Files
cmos/tests/test_note_formatter.py
T
Mark Eaton 28ca3ac928 v2 chunk 1: scaffold note formatter (Path B parallel artifact)
Begin v2 (in-text citation note form) as a parallel artifact to the
v1 bibliography formatter, per the Path B architectural decision:
no shared mutable state, no shared prompt content, no v1 changes.

New artifacts:

- src/cmos/parser.py: add find_notes() / NoteDefinition /
  NotesParseResult as siblings to split_bibliography(). Targets
  pandoc-style markdown footnote definitions [^marker]: text.
  Single-line only; multi-line continuation deferred.

- src/cmos/note_formatter.py: new module mirroring formatter.py.
  17-rule SYSTEM_PROMPT_NOTES for CMOS 18 first-occurrence note
  form. Reuses cmos.runtime_validator.validate() unchanged — its
  structural checks all apply to note form too. Caller injection,
  retry loop, and model selection mirror v1.

- tests/test_parser.py: 7 new tests for find_notes().

- tests/test_note_formatter.py: 11 new tests mirroring v1 test
  discipline (no API calls, fake-caller injection, retry semantics,
  prompt smoke checks).

- exemplars/notes/: new subdirectory with 3 hand-synthesized
  first-occurrence note exemplars (book, journal article, chapter
  in edited book). Uses expected_note as the field name (not v1's
  expected_bibliography). harness/score.py:load_exemplars uses a
  non-recursive glob, so the v1 canary loader does not see these —
  Path B isolation is automatic.

v1 untouched. v1 canary still loads exactly 17 exemplars. Full
test suite: 96 v1 + 18 new v2 = 114 passing.

Smoke-tested all 3 v2 exemplars against real GPT-5 (not fakes),
2 runs each. book_first_yu and journal_first_kwon: 4/4 byte-perfect
on the first try. chapter_first_doyle: 0/2, surfacing two known
prompt gaps for the next iteration:

  1. Publisher abbreviation not expanded ("U of Chicago Press"
     preserved instead of "University of Chicago Press"). v1's
     formatter.py rule 14 is missing from the v2 prompt.

  2. "ed." pluralized to "eds." for multiple editors. CMOS NB
     uses "ed." invariantly regardless of editor count.

Both gaps are addressable prompt edits, not architectural problems —
exactly the kind of finding the dev loop is designed to surface.

Out of scope (deferred to chunk 2 and later): CLI extension,
shortened-form generation, document reassembly, harness scoring
loop integration, v2 linter rules, real-draft testing.
2026-04-11 18:20:50 -04:00

118 lines
4.3 KiB
Python

"""Tests for src/cmos/note_formatter.py — the v2 inner-loop iterable artifact.
Mirrors tests/test_formatter.py in shape and discipline. The note formatter
is iterated independently of the bibliography formatter (Path B); these
tests pin the stable seams (caller injection, retry loop, prompt smoke
checks) and leave LLM output shape testing to the v2 exemplar canary.
No real API calls in this file. Tests use injected fake callers.
"""
from cmos.note_formatter import (
MODEL,
SYSTEM_PROMPT_NOTES,
build_user_message,
format_note_entry,
)
def test_default_model_is_gpt5():
# Same default as v1. Override with OPENAI_MODEL=... in .env.
assert MODEL == "gpt-5"
def test_system_prompt_mentions_cmos_18_note_form():
assert "CMOS" in SYSTEM_PROMPT_NOTES
assert "18" in SYSTEM_PROMPT_NOTES
assert "NOTE" in SYSTEM_PROMPT_NOTES
def test_system_prompt_mentions_no_inversion():
# The single most distinctive difference from bibliography form: notes
# use normal-order author names ("First Last"), not inverted.
lower = SYSTEM_PROMPT_NOTES.lower()
assert "normal" in lower or "not inverted" in lower
def test_system_prompt_mentions_comma_separation():
# The second most distinctive difference: commas between elements,
# not periods.
lower = SYSTEM_PROMPT_NOTES.lower()
assert "comma" in lower
def test_system_prompt_excludes_leading_number():
# The note number prefix ("1. ", "2. ") is supplied externally; the
# formatter must not include it. The prompt must say so.
lower = SYSTEM_PROMPT_NOTES.lower()
assert "leading number" in lower or "do not include" in lower
def test_system_prompt_mentions_no_ibid():
# CMOS 18 deprecates Ibid., same as v1.
assert "Ibid" in SYSTEM_PROMPT_NOTES or "ibid" in SYSTEM_PROMPT_NOTES.lower()
def test_build_user_message_contains_the_messy_input():
msg = build_user_message("yu, charles. interior chinatown. 2020, 45")
assert "yu, charles. interior chinatown. 2020, 45" in msg
def test_note_formatter_uses_injected_caller():
"""Same caller-injection seam as v1. Tests can substitute a fake without
touching the network."""
recorded: dict = {}
def fake_caller(system: str, user: str) -> str:
recorded["system"] = system
recorded["user"] = user
return "Charles Yu, *Interior Chinatown* (Pantheon Books, 2020), 45."
output = format_note_entry(
"yu, charles. interior chinatown. New York: Pantheon Books, 2020, p 45.",
caller=fake_caller,
)
assert output == "Charles Yu, *Interior Chinatown* (Pantheon Books, 2020), 45."
assert recorded["system"] == SYSTEM_PROMPT_NOTES
assert "yu, charles" in recorded["user"]
def test_note_formatter_strips_whitespace_from_caller_output():
def fake_caller(system: str, user: str) -> str:
return " Charles Yu, *Interior Chinatown* (Pantheon Books, 2020), 45. \n"
output = format_note_entry("anything", caller=fake_caller)
assert output == "Charles Yu, *Interior Chinatown* (Pantheon Books, 2020), 45."
def test_note_formatter_retries_on_validator_failure():
"""If the runtime validator rejects the first attempt (e.g., italics
dropped), the formatter retries. Same retry semantics as v1."""
attempts = []
def flaky_caller(system: str, user: str) -> str:
attempts.append(len(attempts) + 1)
if len(attempts) == 1:
# First attempt: italics dropped — validator should fail.
return "Charles Yu, Interior Chinatown (Pantheon Books, 2020), 45."
return "Charles Yu, *Interior Chinatown* (Pantheon Books, 2020), 45."
output = format_note_entry("messy yu", caller=flaky_caller)
assert "*Interior Chinatown*" in output
assert len(attempts) == 2
def test_note_formatter_returns_last_attempt_after_max_retries():
"""If every attempt fails validation, return the last attempt and don't
loop forever. Same as v1 — the failure surfaces via scoring or human
review, not via a runtime exception."""
attempts = []
def always_bad(system: str, user: str) -> str:
attempts.append(1)
return "Charles Yu, Interior Chinatown (Pantheon Books, 2020), 45."
output = format_note_entry("messy", caller=always_bad, max_retries=2)
assert len(attempts) == 3 # 1 initial + 2 retries
assert "*" not in output