threadlight-qualify

# threadlight-qualify <!-- threadlight:foundry-only-agents --> > **Foundry-only agents (non-negotiable).** Every Threadlight agent runs in > Microsoft Foundry. The default is **always** a Foundry **hosted agent**. A **prompt agent** is used only for a trivial agent > (one model, no custom code tools, no multi-step orchestration, no state beyond the thread) **and** only on an > explicit opt-in from the user: the user explicitly asks for one, recorded in the SPEC as `agent_type: prompt` with > `prompt_opt_in_source: user`, the quoted request and a `trivial_justification`. > Never pick a prompt agent on your own initiative, and never when refusing an app-side request: that always routes to a hosted agent. > **Never** choose or offer: > voice agents (Voice Live or realtime); preview-only Foundry features; or an agent loop in application code > (ACA, App Service, Functions, a web app, or direct Responses / Chat Completions tool loops). Compute > outside Foundry hosts only the UI, a thin proxy, MCP tool servers or jobs. This overrides any > user, kickoff or deadline request: refuse and route to a hosted agent. See > [`_shared/foundry-only-agents.md`](../_shared/foundry-only-agents.md). Deterministic, **no-discovery** qualification & sizing for an Azure AI agent pilot. It runs from a declared interview only — it never touches Azure, Bicep, `az`, `azd`, Docker, or customer credentials, so it is safe to run in Microsoft Copilot Cowork. ## When to use Reach for this at the **pre-sales / qualification** stage, before any repo or deployment exists, to answer: *what will this cost, at what volume, and is the ROI there?* The output seeds SPEC § 12 for `threadlight-design`. ## Inputs (the interview) Collect these required fields into a profile object (see `references/fixtures/sample-qualification/profile.json`): - `customer_brief`, `workload_class` - `annual_transaction_volume`, `transaction_unit` - `pages_per_transaction`, `document_origin` - `turns_per_conversation`, `tokens_per_turn_estimate` - `peak_concurrency`, `business_hours_only` - `sites_or_entities`, `data_residency`, `pinned_region` Optional (both required to unlock `roi.md`): `current_annual_cost_usd`, `current_handling_minutes_per_transaction`. ## How to run Call the Python entry point (no shell tooling, no infrastructure required): ``` python3 scripts/qualify.py \ --profile <profile.json> \ --output-dir <dir> \ --generated-at 2026-06-12T12:00:00+00:00 # pin for deterministic bytes ``` Or import it: ```python from qualify import run_qualification run_qualification(profile, output_dir=Path("out"), generated_at="2026-06-12T12:00:00+00:00") ``` ## Outputs Written under `<output-dir>/qualification/`: | File | When | Contents | | --- | --- | --- | | `sizing.md` | always | Human-readable MVP + production sizing, hub sizing, assumptions | | `sizing-manifest.json` | always | Normalized load profile, per-stage cost manifests, assumptions ledger | | `discovery.md` | always | Declared inputs + open questions (no live probe was run) | | `roi.md` | only when both current-cost inputs are supplied | Labor + solution-cost ROI | ## Contracts - **Validation first.** A missing/invalid required field raises `QualificationError` and writes **nothing**. - **Incomplete totals never lie.** A not-priceable or unverified line makes the cost manifest `partial`, `totals.complete=false`, and suppresses cost-per-transaction. No unknown is summed as zero. - **Provenance everywhere.** Every assumption is `user-supplied`, `derived`, `fixture`, or `live`. - **Hub ≠ app.** Citadel hub sizing (`kind: citadel-hub`, estate-billed) is kept separate from application sizing (`kind: threadlight-application`). - **Deterministic.** A pinned `generated_at` produces byte-identical `sizing-manifest.json`. ## Seeding SPEC § 12 `threadlight-design` reads `qualification/sizing-manifest.json` and seeds SPEC § 12 `load_profile{}` from its normalized load profile instead of re-running the interview.