Does it fit?

A candidate is checked against what is needed, one rule at a time, and gets one verdict from the ladder. The rules come from the item's category profile; the evidence comes from the inventory or from values written in. No model takes part.

Two sides

• The need: what the part must satisfy. Either another inventory item (the laptop that needs an adapter, the frappé maker whose agitator wore out, with the required values recorded on it under Add what you know), or values written straight into the form.
• The candidate: the item whose page you are on. Its values are the ones it currently shows: accepted ones first, else the latest observation still to confirm, each with its provenance. A value that came from a measurement carries its uncertainty into the check.

Needed values are written in the candidate's terms: for a vacuum tool, the needed opening_inner_diameter is the wand's outer diameter; for an agitator, the needed tab_span is the worn rod's own span, the one the drive takes. A side can be an item and values at once; a written value replaces the item's for that key. Numbers are compared in the rule's unit: a value on either side recorded or written in another unit (3300 mA against a rule in amps) stays unknown and asks for the right one.

Rules

Each rule names a key, whether it is hard (decides the verdict) or soft (reported), and how it compares:

Kind	Passes when
equal	both sides read the same after the label normaliser (polarity, connector, part number); a material key reads as metal or plastic however it is spelled, and a mix (including wood or ceramic), a negation, an uncertain claim, a component or a surface finish stays as written
within	the candidate is within the tolerance of the needed value: an absolute amount, a percentage, or the larger of the two
at_least	the candidate is at least the needed value, less any slack
at_most	the candidate is at most the needed value, plus any slack

A numeric check takes the candidate's uncertainty into account. A measured 4.55 mm ± 0.3 mm against 4.5 mm ± 0.1 mm is neither a pass nor a fail: the check stays unknown and asks for a caliper reading. A rule marked identity (part or model number) is soft, and a pass on it makes the match EXACT_MATCH.

What comes back

The verdict with the one line that decided it, every check with the needed value, the candidate's value and its backing (provenance, whether it is still to confirm, its uncertainty), the result and the basis in words; the questions to ask the seller for every unknown on the candidate's side; the rules the need gives no value for; and the checks that rest on observations not yet accepted.

In the web app

On an item's page, Does it fit? lists the other items under Needed by; pick one, or leave Values I type below and fill the rules' fields, then Check fit. The verdict, the checks and the questions appear under the form. The item is the candidate, so the panel is behind the access code; the demo visitor's tries go through POST /api/assess with values on both sides.

From the API

curl -s localhost:8000/api/assess -H 'content-type: application/json' -d '{
  "need": {"profile": "power_adapter", "values": {"output_voltage": "19.5 V", "polarity": "center positive"}},
  "candidate": {"item_id": 3}
}'
curl -s localhost:8000/api/profiles/power_adapter/rules

See HTTP API and the profile schema.