portfolio-optimiser/tests/test_prose_code_form_loadbearing.py
Kjell Tore Guttormsen da0ccd0489 feat(p17b): a context set that spans TWO bases, and a judge told which one [skip-docs]
``contexts/dekke-og-kontrakt-lindaas-2027`` is the first set whose approaches
route at more than one knowledge base: a1/a2 at n200-2024 (material requirements)
and a3/a4 at r761-2025 (the rig, and the falsification arm). That is the whole
reason it exists -- P17b measures that ONE commission can be run across several.

``bundle.txt`` grows a block per base; a set naming one base is one block, so the
four pre-P17b files parse byte-identically. The reader now has ONE home
(``stress.read_bundle_declarations``): it used to be a private copy in the P14
gate and a second, looser one inside ``stress.main``, and the multi-base form is
exactly the change that would have let them drift.

Rule U becomes the UNION of every declared base, and that is not a formality.
MEASURED 15.09: ``enhetspris`` is absent from n200-2024 and carried by 70 of
r761-2025's 2 756 concepts, so anchors admitted per base would have admitted a
question the pass as a whole CAN ground. It was dropped from the fifth set's
anchors for that reason.

``score_context_set(bundle_id=...)`` restricts the judgement to the approaches
routed at THIS base. Without it, judging the n200 outbox reports the r761
approach as ``not_evaluated``/``absent`` -- a false finding, because that
approach WAS evaluated, against the other base, under the other run_id. That
defect is pinned by its own arm. The judge's CLI refuses to guess when a set
declares several bases, with an rc-0 control on ``--bundle``.

Arm (d) gained a second half: every DECLARED base must be named by some
approach, because a base no approach names is never run.

The P19/B2 fasit denominator moved 26 -> 32 and is asserted, not dropped: six new
references, two of them bare ``prosessnr`` (12.11, 12.12), so B1's
punctuation-and-digits form is now exercised by a fasit and not only by a
known-positive.

Suite 1774/5, golden byte-unchanged.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-15 04:44:37 +02:00

211 lines
9.7 KiB
Python

"""P19 DEL B — a "cost code" must have a FORM when the input offers forms.
**The measured defect.** P18's round 2 ended with two ``validated`` proposals whose
``affected_item`` codes were ordinary words out of a road standard's prose:
``impulsventilator`` (in 4 of 270 N500 documents) and ``bituminøst bærelag`` (4 of 1 133 N200).
Both are GROUNDED in P7's sense — they appear verbatim in the input, which is all that stage asks —
and neither is INERT in P18/B1's sense, because neither is anywhere near the 5 % document share.
They are simply not identifiers of a cost line. The gate had no stage that could say so.
**Replayed offline against the very bases those runs were given** (the known positive, measured
15.09 rather than asserted): both come back ``Rejection`` naming the denominator —
``…offers 391 identifiers of its own`` / ``…offers 1359``.
**The generality guard is the half that makes this a rule rather than a preference.** The gate
fires ONLY where the input demonstrably offers identifier forms. A corpus that carries none cannot
be answered in a form it does not use, and refusing there would be a rule about SHAPES rather than
about grounding — exactly what ``_ground_against_input``'s own docstring refuses for the stage it
sits inside.
**B1 — the third and fourth forms, transcribed from a measurement.** R761's requirement numbers are
bare dotted numbers (``12.1``, ``52.11``); ALL SIX ``ref`` values in
``contexts/kontrakt-sorasen-2027/fasit.json`` are of that shape and NEITHER pre-P19 form matched
one, so r761's whole offer was **3 identifiers over 6.5 MB**. Measured after: **2 332**. And the
first form was WIDENED in the same pass, because P19/B2 made these forms decide ``prose`` as well
as count an offer: this repo's own ``ENERGI-TOTAL-EL`` matched neither, so the classifier called a
real cost code prose and the new gate refused it. A gate may only be wrong in the direction that
admits too much.
**Two exemptions, both load-bearing:**
* a code the BASELINE carries is never refused here — stage 0 has already ruled it a real line of
this project, and the weaker stage must not overrule the stronger falsifier (the sentence
``_grounding_text`` already carries about its third source). A derived schedule whose codes are
bare section numbers would otherwise be refused wholesale by the gate meant to protect it;
* ``has_identifier_form`` FULL-matches. ``impulsventilator 12.1`` carrying a process number does
not make the word a cost code, and a substring rule would let any prose code smuggle one along.
What each arm pins:
(a) the two MEASURED defects are refused, with the denominator in the reason (Step 5 feeds that
reason verbatim into the next attempt, and "ungrounded" alone teaches nothing);
(b) the generality guard: a base offering no identifier form leaves the gate OFF;
(c) an anchored code is exempt — the stronger falsifier wins;
(d) the 26 known negatives: every ``ref`` in all four fasit files still classifies as an identifier,
and so does this repo's own ``ENERGI-TOTAL-EL``;
(e) B1's known positives and negatives, one by one;
(f) B2: the classification is reported, and it is the SAME classifier the gate uses.
"""
from __future__ import annotations
import glob
import json
import pytest
from portfolio_optimiser.ir import CostBaseline, CostBaselineLine, SavingsProposal
from portfolio_optimiser.validator import (
Grounding,
Rejection,
ValidatedProposal,
classify_codes,
has_identifier_form,
identifier_tokens,
validate_proposal,
)
#: A base that OFFERS identifier forms: eleven documents carrying requirement numbers, one of which
#: also mentions the prose word. Eleven because ``_GROUNDING_MIN_INERT_DOCUMENTS`` is 10 — with
#: fewer, P18/B1's share rule cannot fire at all and this file would be measuring that instead.
_OFFERING = tuple(f"Krav 10.4.3—{i} om ventilasjon i tunnelen." for i in range(1, 12)) + (
"Krav 8.4.2—1 sier at impulsventilator skal dimensjoneres for brannlast.",
)
#: The same corpus with every identifier removed — prose only.
_FORMLESS = ("ingen koder her, bare tekst om ventilasjon", "og enda mer prosa om tunnelen")
def _proposal(code: str, *, quantity: float = 4.0, unit_cost: float = 250_000.0) -> SavingsProposal:
return SavingsProposal(
project_id="t",
measure="m",
affected_items=[{"code": code, "quantity": quantity, "unit_cost": unit_cost}],
claimed_saving_nok=100_000.0,
)
# ---------------------------------------------------------------------------------------------
# (a) the measured defect
# ---------------------------------------------------------------------------------------------
def test_a_word_from_the_prose_is_not_a_cost_code() -> None:
"""(a) P18's ``impulsventilator``, in the shape that reached ``validated``."""
verdict = validate_proposal(_proposal("impulsventilator"), grounding=Grounding(_OFFERING))
assert isinstance(verdict, Rejection)
assert "has no identifier form" in verdict.reason
# The DENOMINATOR, not just a complaint: Step 5 feeds this reason into the next attempt.
assert "offers 14 identifiers of its own" in verdict.reason
def test_the_same_input_still_admits_a_real_reference() -> None:
"""(a), the other half. A gate that can only refuse proves nothing."""
assert isinstance(
validate_proposal(_proposal("Krav 8.4.2—1"), grounding=Grounding(_OFFERING)),
ValidatedProposal,
)
# ---------------------------------------------------------------------------------------------
# (b)-(c) the two exemptions
# ---------------------------------------------------------------------------------------------
def test_a_base_that_offers_no_form_leaves_the_gate_off() -> None:
"""(b) The generality guard. A corpus with no identifiers cannot be answered in one."""
assert isinstance(
validate_proposal(_proposal("ventilasjon"), grounding=Grounding(_FORMLESS)),
ValidatedProposal,
)
def test_a_code_the_baseline_carries_is_never_refused_for_its_shape() -> None:
"""(c) Stage 0 has already ruled it a real line; the weaker stage must not overrule it."""
baseline = CostBaseline(
project_id="t",
items={"impulsventilator": CostBaselineLine(quantity=4.0, unit_cost=250_000.0)},
)
assert isinstance(
validate_proposal(
_proposal("impulsventilator"), baseline=baseline, grounding=Grounding(_OFFERING)
),
ValidatedProposal,
)
# ---------------------------------------------------------------------------------------------
# (d)-(e) known negatives and positives
# ---------------------------------------------------------------------------------------------
def test_every_fasit_reference_in_every_context_set_is_an_identifier() -> None:
"""(d) Every fasit reference across every context set, with the denominator.
One of them — ``Krav 3.3.2—1_1`` — is why the second form grew an optional ``_<n>`` suffix.
Measured, not anticipated: before that it was the single reference the classifier called prose.
**The denominator MOVED 26 -> 32 with P17b's fifth context set**, and it is asserted rather
than dropped for the reason it was written down in the first place: a list comprehension over
``contexts/*/fasit.json`` that quietly found fewer rows would make this arm weaker without
making it red. The six new ones are four ``Krav x.y.z—n`` from n200-2024 and TWO bare
``prosessnr`` from r761-2025 (``12.11``, ``12.12``) — the punctuation-and-digits form B1 added,
now exercised by a fasit and not only by a known-positive.
"""
refs = [
concept["ref"]
for path in sorted(glob.glob("contexts/*/fasit.json"))
for row in json.loads(open(path, encoding="utf-8").read())["must_cite"]
for concept in row["concepts"]
]
assert len(refs) == 32, f"denominator moved: {len(refs)}"
assert [r for r in refs if not has_identifier_form(r)] == []
assert has_identifier_form("ENERGI-TOTAL-EL"), "this repo's own reference cost code"
@pytest.mark.parametrize(
"token",
["12.1", "12.12", "22.1", "52.1", "52.11", "51.1", "65 ASFALTDEKKER", "SHA-01", "B-20-00-00"],
)
def test_b1_known_positives(token: str) -> None:
"""(e) The forms the delivered corpora carry."""
assert has_identifier_form(token)
@pytest.mark.parametrize(
"token",
[
"2027",
"250000",
"15.09.2026",
"2026-09-15",
"impulsventilator",
"bituminøst bærelag",
"0.70",
],
)
def test_b1_known_negatives(token: str) -> None:
"""(e) Bare numbers and dates are not identifiers — the K2-measured inert class."""
assert not identifier_tokens(token), f"{token!r} was counted as an identifier"
# ---------------------------------------------------------------------------------------------
# (f) B2 — the report
# ---------------------------------------------------------------------------------------------
def test_the_classification_is_reported_and_is_the_gates_own() -> None:
"""(f) One classifier, two consumers (kø-(p)): a report that disagreed with the gate about one
proposal would be evidence about nothing."""
forms = classify_codes(["impulsventilator", "Krav 8.4.2—1", "12.1", "ENERGI-TOTAL-EL"])
assert forms == {
"impulsventilator": "prose",
"Krav 8.4.2—1": "identifier",
"12.1": "identifier",
"ENERGI-TOTAL-EL": "identifier",
}
# And the gate agrees, code for code, on the SAME input.
for code, kind in forms.items():
verdict = validate_proposal(_proposal(code), grounding=Grounding(_OFFERING + (code,)))
refused = isinstance(verdict, Rejection) and "has no identifier form" in verdict.reason
assert refused == (kind == "prose"), (code, kind, verdict)