test: fix fallback chain and stale library snapshot drift in the test suite (#322)

Eight tests failed on main, all from drift rather than a code regression
(verified: the same failures reproduced on a clean HEAD worktree before this
change). Three independent causes:

1. Model fallback chain missed Fable
------------------------------------
langgraph_engine/model_fallback.py encoded
TIER_CHAIN = ("haiku", "sonnet", "opus"), but the authored rule
(~/.claude/rules/model-fallback.md) and the library now declare
haiku -> sonnet -> opus -> fable -> escalate to user (Fable is the tier
above Opus). test_the_encoded_chain_matches_the_rule_document failed.

- TIER_CHAIN gains "fable". next_tier/decide/format_fallback_log are
  chain-driven, so opus now retries at fable and fable is the top tier that
  escalates to the user; no other logic change was needed.
- tests/test_model_fallback.py: the inline copy of the documented chain block
  now mirrors the rule (four lines, fable top tier) and the parsed length
  assertion is 4. The mutated-chain negative controls still fail as intended.

2. kg_selection snapshot constants stale vs claude-global-library
-----------------------------------------------------------------
tests/test_kg_selection.py pins counts measured against library 29.73.0;
the live library is 29.97.4. Re-measured with the same catalogue/adapter:
  EXPECTED_DOMAIN_COUNT          100  -> 104
  EXPECTED_TOTAL_EDGES          7266  -> 7527
  EXPECTED_AGENT_COUNT           508  -> 528
  EXPECTED_SKILL_COUNT           996  -> 1034
  NAIVE_READER_BLIND_SPOT_EDGES  486  -> 490   (domains unchanged at 7)
  MEASURED_CONTAINER_CENSUS[("bare","type")] 58 -> 62 (others unchanged;
  total still equals the domain count)
Docstrings that quoted the old figures/version were updated to match.
tests/test_kg_selection_fallback.py: MEASURED_REAL_MAX_CONFIDENCE
0.864 -> 0.865 (real-task maximum is now 0.8643; minimum 0.685 unchanged,
still well clear of the floor).

3. Advertised file counts one behind the tree
---------------------------------------------
test_doc_counts_match_the_tree: docs said 105 test files / 96 unit /
493 .py repo-wide / 126 files under tests/; the tree has 106 / 97 / 494 / 127.
Updated CLAUDE.md Quick Info and README.md (tests badge, directory tree,
Testing section).

Verification
------------
Full suite, run in four shards (pytest.ini already sets -q, so progress-line
counts were used): 2757 passed, 48 skipped, 0 failed, 0 errors, including
tests/integration and tests/e2e. The four previously failing files alone:
104 passed.

Closes #322

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013u46KPovSoGvJBAxyJiFoX

Author: piyushmakhija28

CLAUDE.md

@@ -20,8 +20,8 @@ Claude Workflow Engine is a 3-level LangGraph-based orchestration pipeline for a
 | **Status** | Active Development |
 | **Primary Location** | langgraph_engine/ |
 | **MCP Servers** | 13 servers -- all in separate repos under [techdeveloper-org](https://github.com/orgs/techdeveloper-org/repositories); 1 also keeps an in-engine copy in `src/mcp/` |
-| **Total Python Files** | 248 (langgraph_engine/); 493 repo-wide |
-| **Test Files** | 105 (96 unit, 5 integration, 3 e2e, 1 load) |
+| **Total Python Files** | 248 (langgraph_engine/); 494 repo-wide |
+| **Test Files** | 106 (97 unit, 5 integration, 3 e2e, 1 load) |
 | **Call Graph** | 578 classes, 3,985 methods, 4 languages (Python/Java/TS/Kotlin) |
 
 ---

README.md

@@ -6,7 +6,7 @@
 [![Python](https://img.shields.io/badge/Python-3.10%2B-green)](https://python.org)
 [![CI](https://github.com/techdeveloper-org/claude-workflow-engine/actions/workflows/ci.yml/badge.svg)](https://github.com/techdeveloper-org/claude-workflow-engine/actions/workflows/ci.yml)
 [![License](https://img.shields.io/badge/License-MIT-yellow)](LICENSE)
-[![Tests](https://img.shields.io/badge/Tests-105%20files%20·%205%20integration-brightgreen)](tests/)
+[![Tests](https://img.shields.io/badge/Tests-106%20files%20·%205%20integration-brightgreen)](tests/)
 [![Discussions](https://img.shields.io/badge/Discussions-GitHub-blueviolet)](../../discussions)
 
 ---
@@ -469,8 +469,8 @@ claude-workflow-engine/           # 369 Python files total
 ├── policies/                     # non-.md policy state only (failure-kb.json) — see note below
 │   └── 03-execution-system/failure-prevention/failure-kb.json
 │
-├── tests/                        # 105 test_*.py files (126 total Python files)
-│   ├── test_*.py                 # 96 unit tests
+├── tests/                        # 106 test_*.py files (127 total Python files)
+│   ├── test_*.py                 # 97 unit tests
 │   ├── integration/              # 5 integration tests (GitHub, MCP, runtime verification)
 │   ├── e2e/                      # 3 end-to-end scenario tests
 │   └── load/                     # 1 concurrency / load test
@@ -628,11 +628,11 @@ ENABLE_HEALTH_SERVER=1 ENABLE_METRICS=1 LOG_FORMAT=json \
 
 ## Testing
 
-105 test_*.py files (126 total Python files in `tests/` including conftest and `__init__` files):
+106 test_*.py files (127 total Python files in `tests/` including conftest and `__init__` files):
 
 | Category | Files | Notes |
 |---|:---:|---|
-| Unit tests | 96 | In `tests/test_*.py` — no external dependencies required |
+| Unit tests | 97 | In `tests/test_*.py` — no external dependencies required |
 | Integration tests | 5 | `tests/integration/` — require live GitHub token or MCP servers |
 | E2E tests | 3 | `tests/e2e/` — require full pipeline environment |
 | Load tests | 1 | `tests/load/` — enabled with `RUN_LOAD_TESTS=1` |

langgraph_engine/model_fallback.py

@@ -71,7 +71,7 @@
 
 _log = logging.getLogger(__name__)
 
-TIER_CHAIN: Tuple[str, ...] = ("haiku", "sonnet", "opus")
+TIER_CHAIN: Tuple[str, ...] = ("haiku", "sonnet", "opus", "fable")
 
 RATE_LIMIT_STATUS_CODES: Tuple[int, ...] = (429, 529)
 

tests/test_kg_selection.py

@@ -7,7 +7,7 @@
 1. No agent or skill name appears as a string literal on the selection code
    path outside test fixtures. Asserted by parsing every module in
    ``langgraph_engine/selection`` and intersecting its string literals and
-   comments against the live 508-agent / 996-skill catalogue. The companion
+   comments against the live 528-agent / 1034-skill catalogue. The companion
    negative plants a real name into a temporary module and requires the same
    detector to flag it.
 2. Ten sample task descriptions each return a ranked agent set in which every
@@ -25,7 +25,8 @@
 its negative substitutes a deliberately non-discriminating scorer and requires
 the same measurement to fail.
 
-Every count asserted here was measured against claude-global-library 29.73.0.
+Every count asserted here was measured against claude-global-library 29.97.4
+(re-measured 2026-09-13; first measured against 29.73.0).
 Where a figure could drift with an upstream library release the assertion says
 so and is written as a bound rather than an equality.
 
@@ -70,15 +71,15 @@
 SAMPLE_STRIDE = 4
 SAMPLE_SIZE = 10
 
-EXPECTED_DOMAIN_COUNT = 100
-EXPECTED_TOTAL_EDGES = 7266
-EXPECTED_AGENT_COUNT = 508
-EXPECTED_SKILL_COUNT = 996
-NAIVE_READER_BLIND_SPOT_EDGES = 486
+EXPECTED_DOMAIN_COUNT = 104
+EXPECTED_TOTAL_EDGES = 7527
+EXPECTED_AGENT_COUNT = 528
+EXPECTED_SKILL_COUNT = 1034
+NAIVE_READER_BLIND_SPOT_EDGES = 490
 NAIVE_READER_BLIND_SPOT_DOMAINS = 7
 
 MEASURED_CONTAINER_CENSUS = {
-    ("bare", "type"): 58,
+    ("bare", "type"): 62,
     ("edges", "type"): 23,
     ("bare", "edge_type"): 7,
     ("relationships", "type"): 7,
@@ -308,7 +309,7 @@ class TestKgAdapterConformance:
     """The adapter reads all six measured schema shapes and never empties out."""
 
     def test_every_domain_parses(self, adapter):
-        """All 100 domain graphs parse; none degrades to an empty result."""
+        """All 104 domain graphs parse; none degrades to an empty result."""
         domains = _all_domains()
         assert len(domains) == EXPECTED_DOMAIN_COUNT
 
@@ -322,7 +323,7 @@ def test_every_domain_parses(self, adapter):
     def test_container_and_edge_type_census_matches_the_measured_shapes(self, adapter):
         """The six container/edge-type combinations occur at their measured counts.
 
-        Counts are for library 29.73.0. An upstream release may legitimately
+        Counts are for library 29.97.4. An upstream release may legitimately
         move a domain between shapes; what must not change is that all six
         forms are handled and every domain lands in one of them.
         """
@@ -355,7 +356,7 @@ def test_edge_totals_agree_with_the_master_catalogue_declaration(self, adapter):
         assert total == EXPECTED_TOTAL_EDGES
 
     def test_the_naive_edges_only_reader_loses_edges_this_adapter_keeps(self, adapter):
-        """NEGATIVE CONTROL: the pre-ADR-015 reader silently drops 486 edges.
+        """NEGATIVE CONTROL: the pre-ADR-015 reader silently drops 490 edges.
 
         If this control ever passes -- if the naive reader loses nothing -- then
         the assertion above is no longer testing anything, because the corpus

tests/test_kg_selection_fallback.py

@@ -71,7 +71,7 @@
 EXPECTED_SPRINT_TITLES = 37
 
 MEASURED_REAL_MIN_CONFIDENCE = 0.685
-MEASURED_REAL_MAX_CONFIDENCE = 0.864
+MEASURED_REAL_MAX_CONFIDENCE = 0.865
 
 FLOOR_ABOVE_EVERY_REAL_TASK = 0.95
 FLOOR_BELOW_EVERY_CANDIDATE = 0.0

tests/test_model_fallback.py

@@ -87,9 +87,10 @@ def _chain_line(*tokens):
     [
         "## Fallback Chain",
         "```",
-        _chain_line("haiku", "sonnet", "opus", "escalate to user"),
-        _chain_line("sonnet", "opus", "escalate to user"),
-        _chain_line("opus", "escalate to user (no further fallback)"),
+        _chain_line("haiku", "sonnet", "opus", "fable", "escalate to user"),
+        _chain_line("sonnet", "opus", "fable", "escalate to user"),
+        _chain_line("opus", "fable", "escalate to user"),
+        _chain_line("fable", "escalate to user (no further fallback, Fable is the top tier)"),
         "```",
         "",
     ]
@@ -220,7 +221,7 @@ def test_the_parser_reads_an_inline_copy_of_the_documented_block(self):
         """
         parsed = parse_documented_chain(DOCUMENTED_CHAIN_BLOCK)
         assert parsed == TIER_CHAIN
-        assert len(parsed) == 3
+        assert len(parsed) == 4
 
     def test_the_parser_rejects_a_mutated_chain(self):
         """NEGATIVE CONTROL: the same parser must disagree when the chain is wrong.