Kjell Tore Guttormsen
38df79126f
B-gate row 1's premise, made callable. Every path through the framework CLI constructs a chat
client, so an outside caller -- a human at a terminal, or an agent that is NOT po -- could not
reach a single run-path step without paying for a model. These four steps need no model at all.
One CLI, four subcommands, one core call each:
navigate-bundle --bundle-dir -> okf.navigate_bundle
cost-baseline --bundle-dir --project-id -> okf.derive_cost_baseline
retrieve-chunks --query --docs-dir [--top-k] -> datasource.retrieve_chunks
prepass-admit --payload --bundle-dir [--dimension] -> prepass.admit_payload
Each handler is a thin adapter: strings in, the SAME function the run path calls, JSON on stdout,
and an exit code that says what happened (0 ran, 2 malformed call, 3 the step refused, named).
A handler that computed anything of its own would be a second implementation of a run-path step,
and the outside caller would stop getting what the debate gets.
Dispatch is an explicit branch per command, not argparse's `set_defaults(handler=...)`: the table
hides the one thing a reader wants to see, and B-gate row 1 asks the same question of the source
(it walks the call graph from `main` down to the step's symbol), where a callable in a Namespace
is a hop neither can follow.
Probes (`tests/test_toolbox_doors.py`, 10 arms): each starts the door as a SUBPROCESS with the
subcommand in argv and asserts on what it wrote -- never by importing the core function, which is
the whole difference the gate exists to measure. The yardstick is outside the door in every arm:
the filesystem (navigate-bundle, including the one deliberate outside-bundle link), a table
transcribed from the priced fixture (cost-baseline), the in-process seam it must equal byte for
byte (retrieve-chunks), and the producer's own checked-in payload (prepass-admit). Every refusal
arm has an rc-0 control beside it.
`portfolio-optimiser-toolbox` is the THIRD console script, and the pin test now says why: it is
the door the other two cannot be used for. README and CLAUDE.md updated with the command and the
reason it exists; every documented invocation was run.
Row 1: 1 -> 5 of 17 (four subcommands + `gate`, which the class fix in e47be68 stopped rejecting
on a name technicality). No other row moved; exit 1.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
135 lines
9.6 KiB
TOML
135 lines
9.6 KiB
TOML
[project]
|
||
name = "portfolio-optimiser"
|
||
version = "1.1.0"
|
||
description = "Generic framework on Microsoft Agent Framework for per-project cost-savings optimization"
|
||
readme = "README.md"
|
||
requires-python = ">=3.10"
|
||
dependencies = [
|
||
# MAF — GA-pakker, IKKE meta-pakken `agent-framework`. Metaen drar `agent-framework-core[all]`,
|
||
# og `[all]` trekker inn de fortsatt-beta integrasjonene (azure-ai-search/cosmos/ollama/…) som
|
||
# tvinger pre-releases og drar med en ALPHA pydantic. Offisiell guide: installer kun det du trenger.
|
||
# Beta-integrasjoner legges til per-fase (med snevert pre-release-scope) når de faktisk trengs.
|
||
"agent-framework-core>=1.18.0,<2", # kjerne (GA) — to-sidig pin (S2.5): major-bump krever re-verifisering av privat-API-premissene (test_maf_version_guard). F16 (12.09): løftet fra 1.16.0; 17 private/ugaranterte former, 16 sjekket mekanisk av proben, se docs/2026-09-12-f16-maf-1180.md. F15 (02.09) løftet 1.9.0→1.16.0.
|
||
# F16 (12.09): løftet 1.8.2 → 1.13.0 SAMMEN med `-openai` og `core`, og koblingen er MÅLT, ikke antatt.
|
||
# foundry 1.13.0 krever selv `core<2,>=1.17.0` OG `openai<2,>=1.14.2`, så de tre kan ikke løftes hver for
|
||
# seg. Resolveren ville BEHOLDT 1.8.2 (den krever bare `core<2,>=1.9.0`, målt) — men to av 1.17/1.18s
|
||
# BREAKING-endringer navngir core OG foundry i SAMME punkt (#7918 sequence-only middleware-inputer,
|
||
# #8127 `SecretString`), så et 1.9.0-æra foundry mot en 1.18.0 core er halvparten av en koordinert
|
||
# endring — nøyaktig «importerer fortsatt, semantikken har flyttet i stillhet»-klassen F15 § 4 målte.
|
||
"agent-framework-foundry>=1.13.0", # Azure/Foundry-profil: FoundryChatClient (GA)
|
||
"azure-identity>=1.25", # S4.1: AzureFoundryBackend passes an explicit AzureCliCredential (Foundry requires it); already transitive via foundry — promoted to a declared direct dep (zero new install weight)
|
||
"agent-framework-openai>=1.14.3", # OpenAI + OpenAI-kompatible lokale endpoints (GA) → lokal profil; F16: løftet 1.8.2 → 1.14.3 (foundry 1.13.0 krever `openai>=1.14.2`; se raden over)
|
||
# Promotert dev→core i Fase 2 (MVP-runtime, ikke lenger spike-only):
|
||
# F16 (12.09): dette gulvet STÅR på 1.1.1, og det er en MÅLING, ikke en utelatelse: 1.1.1 (2026-08-21) er
|
||
# fortsatt siste utgivelse på PyPI, så det finnes ingenting å løfte det til. Kravet `core<2,>=1.15.0` er
|
||
# tilfredsstilt av core 1.18.0. F15s formulering «de to gulvene løftes SAMMEN — det finnes ingen delvis
|
||
# bump» var sann for F15s bump; etter F16 er den MÅLT SVAKERE: core kan løftes uten at orchestrations kan.
|
||
"agent-framework-orchestrations>=1.1.1", # GA orchestration builders (GroupChat/Concurrent/Magentic)
|
||
"pulp>=2.8", # deterministisk validator-solver; PuLP bundler CBC i wheelen (R2). Installert 3.3.2.
|
||
# PuLP 4.0 vil kreve `pip install pulp[cbc]` + COIN_CMD (Fase-migrasjonsnotat).
|
||
"mcp>=1.28.0", # tynn lokal-mappe MCP-server (Step 7) — GA (resolverte 1.28.0) per Step 1-beslutning
|
||
"anyio>=4.14", # kø-(z): ingest_mcp.py bruker anyio.fail_after direkte (MCP-timeout-stien); allerede transitiv via mcp — promotert til deklarert direkte dep (zero new install weight, resolverte 4.14.0)
|
||
"pydantic>=2.11,<3", # IR/validering (B1) — eksplisitt pin til STABIL 2.x, aldri alpha
|
||
# S3.1: brute-force cosine for the hybrid verdict retriever — MAF-free, offline (D-C).
|
||
# Upper bound is <2.3, NOT <3, and it is load-bearing twice over: numpy 2.3+ requires
|
||
# Python >=3.11 (2.4+ requires >=3.12), so a wider range silently contradicts this project's
|
||
# `requires-python = ">=3.10"`; and numpy 2.3+ stubs use PEP 695 `type` statements, which
|
||
# mypy refuses to parse under `python_version = 3.10` (python/mypy#18701) — breaking
|
||
# `uv run mypy src` outright. Raise both floors together or not at all.
|
||
"numpy>=2.0,<2.3",
|
||
# Door A ingest (§4–§6) — the shared implementation of shared/ingest-spec.md, MAF-free (D7).
|
||
# NOT "zero runtime deps" since 0.4.0: llm-ingestion-guard is okf's one runtime dependency,
|
||
# and okf >=0.8.5 declares `>=1.2,<2.0` for it. That is why the two lines below move together.
|
||
"llm-ingestion-okf",
|
||
"llm-ingestion-guard",
|
||
# U14 (økt 55, operatørbeslutning 23.08): the tracing seam's ONE new runtime dependency.
|
||
# `opentelemetry-api` was already here transitively via agent-framework-core, but the SDK is
|
||
# what `configure_otel_providers()` needs — without it MAF raises ModuleNotFoundError, so the
|
||
# seam cannot land as an opt-in without declaring it. The console exporter this framework uses
|
||
# (`opentelemetry.sdk.trace.export.ConsoleSpanExporter`) ships INSIDE this package, so console
|
||
# mode costs exactly one declaration. The OTLP exporter packages are deliberately NOT declared:
|
||
# they are egress, they drag grpc/protobuf into a published wheel, and MAF already raises a
|
||
# named ImportError telling the operator which one to install. Stated honesty limit, not an
|
||
# oversight — see tracing.py.
|
||
"opentelemetry-sdk>=1.42,<2",
|
||
# Door A content gate — scans materialized concepts before they reach the bundle; zero runtime deps, MAF-free (D7)
|
||
]
|
||
|
||
# Console entry points — the install surface a fresh clone gets from `uv sync` (P4 pkt. 5).
|
||
# Deliberately TWO, not five: `run` is the framework CLI (three documented modes) and `simulation`
|
||
# is the offline end-to-end proof the README points a newcomer at. `costsim` / `hitl` / `preflight`
|
||
# keep the `python -m` form — they are operator utilities, not the product's front door, and every
|
||
# name here is a name the freeze has to carry. Pinned by tests/test_console_entry_points.py.
|
||
[project.scripts]
|
||
portfolio-optimiser = "portfolio_optimiser.run:main"
|
||
portfolio-optimiser-demo = "portfolio_optimiser.simulation:main"
|
||
portfolio-optimiser-toolbox = "portfolio_optimiser.toolbox:main"
|
||
|
||
# Distribution channel for the shared ingest library (mirrors portfolio-optimiser-claude,
|
||
# verified in consumer CI): git pin against the public Forgejo repo — reproducible for every
|
||
# consumer, uv.lock pins the exact commit behind the tag. Bump the rev on a new library tag.
|
||
[tool.uv.sources]
|
||
llm-ingestion-okf = { git = "https://git.fromaitochitta.com/open/llm-ingestion-okf.git", rev = "v0.8.5" }
|
||
# Pinned to a TAG, never a floating ref — a security component whose version can move under a
|
||
# `uv sync` is a gate that can stop gating without a local diff (the okf `generated`-literal
|
||
# lesson, measured 2026-08-09). Bumping this is a decision, not maintenance.
|
||
# `tag =`, never `rev =`, and never a lower 1.x: okf >=0.8.5 pins this library itself via its own
|
||
# [tool.uv.sources] `tag = "v1.4.0"`, and uv treats `rev = "v1.4.0"` as a DIFFERENT url from
|
||
# `tag = "v1.4.0"` at the same value — both spellings were measured to fail resolution as
|
||
# "conflicting URLs" (P13 § 2a). The declared floor okf asks for is >=1.2, but 1.2.0 is not
|
||
# choosable for the same reason.
|
||
llm-ingestion-guard = { git = "https://git.fromaitochitta.com/open/llm-ingestion-pipeline-security.git", tag = "v1.4.0" }
|
||
|
||
# Dev tooling as a PEP 735 dependency-group (uv includes it by default in `uv sync`/`uv run`),
|
||
# so the documented bare `uv sync` + `uv run pytest` workflow installs it without `--extra`.
|
||
#
|
||
# INTENTIONAL, RECORDED DEVIATION (Fase 2 review F4): the original plan Step 1 authorized only
|
||
# moving `agent-framework-orchestrations` + `pulp` from the dev extra into core. Migrating the
|
||
# dev group from `[project.optional-dependencies]` to this PEP 735 `[dependency-groups]` block
|
||
# went beyond that stated scope, but is KEPT (not reverted) because it is what makes the brief's
|
||
# Success Criterion SC1 ("bare `uv sync` + `uv run pytest`") install dev tooling without
|
||
# `--extra` — reverting would break SC1. Logged here so the deviation is no longer silent.
|
||
[dependency-groups]
|
||
dev = [
|
||
"pytest>=8",
|
||
"ruff>=0.6",
|
||
"mypy>=1.11",
|
||
"pytest-asyncio>=0.24", # MAF orchestrations are async (await workflow.run(...)) — test-only
|
||
]
|
||
|
||
[build-system]
|
||
requires = ["hatchling"]
|
||
build-backend = "hatchling.build"
|
||
|
||
[tool.hatch.build.targets.wheel]
|
||
packages = ["src/portfolio_optimiser"]
|
||
|
||
# Fase 4a: the wheel CARRIES shared/ (spec, persona skill, example bundles) as packaged data —
|
||
# a byte-identical mirror under portfolio_optimiser/_shared/, resolved by shared_root() only when
|
||
# no working tree is present (checkout wins; PORTFOLIO_SHARED_ROOT wins over both). This is what
|
||
# makes an installed wheel — and a container — work without a clone, while shared/ itself stays a
|
||
# pull-only subtree at the repo root. Pinned by tests/test_shared_packaged_data_loadbearing.py.
|
||
[tool.hatch.build.targets.wheel.force-include]
|
||
"shared" = "portfolio_optimiser/_shared"
|
||
|
||
[tool.ruff]
|
||
line-length = 100
|
||
src = ["src", "tests", "spikes"]
|
||
|
||
[tool.pytest.ini_options]
|
||
pythonpath = ["src", "."]
|
||
testpaths = ["tests"]
|
||
asyncio_mode = "auto"
|
||
|
||
[tool.mypy]
|
||
python_version = "3.10"
|
||
|
||
# Third-party libs without bundled type stubs (py.typed) — analysed as untyped, not errors.
|
||
[[tool.mypy.overrides]]
|
||
module = ["pulp.*", "agent_framework_foundry.*", "llm_ingestion_guard.*"]
|
||
ignore_missing_imports = true
|
||
# NOTE on `llm_ingestion_guard`: the override alone would make the seam type-BLIND, not
|
||
# type-safe — every symbol arrives as `Any`, so a field rename upstream would type-check
|
||
# happily and fail at runtime. `ingest.materialize_gated` therefore coerces each value it
|
||
# reads off the guard's result objects to a concrete type at the boundary (the okf precedent),
|
||
# and the seam's behaviour is pinned by tests/test_ingest_content_gate_loadbearing.py.
|