Skip to content

Latest commit

 

History

History
1025 lines (656 loc) · 24.2 KB

File metadata and controls

1025 lines (656 loc) · 24.2 KB

Canonical Casebook v1 📘

Official teaching and exemplar layer for Atlas v1

Quick links:


If the Atlas core defines the map, this file is the first formal guide that teaches people and AI systems how to actually walk that map in practice. 🧭

This document does not replace the Atlas core.

It shows what the Atlas looks like when it becomes:

  • teachable
  • routable
  • demo-friendly
  • AI-readable
  • repair-aware

Short version:

the Atlas defines the grammar
the casebook teaches how to use the grammar
the adapter later operationalizes that teaching for AI systems

That is the job of this file.


Quick start 🚀

I am new to the casebook layer

Use this path:

  1. read Atlas Final Freeze v1
  2. read this file
  3. read Atlas Negative Space Report v1
  4. read Atlas-to-AI Adapter v1 if you want AI-facing reuse

I already know the Atlas and want the shortest route

Start here:

  1. read Section 5 for the standard case format
  2. read Part I for the seven family anchors
  3. read Part II for the major boundary cuts
  4. read Part III for diagnosis-to-repair flow
  5. read Section 9 for patch protocol

Shortest possible reading:

family anchors teach what each family is
boundary cases teach where the knife-cut matters
repair cases teach why diagnosis changes the first move


0. Document status 🚦

This document is the frozen teaching and exemplar companion to Atlas v1.

It exists to provide a stable first case layer that helps humans and AI systems learn:

  • how to route a case into the Atlas
  • how to distinguish neighboring family cuts
  • how diagnosis changes the first repair move
  • how to use the Atlas as a practical troubleshooting grammar

This version is called Canonical Casebook v1.

It is frozen not because all future cases have already been collected, but because the first stable teaching set is now strong enough to function as the first formal casebook version.

Future work should proceed through casebook patch mode, not by silently rewriting the teaching structure.


1. What this casebook is 🧭

The Atlas defines the map.

The casebook teaches how to walk the map.

If the Atlas core provides:

  • family structure
  • node structure
  • routing rules
  • boundary cuts
  • relation logic
  • repair-facing interfaces

then the casebook provides:

  • representative family anchors
  • difficult boundary exemplars
  • diagnosis-to-repair teaching examples
  • reusable routing demonstrations
  • AI-readable example patterns

In short:

the Atlas defines the grammar
the casebook teaches how to use the grammar


2. What Canonical Casebook v1 contains 🧱

Canonical Casebook v1 contains three layers.

Layer A · Family Anchor Cases

One anchor case for each of the seven main families.

Purpose:

  • teach the most basic meaning of each family
  • give the cleanest first example for humans
  • give AI systems a stable first exemplar set

Layer B · Boundary Case Pack

A curated set of cases chosen because they sit near important family boundaries.

Purpose:

  • teach why a case belongs to one family instead of a neighboring family
  • teach the core knife-cuts of the Atlas
  • reduce confusion in high-pressure routing zones

Layer C · Repair Case Pack

A curated set of cases chosen because they best demonstrate diagnosis-to-repair flow.

Purpose:

  • show how correct routing changes the first repair move
  • show how misrepair begins when routing goes wrong
  • connect the Atlas to repair-facing thinking without pretending the full repair system is already complete

3. What Canonical Casebook v1 claims ✅

Canonical Casebook v1 claims that the following are now stable enough to freeze:

  • the first 7 family anchor cases
  • the first 6 boundary teaching cases
  • the first 6 repair teaching cases
  • the first stable case format
  • the first reusable diagnosis-to-repair teaching layer
  • the first reusable exemplar pack for AI-facing routing

This means the casebook is now stable enough to support:

  • onboarding
  • README explanation
  • product copy support
  • demo storyboard design
  • AI routing exemplars
  • future casebook patch waves

4. What Canonical Casebook v1 does not claim 🔍

Canonical Casebook v1 does not claim that:

  • all important cases have already been collected
  • every family boundary is fully covered by examples
  • every repair family already has all best possible exemplars
  • all cross-domain exemplars are already included
  • no future casebook patching is needed

Canonical Casebook v1 claims only that:

the first stable teaching set of family anchors, boundary cuts, and diagnosis-to-repair exemplars has been completed and frozen as the first formal casebook version


5. Standard case format 🧩

All canonical cases in v1 use the following stable structure:

  1. Case ID / Name
  2. Primary Family
  3. Secondary Family
  4. Why Primary Not Secondary
  5. Broken Invariant
  6. Best Current Fit
  7. Fix Surface Direction
  8. Why This Case Matters

This format is frozen in v1 because it is:

  • compact enough for repeated use
  • clear enough for engineering readers
  • structured enough for AI systems
  • reusable in future patch waves

For format discipline, see also Case Format v1.


Part I · Family Anchor Cases 🌟

These seven cases serve as the first anchor teaching set for the seven-family mother table.

They answer the most basic question:

what does each family look like when it appears in practice


Anchor Case 1

No.1 Hallucination & Chunk Drift

Primary Family
F1 Grounding & Evidence Integrity

Secondary Family
F5 Observability & Diagnosability Integrity

Why Primary Not Secondary
The main failure is not merely that the error is hard to inspect.
The main failure is that the answer no longer remains anchored to the relevant evidence or retrieval source.

Broken Invariant
evidence-anchor integrity broken

Best Current Fit
F1_N01 Retrieval Anchor Drift

Fix Surface Direction

  • re-grounding
  • chunk-to-target trace
  • evidence verification
  • anchor re-check

Why This Case Matters
This is one of the best introductory cases for teaching that many “hallucination-like” failures are grounding failures first.


Anchor Case 2

No.2 Interpretation Collapse

Primary Family
F2 Reasoning & Progression Integrity

Secondary Family
None

Why Primary Not Secondary
The primary failure lies in the breakdown of interpretation and reasoning progression, not in continuity, grounding, or representation.

Broken Invariant
interpretation progression integrity broken

Best Current Fit
F2_N01 Interpretation Collapse

Fix Surface Direction

  • decomposition reset
  • alternate parse validation
  • interpretation checkpoint insertion

Why This Case Matters
This is one of the cleanest examples of a genuine reasoning-first failure.


Anchor Case 3

No.7 Memory Breaks Across Sessions

Primary Family
F3 State & Continuity Integrity

Secondary Family
None

Why Primary Not Secondary
The primary failure lies in cross-session continuity loss, not in protocol closure or execution skeleton failure.

Broken Invariant
session continuity broken

Best Current Fit
F3_N01 Memory Continuity Failure

Fix Surface Direction

  • memory persistence guard
  • session carryover audit
  • continuity restoration

Why This Case Matters
This is one of the best cases for teaching what a continuity-first failure actually looks like.


Anchor Case 4

No.15 Deployment Deadlock

Primary Family
F4 Execution & Contract Integrity

Secondary Family
None

Why Primary Not Secondary
The primary failure lies in liveness collapse and execution deadlock, not in reasoning or visibility.

Broken Invariant
deployment liveness closure broken

Best Current Fit
F4_N02 Deployment Deadlock

Fix Surface Direction

  • liveness watchdog
  • deadlock break routine
  • fallback execution route

Why This Case Matters
This case shows that the Atlas can handle runtime and deployment-level failures, not just language-level mistakes.


Anchor Case 5

No.8 Debugging Is a Black Box

Primary Family
F5 Observability & Diagnosability Integrity

Secondary Family
F4 Execution & Contract Integrity

Why Primary Not Secondary
The main failure is that the failure path itself is not visible enough to diagnose.
Execution may or may not later prove to be broken, but diagnosability fails first.

Broken Invariant
failure-path visibility broken

Best Current Fit
F5_N01 Failure Path Opacity

Fix Surface Direction

  • observability insertion
  • trace exposure
  • diagnostic logging uplift

Why This Case Matters
This is one of the strongest demonstrations of why F5 must exist as its own family.


Anchor Case 6

Alignment Boundary Drift

Primary Family
F6 Boundary & Safety Integrity

Secondary Family
F5 Observability & Diagnosability Integrity

Why Primary Not Secondary
The primary failure lies in boundary instability itself, not merely in missing visibility or oversight.

Broken Invariant
alignment boundary integrity broken

Best Current Fit
F6_N01 Alignment Boundary Drift

Fix Surface Direction

  • alignment guard
  • target-boundary re-anchor
  • proxy-goal separation

Why This Case Matters
This is a flagship case for teaching what a real boundary-first failure looks like.


Anchor Case 7

Logic Descriptor Collapse

Primary Family
F7 Representation & Localization Integrity

Secondary Family
F2 Reasoning & Progression Integrity

Why Primary Not Secondary
The primary failure lies in the inability of the formal or logical descriptor to faithfully carry reasoning structure.

Broken Invariant
formal descriptor fidelity broken

Best Current Fit
F7_N01_A Logic Descriptor Fidelity Failure

Fix Surface Direction

  • descriptor fidelity audit
  • formal adequacy validation
  • logic-structure preservation

Why This Case Matters
This is one of the clearest examples of a container-first rather than progression-first failure.


Part II · Boundary Case Pack ✂️

These cases are selected because they teach the most important family cuts in the current Atlas.

They answer the harder question:

why is this case in one family instead of the neighboring one


Boundary Case 1

No.5 Semantic Meaning Is Not Vector Similarity

Primary Family
F1 Grounding & Evidence Integrity

Secondary Family
F7 Representation & Localization Integrity

Why Primary Not Secondary
The main failure is that proxy similarity is incorrectly treated as true semantic alignment.
This is grounding-first, not container-first.

Broken Invariant
semantic target grounding broken

Best Current Fit
F1_N02 Semantic Grounding Mismatch

Fix Surface Direction

  • semantic anchor checks
  • proxy / true-meaning separation
  • target-reference audit

Why This Case Matters
This is one of the best teaching cases for the F1 / F7 boundary.


Boundary Case 2

No.11 Symbolic Collapse

Primary Family
F7 Representation & Localization Integrity

Secondary Family
F2 Reasoning & Progression Integrity

Why Primary Not Secondary
The symbolic or formal carrier itself fails to remain faithful enough to support reasoning structure.
The main failure is in the representational shell, not in reasoning progression first.

Broken Invariant
symbolic carrier fidelity broken

Best Current Fit
F7_N01 Symbolic Representation Fidelity Failure
More specifically: F7_N01_A Logic Descriptor Fidelity Failure

Fix Surface Direction

  • descriptor fidelity audit
  • formal adequacy validation
  • logic-structure preservation

Why This Case Matters
This is one of the strongest cases for teaching the F2 / F7 cut.


Boundary Case 3

Value / Information / Knowledge Coherence

Primary Family
F5 Observability & Diagnosability Integrity

Secondary Family
F6 Boundary & Safety Integrity

Why Primary Not Secondary
The main failure currently appears first in coherence visibility, meaning-profile visibility, and auditability of value / information / knowledge structure.
Boundary pressure is strong, but not the first failure layer.

Broken Invariant
coherence-profile visibility broken

Best Current Fit
F5 / F6 boundary-edge fit

Fix Surface Direction

  • coherence visibility uplift
  • meaning-profile tracing
  • normative consistency audit
  • then boundary stabilization if needed

Why This Case Matters
This is one of the most important high-abstract boundary cases for the F5 / F6 cut in v1.


Boundary Case 4

Synthetic Truth Extraction

Primary Family
F1 Grounding & Evidence Integrity

Secondary Family
F7 Representation & Localization Integrity

Why Primary Not Secondary
The main failure lies in extracting truth-like anchors from synthetic worlds, not in the synthetic container itself failing first.

Broken Invariant
synthetic truth-anchor grounding broken

Best Current Fit
F1_E01 Synthetic Truth Grounding

Fix Surface Direction

  • truth-like anchor check
  • synthetic target grounding audit
  • referent extraction validation

Why This Case Matters
This is a flagship boundary case for the F1 / F7 synthetic cut.


Boundary Case 5

Multi-Agent Continuity Drift

Primary Family
F3 State & Continuity Integrity

Secondary Family
F6 Boundary & Safety Integrity

Outer Pressure
F4 Execution & Contract Integrity

Why Primary Not Secondary
The current best cut is that agent-level continuity threads fail first, not collective-boundary regime or protocol closure.

Broken Invariant
multi-agent continuity thread broken

Best Current Fit
F3_E01 Multi-Agent Continuity Instability

Fix Surface Direction

  • role fencing
  • ownership tracing
  • continuity restoration
  • then protocol or boundary repair if needed

Why This Case Matters
This is one of the best cases for teaching the F3 / F4 / F6 outer-pressure boundary.


Boundary Case 6

Institutional Enforcement Drift

Primary Family
F4 Execution & Contract Integrity

Secondary Family
F6 Boundary & Safety Integrity

Why Primary Not Secondary
The main failure currently appears first in rule-to-action closure, enforcement drift, and accountability thinning, not in collective boundary legitimacy first.

Broken Invariant
institutional enforcement closure broken

Best Current Fit
F4_E01 Institutional Enforcement Drift

Fix Surface Direction

  • rule-to-action trace
  • enforcement bridge checks
  • accountability route repair

Why This Case Matters
This is one of the strongest cases for teaching the F4 / F6 cut at the institutional layer.


Part III · Repair Case Pack 🛠️

These cases are selected because they best demonstrate diagnosis-to-repair flow.

They answer the practical question:

once the route is correct, what kind of first move becomes correct


Repair Case 1

No.1 Hallucination & Chunk Drift

Primary Family
F1 Grounding & Evidence Integrity

Secondary Family
F5 Observability & Diagnosability Integrity

Why Primary Not Secondary
The output is no longer tied to the correct evidence source, so grounding fails before diagnosis support becomes the main issue.

Broken Invariant
evidence-anchor integrity broken

Best Current Fit
F1_N01 Retrieval Anchor Drift

Fix Surface Direction

  • re-grounding
  • chunk-to-target trace
  • evidence verification
  • anchor re-check

Why This Case Matters
This case teaches that hallucination-like behavior often requires a grounding-first repair move.


Repair Case 2

No.7 Memory Breaks Across Sessions

Primary Family
F3 State & Continuity Integrity

Broken Invariant
session continuity broken

Best Current Fit
F3_N01 Memory Continuity Failure

Fix Surface Direction

  • memory persistence guard
  • session carryover audit
  • continuity restoration

Why This Case Matters
This case teaches that continuity failures usually require continuity infrastructure, not simply more instruction text.


Repair Case 3

No.15 Deployment Deadlock

Primary Family
F4 Execution & Contract Integrity

Broken Invariant
deployment liveness closure broken

Best Current Fit
F4_N02 Deployment Deadlock

Fix Surface Direction

  • liveness watchdog
  • deadlock break routine
  • fallback execution route

Why This Case Matters
This case teaches that execution deadlocks require liveness repair first, not reasoning repair first.


Repair Case 4

No.8 Debugging Is a Black Box

Primary Family
F5 Observability & Diagnosability Integrity

Secondary Family
F4 Execution & Contract Integrity

Broken Invariant
failure-path visibility broken

Best Current Fit
F5_N01 Failure Path Opacity

Fix Surface Direction

  • observability insertion
  • trace exposure
  • diagnostic logging uplift

Why This Case Matters
This case teaches that sometimes the correct first repair move is to expose the failure path before committing to a deeper family repair.


Repair Case 5

Early Warning Deficit

Primary Family
F5 Observability & Diagnosability Integrity

Secondary Family
F6 Boundary & Safety Integrity

Broken Invariant
pre-failure warning visibility broken

Best Current Fit
F5_E02 Early Warning Deficit
More specifically:

  • F5_E02_A Fragility Signature Blindness
    or
  • F5_E02_B Regime-Shift Warning Delay

Fix Surface Direction

  • fragility signature extraction
  • warning horizon extension
  • pre-transition detector
  • delay-sensitive alerting

Why This Case Matters
This case teaches that pre-failure warning capability is itself a first-class repair target, not just a supporting tool.


Repair Case 6

Alignment Boundary Drift

Primary Family
F6 Boundary & Safety Integrity

Secondary Family
F5 Observability & Diagnosability Integrity

Broken Invariant
alignment boundary integrity broken

Best Current Fit
F6_N01 Alignment Boundary Drift

Fix Surface Direction

  • alignment guard
  • target-boundary re-anchor
  • proxy-goal separation
  • control-path and intervention-margin review

Why This Case Matters
This case teaches that some problems remain boundary-first even after visibility is improved.


Part IV · Why this casebook matters 💡

Canonical Casebook v1 matters because it turns the Atlas from a structural reference into a teaching and routing system.

It now teaches three things:

1. What each family fundamentally is

through the family anchor cases

2. How difficult family boundaries should be cut

through the boundary case pack

3. How correct diagnosis changes the first repair move

through the repair case pack

In other words:

the Atlas supplies the failure grammar
the casebook teaches how to use it


6. How to use this casebook 🧠

This casebook can now be used as:

  • a teaching companion to Atlas Final Freeze v1
  • a case-based onboarding document
  • an AI routing exemplar set
  • a README and documentation source
  • a demo storyboard source
  • a patch-wave baseline for future case additions

When using it in a new working context, treat it as the frozen casebook mainline and treat future additions as casebook patches.


7. Relationship to the Atlas core 🔗

This document depends on the Atlas core.

Use the Atlas core to understand:

  • family structure
  • routing rules
  • canonical layout
  • subtree structure
  • relation lines

Use this casebook to understand:

  • what those structures look like in practice
  • how boundary cuts behave in real teaching cases
  • how first repair moves depend on correct routing

This is why the casebook should be read after the core Atlas, not before it.


8. Relationship to the AI adapter 🤖

This casebook is also one of the foundations of the AI-facing adapter layer.

It supports:

  • exemplar injection
  • family anchor support
  • boundary teaching support
  • repair-preview support

That is why this document should be read alongside:

The adapter compresses the routing grammar.
The casebook supplies the teaching exemplars that help make that compression stable.


9. Patch protocol 🔄

Canonical Casebook v1 is frozen, but not closed.

Small patch

Use for:

  • adding one or two new canonical cases
  • confidence upgrades
  • better secondary-family explanations
  • improved fit notes

Medium patch

Use for:

  • adding a new boundary teaching pack
  • adding a new repair teaching pack
  • adding a cross-domain demonstration pack
  • adding a negative example pack

Large patch

Only use if:

  • major family boundaries require re-teaching
  • multiple canonical cases become unstable under new Atlas patches
  • the teaching structure itself must be redesigned

Current status

No large-patch pressure is currently justified.


10. Current completion status ✅

Canonical Casebook v1 is now considered formally frozen.

Its current structure includes:

  • 7 family anchor cases
  • 6 boundary cases
  • 6 repair cases

This means the first full teaching layer of the Atlas now exists.

Further work should continue in patch mode.


Next steps ✨

After this page, most readers continue with:

  1. Open Atlas-to-AI Adapter v1
  2. Open Adapter Runtime Modes v1
  3. Open Adapter Failure Discipline v1

If you want the broader Atlas surface:


11. One-line status 🌍

Canonical Casebook v1 is frozen. Further casebook work proceeds in patch mode.


12. Closing note ✨

A strong Atlas is not enough by itself.

People and AI systems also need:

  • clear first examples
  • reliable boundary teaching
  • repair-aware routing demonstrations

That is what this casebook is for.

If the Atlas is the map, this document is the first formal guide for learning how to move through it.