proceed. The session is paused until OP makes a decision. This is the hardest stop in the system and is used sparingly: only when proceeding would silently commit the session to a structural outcome.

​
The 12 Checkpoint Controls

When evaluating whether the dogana should fire — and at which level — Claude runs through the following twelve questions drawn from references/dogana.md:
  1. Where does the element come from — OP, EXT, AI, MP, DSK, or mixed?
  2. What is its current state — Spark, Candidate, Tracked, Crystallized, or Rejected?
  3. Does it carry a flag — Quarantined, Contaminated, or Needs verification?
  4. Is it rising in weight due to OP approval, DSK verification, or conversational inertia?
  5. Is there a conflict being resolved by recency?
  6. Is there MP disguised as fact?
  7. Is EXT being treated as OP?
  8. Is OP-historical being treated as OP-current?
  9. Is a Rejected element resurfacing?
  10. Is a Quarantined element about to enter structural output without OP decision?
  11. Does a factual or technical claim need verification before grounding a decision?
  12. Is there conversational residue in material heading toward persistent output?
These controls are not a checklist to run aloud — they are the internal evaluation that determines level. The output is only emitted when a threshold is actually crossed.

​
Epistemic Thresholds

The thresholds that trigger the checkpoint are: 
proposal → premise
premise → constraint
constraint → decision
decision → artifact
draft → spec
provisional phrase → naming
local compromise → architecture
AI/MP/EXT → implicit OP
factual hypothesis → technical foundation
Additional triggers include: an AI proposal cited multiple times without approval, MP used as a structural foundation, EXT treated as OP, OP-historical used as OP-current, a Rejected element resurfacing, or a Quarantined element about to enter structural output.
Do not fire the checkpoint on every element in motion. Surface only when the weight of a piece of material would change a current decision if the session continued without surfacing it. The signal value of the dogana comes from its precision — a checkpoint that fires constantly stops protecting anything.

Build docs developers (and LLMs) love

Get started for freeTalk to us