open/portfolio-optimiser
1
0
Fork 0
Code Issues Releases 3 Activity

feat(prepass): the debate is handed the declared cut and the navigator tools are withdrawn [skip-docs]

Browse source 
[skip-docs]: CLI-flagget og README-blokka kommer i steg 6/7; `prepass_payload` er
foreloepig bare naabar for en bibliotek-kaller.

`run_project(prepass_payload=...)` forgrener bundle-armen. UTEN payload er hver linje
uendret -- pekeren, de fire verktoeyene, siteringer over hele den navigerte basen. MED
et payload faar debatten et DEKLARERT KUTT og verktoeyene trekkes (SS 2.2: kontekst
pre-passet holdt tilbake ble holdt tilbake med vilje; en debatt som holder BEGGE er fri
til aa gaa rundt kuttet den nettopp erklaerte).

Nekten PROPAGERER, aldri en stille degradering tilbake til pekeren -- `load_mandate`s
regel. Maalt paa NULL modellkall, ikke paa exit-koden.

`delivered == 0` nektes ved NAVN foer debatten, med nevnerne, spoersmaalet og ref-en
sitert: maalt er den tilstanden bare naabar naar hvert konsept feilet leksikalsk (den
andre tomme saken nekter produsenten selv), altsaa bevis for FRAVAER. Uten den falt
kjoeringen gjennom til `run.py`s siteringsvakt, hvis melding navngir `docs_dir` -- som
er `None` paa denne stien.

`bundle_excerpt_citations` (datasource) siterer de LEVERTE konseptene alene: et stempel
som siterer hele korpuset for et forslag som saa fire, gjenoppfinner den uerklaerte
paastanden sømmen finnes for. Kroppen tas fra den NAVIGERTE `BundleFile`, ikke fra
payloadets `text` -- den leverte teksten er NFC-normalisert med strippet hale, saa en
locator over den ville ikke indeksert fila den navngir. Deler dermed ogsaa
`bundle_citations`' verdict-eksklusjon i stedet for aa gjenta den.

MCP-appenden ligger BEVISST under forgreningen: dette trekker navigatoerverktoeyene,
ikke verktoeylista. Maalt: en tom liste naar traaden som `tools: None`, saa ingen
uproevd tom-array-form innfoeres.

`RunResult.prepass` og `DryRunReport.prepass` DEFAULTER (`skipped_links`-halvdelen:
`None` er det sanne utsagnet "ingen payload ble gitt"), bundet i BEGGE grener saa ingen
`NameError` venter paa veg-stien. `ProvenanceStamp` er BEVISST urørt -- stempelet
beskriver gaten som doemte EN kandidat, dette er et RUN-nivaa-faktum om hva kjoeringen
i det hele tatt fikk lese.

Tilbaketrekkingen asserteres paa `fresh_workflow(tools=...)`, ALDRI paa
`debate_tool_calls`: maalt er det sporet allerede tomt MED alle fire verktoeyene, fordi
en `ScriptedChatClient` aldri emitterer et verktoeykall. En arm skrevet paa det kan
ikke skille de to implementasjonene. GATE, IKKE VEGG: en egen arm beviser at den gatede
ExpeL-folden fortsatt naar hypotese-prompten (0.82) under et payload.

1440 passed / 5 skipped (fra 1425/5, +15, 0 fjernet). ruff + mypy rene. Golden
`shasum -a 1` av INNHOLDET = ea8c534773acdbe41ae68f2c55724d69aaf8be4f, BYTE-UENDRET.

Co-Authored-By: Claude <claude-opus-5>
This commit is contained in:
Kjell Tore Guttormsen 2026-09-07 11:10:54 +02:00
commit 7aa06d581f
3 changed files with 513 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

101
src/portfolio_optimiser/run.py
View file

@ -49,6 +49,7 @@ from portfolio_optimiser.contracts import GoalConfig, GoalContract, load_contrac
from portfolio_optimiser.ledger import SavingsLedger, to_ore
from portfolio_optimiser.datasource import (
bundle_citations,
bundle_excerpt_citations,
chunk_dict_to_citation,
make_retrieval_tool,
retrieve_chunks,
@ -118,7 +119,7 @@ from portfolio_optimiser.validator import (
baseline_from_project,
validate_proposal,
)
from portfolio_optimiser import hitl, okf, outbox
from portfolio_optimiser import hitl, okf, outbox, prepass
from portfolio_optimiser.semretrieval import (
SEMANTIC_WEIGHT_DEFAULT,
Embedder,
@ -234,6 +235,17 @@ class RunResult:
#: Carried HERE and on neither other carrier: ``ProvenanceStamp`` describes the gate that
#: judged ONE candidate, and ``DryRunReport`` returns above generation entirely.
expert_revisions: tuple[ProposalReview, ...] = ()
#: The cut this run was GIVEN, when it was given one (order 20260907T080223Z): the base's ref,
#: the question it was computed for, the three denominators and the withheld rules by count.
#:
#: DEFAULTED, unlike ``cost_baseline_anchored``, and the difference is the one that row states:
#: both of that field's possible defaults would assert something about an event, whereas
#: ``None`` here is the true statement "no payload was supplied" — and there is exactly one way
#: to supply one. This is ``skipped_links``' half of the rule.
#:
#: Carried HERE and not on ``ProvenanceStamp``: the stamp describes the gate that judged ONE
#: candidate, while this is a RUN-level fact about what the run was allowed to read at all.
prepass: prepass.PrepassDeclaration | None = None
@property
def verdict_key(self) -> str:
@ -294,6 +306,11 @@ class DryRunReport:
#: (and both claims would sometimes be false), while a missing trace asserts only that the event
#: list is empty. The road path navigates no bundle, so empty is literally true there too.
skipped_links: tuple[okf.SkippedLink, ...] = ()
#: The cut a dry run was given, when it was given one. Carried here for ``skipped_links``'
#: reason and NOT for ``unkeyed_verdicts``' one: the dry-run cut returns BELOW the fork that
#: resolves a payload, so a dry run can honestly report what it would have read — whereas a
#: field resolved above that cut could only ever report zero.
prepass: prepass.PrepassDeclaration | None = None
@dataclass(frozen=True)
@ -889,6 +906,14 @@ async def run_project(
#: every commissioned approach, and a ``revise`` buys ONE more attempt out of the budget the
#: loop already has. It mints no verdict and gates nothing (F2).
proposal_reviewer: ProposalReviewer | None = None,
#: A verified OKF consumption pre-pass payload (order 20260907T080223Z). Bundle path only, and
#: OPT-IN by construction: ``None`` (the default) leaves the debate on ``_bundle_pointer``'s
#: pointer and the four navigator tools, byte-identically. Given one, the debate is handed a
#: DECLARED CUT and the tools are withdrawn — a debate holding both would be free to walk
#: around the cut it just declared, which is the undeclared cut in a costume (contract SS 2.2).
#: A LOADED object, never a path: the library seam takes the validated artefact and the CLI
#: owns the file, exactly as ``mandate=`` and ``dimension=`` already do.
prepass_payload: prepass.PrepassPayload | None = None,
) -> RunResult | DryRunReport:
"""Run the vertical slice for ONE project. ``client_factory`` is the test-injection seam
(defaults to the real backend). ``verdict_input`` carries the expert decision/rationale
@ -957,6 +982,14 @@ async def run_project(
# * bundle path: anchored only when the bundle SHIPS a ``cost-baseline.json``. A bundle written
# before the amendment (every commons-owned golden) is legitimately un-anchored -> None =
# pre-S4.0 behaviour. A baseline that exists but is malformed still raises (fail-closed).
if prepass_payload is not None and bundle_dir is None:
# Hoisted above the arm below, and refused rather than ignored: a payload is a cut OF a
# knowledge base, and the road path has none for it to agree with. A silently dropped
# payload would leave the caller with a navigating run that reported a declared one.
raise prepass.PrepassRefused(
"a pre-pass payload declares a cut of a knowledge base, so it needs the bundle it "
"was cut from; this run was given no bundle_dir"
)
if bundle_dir is not None:
bundle = okf.navigate_bundle(bundle_dir)
# ONE bundle-id rule (Step 10, slackened S7a-3 pkt. 1): the DECLARED id is the identity and
@ -985,15 +1018,63 @@ async def run_project(
# tools the exploration uses; ``gen_context = debate_output or context`` below means the
# generation fallback is bounded by the same change rather than by a second policy.
dimension_id = dimension.id if dimension else None
context = _bundle_pointer(bundle, resolved.id, dimension=dimension_id)
citations = bundle_citations(bundle)
# The pre-pass fork (order 20260907T080223Z). WITHOUT a payload every line below is what
# it was: the pointer, the four tools, and citations over the whole navigated base.
#
# WITH one, the debate is handed a DECLARED CUT and the tools are withdrawn. The refusal
# PROPAGATES — never a silent degrade back to the pointer, which is ``load_mandate``'s
# rule: a caller who asked for a declared cut and got a navigating run instead was
# answered by a silently downgraded order.
prepass_declaration: prepass.PrepassDeclaration | None = None
debate_tools: list[Any]
if prepass_payload is not None:
prepass.check_payload_shape(prepass_payload)
prepass.verify_against_bundle(
prepass_payload,
bundle_dir=bundle_dir,
resolved_id=resolved,
dimension=dimension_id,
)
if not prepass_payload.excerpts:
# Measured: ``delivered == 0`` is reachable only when every concept failed to
# match (the producer REFUSES the other empty case, where concepts matched and the
# budget admitted none). So this is evidence of ABSENCE for this question at this
# ref, and saying it is better than two silent alternatives: an empty prompt, or
# falling through to the citation guard below, whose message names ``docs_dir`` —
# ``None`` on this path. SS 7.3's own posture: the skill stops and says so.
raise prepass.PrepassRefused(
f"the pre-pass delivered 0 of {prepass_payload.denominators.considered} "
f"concepts for the question {prepass_payload.question!r} at ref "
f"{prepass_payload.bundle.ref}; an empty cut is evidence that this knowledge "
"base does not answer that question, not something to run a debate over"
)
prepass_declaration = prepass.declaration_of(prepass_payload)
context = prepass.render_context(prepass_payload)
# Citations over the DELIVERED concepts alone: a stamp citing the whole corpus for a
# proposal that saw eight documents re-creates the undeclared claim this seam removes.
# Built from the MOUNTED bodies in ``bundle_citations``' own shape, so ``snippet ==
# body[start:end]`` stays exact by construction rather than indexing normalised text.
citations = bundle_excerpt_citations(
bundle, [excerpt.concept_id for excerpt in prepass_payload.excerpts]
)
# Contract SS 2.2: "Context the pre-pass withheld was withheld deliberately." A debate
# holding both the payload and the ladder could walk around the cut it just declared.
# Measured: an empty list reaches the wire as ``tools: None``, so no untested empty-
# array form is introduced. The MCP append BELOW this fork is deliberately untouched —
# this withdraws the navigator tools, not the tool list.
debate_tools = []
else:
context = _bundle_pointer(bundle, resolved.id, dimension=dimension_id)
citations = bundle_citations(bundle)
# §4.1a context-scope, carried over: the agents read ONLY dimension-matched knowledge.
# The filter used to live in the rendering; with navigation it lives in the TOOLS, on
# both rungs (``navigator_tools``' own gate), because that is now where the bytes
# leave. Under a payload the SAME two gates are re-raised by
# ``prepass.verify_against_bundle`` on the mounted documents instead.
debate_tools = list(navigator_tools([bundle_dir], dimension=dimension_id))
# What the navigation could NOT reach, taken from the run's ONE walk. The road path below
# navigates no bundle at all, so its empty tuple is literally true rather than a stand-in.
skipped_links: tuple[okf.SkippedLink, ...] = bundle.skipped
# §4.1a context-scope, carried over: the agents read ONLY dimension-matched knowledge. The
# filter used to live in the rendering; with navigation it lives in the TOOLS, on both
# rungs (``navigator_tools``' own gate), because that is now where the bytes leave.
debate_tools: list[Any] = list(navigator_tools([bundle_dir], dimension=dimension_id))
else:
project = _project_by_id(project_id)
baseline = baseline_from_project(project)
@ -1003,6 +1084,10 @@ async def run_project(
skipped_links = ()
# No knowledge base, so no bundle identity — said by ABSENCE rather than by minting one.
resolved_bundle_id = None
# Bound in BOTH branches for ``resolved_bundle_id``'s reason: the ``DryRunReport`` and the
# ``RunResult`` below read it unconditionally, and a name bound in one arm only is a
# ``NameError`` waiting for the other caller.
prepass_declaration = None
debate_tools = [make_retrieval_tool(docs_dir, top_k=top_k)]
# Trekk B2 (krav 3): configured MCP servers become tools the AGENTS can call during the debate.
@ -1077,6 +1162,7 @@ async def run_project(
cost_baseline_anchored=baseline is not None,
bundle_id_source=resolved_bundle_id,
skipped_links=skipped_links,
prepass=prepass_declaration,
)
# The MCP lifecycle (Trekk B2): entered HERE, after the dry-run cut above, so a dry run never
# opens a connection — its promise to stop before the first call covers egress too. Constructed
@ -1407,6 +1493,7 @@ async def run_project(
unkeyed_verdicts=unkeyed_verdicts,
debate_tool_calls=tuple(debate_tool_calls),
expert_revisions=tuple(expert_reviews),
prepass=prepass_declaration,
)