Models
The find engine
Scores a candidate product on three independently sourced legs, and refuses to average them into one comfortable number.
A product surface, not a published identifier. It runs on the shared inference path and carries no identifier of its own.
What it is for
The find engine decides which products enter an operator's feed, and what may be said about each one. It is the surface with the strongest claim to being deterministic: the legs, their floors and their decay are arithmetic over dated evidence rows, not a judgement a model is asked to make.
The same engine runs against a product an operator names themselves. The analyse tool returns the same three-leg reading, on the same evidence rows, for a product someone already sells.
What it can do
Three legs, separately sourced
Urgency asks whether there is a reason to buy now rather than later. Scarcity asks whether supply is constrained in a way a buyer can feel. Feasibility asks whether the thing can actually be sourced, shipped and sold. Each leg must stand on its own dated sources; an aggregate may not rescue a leg that has none.
The weakest leg sets the age
For each leg the engine takes its freshest source, then takes the oldest of those three. A find is only as current as its most neglected axis. Evidence under ninety days counts at full strength and decays towards a floor at one year, so confidence falls as inputs go quiet rather than waiting to be corrected.
A failing leg is ranked, not hidden
A product below the floor on any leg ranks last. It is not silently removed, because a removed candidate cannot be argued with.
Exclusivity as a database property
A hold is a thirty-minute reservation while the operator decides; several people may watch the same unclaimed product at once. A claim is permanent, and exactly one can exist, enforced by a unique index on the product in the claims table rather than by application logic.
How it is measured
Not covered by commerce-v1. The benchmark measures the commerce engine's reasoning over tool evidence; it does not measure how that evidence was gathered or scored.
Because no recorded run exercises this surface, this page states no measurement of it at all. The benchmark and its system cards are published under research; they describe commerce-reasoning-v1 and nothing else.
Its limits
- The score is a reading of evidence rows, not a prediction of sales. It says what has been observed, not what will happen.
- A leg with no sources is missing, and the engine says which leg is one evidence step short rather than filling the gap.
- Exclusivity applies to the whole inventory, so a product under another founder's claim or hold is not shown as available. One walkthrough product is deliberately shared so a first-time visitor can complete the journey without consuming real inventory.
- commerce-v1 does not exercise this surface. Nothing on the benchmark speaks to how well the legs are scored.
Where these facts come from
Each line names a file in the implementation. The descriptions on this page are a transcription of those files and are marked provisional at their source until they have been checked against it again.
- The three-leg scoring — flowfinds-ai-tool/backend/hunt.py
- Capabilities implemented today — flowfinds-ai-tool/PRODUCT-MAP.md
The others
- commerce-reasoning-v1 — The decision engine an operator talks to: it investigates a question with application tools before it answers, and it names the evidence it used.
- The store editor — Turns an instruction in the operator's own words into a proposed change with a cited reason, a rejected alternative, and a reversibility flag.
- The organic research engine — Researches an audience per platform and produces a framing brief. It has no posting function, and no credential it could post with.
- The support agent — Answers a customer from one product's own record, and is scanned on the way out for the things it must never say.
Back to all models, or read what the engine is used for under the product.