The map is still expanding — discovering where robots can operate economically, and what each market requires, is part of the research. A hundred and thirty-five markets, grouped from institutional scale down to the ones a single operator can claim this week. Each entry is a deployment, an experiment or a blueprint at its own evidence level: the state records where P34 has already been pointed — not a ranking, and not a recommendation of where you should start. The industry cases and the enterprise track are deep dives into a handful of these; they come after.
Four criteria, and a market has to clear all of them:
The same list, structured — ids, groups, evidence states, menu shape and data sources per market. Fetch it directly rather than parsing this page:
Read order matters: this catalogue is the overview, cases.html and enterprise.html are the deep dives beneath it. Every entry carries a state — a listing is an evidence level, not a claim that P34 ships that market today.
Read broadly, a market is not necessarily an exchange. It is any recurring environment where an operator sees a large feasible action menu, chooses a small subset, can eventually measure the economic outcome, and can state the constraints well enough for a policy to improve. What makes such an environment computable is that the history of what was chosen — and what was passed over — leaves a signal.
That passed-over set is the reject market: the actions whose gross value was above zero but below the cost, capability, speed or confidence threshold required for anyone to discover and execute them. A computable reject market is one where most opportunities are filtered out by people, rules, budgets or old models — and AI can systematically re-evaluate what was left behind.
Instead of looking only for obvious winners, the system learns the rejection frontier: which opportunities were correctly rejected, which were missed, and which accepted deals should actually be avoided. As markets become more efficient this frontier becomes more important, because the easy profit disappears first. Nearly every entry below collapses onto one of seven reasons a reject exists.
Nobody evaluated it. The menu was larger than the number of decisions a person could make.
Somebody evaluated it and was wrong — the estimate, not the opportunity, was the defect.
The ticket was too small for the incumbent's cost of doing business, whatever its return.
It was bad for them and good for you. A load going the wrong way for one carrier is a backhaul for another.
They could not take it even though it was attractive — no truck, no crew, no balance sheet, no slot.
Too many interacting variables made evaluation itself uneconomic.
Somebody could have taken it, but not quickly enough for it to still be there.
A marketing team, sales team, procurement group or operations department can treat its work as an internal market: capital goes in, a large menu of possible actions is considered, most are rejected, and the selected actions produce measurable economic returns. P34 can optimize these decisions across multiple layers — what to fund, what to avoid, how much to allocate, and when to do nothing — turning each department into a continuously learning internal capital-allocation engine.
Nothing in the catalogue below is a prerequisite for this: an internal market needs no counterparty, only a recorded menu and a measurable outcome. The enterprise track is where that runs inside a governed operating envelope.
A surprising amount of small-business margin exists precisely because opportunities are geographically fragmented, operationally messy, irregular or information-intensive — expensive for efficient capital to pursue. The research question this catalogue exists to test is whether an agent lowers those thresholds enough to convert part of a reject set into an operable market. It is a hypothesis under test, not a claim about returns, and the evidence state on every entry below says how far it has actually been taken.
Start from the market you already operate in, or one you know well. That is the single strongest predictor of whether this works for you. What makes P34 useful on a market is your data, your constraints and your operating knowledge — a market you understand beats a market with a pre-built workflow.
A reference deployment is not a recommendation. Amazon wholesale is the market P34 is most developed on, and it is also one of the hardest to enter. The difficulty is not the model: Amazon account management — ungating, brand and IP complaints, performance metrics, suspension and reinstatement — and wholesale supplier relationships — winning authorised distributor accounts at all, minimums, credit terms — are demanding operating problems, and P34 does not solve either of them. Do not prioritise it because we started there.
Use the reference deployment and the active experiments as evidence that the method works, then run the fit check on your market. If it clears the four criteria and you can bring the telemetry, that is the market to point P34 at — and we want to hear about it.
Ordered by scale and by how much of the work is data rather than hands. Jump to core, tier 1, tier 2, tier 3, tier 4, demand, capacity, infrastructure, industrial, contracts or flagged.
Institutional scale — the markets HyperC and its enterprise partners work directly. Large tickets, deep telemetry, and in several cases a live deployment behind them.
The founding deployment: purchase and shipping portfolio decisions from live wholesale menus, in production since 2023 at roughly $100M/yr reseller scale (company-reported).
Hard to operate: The best-understood market here and one of the hardest to operate — and the hard parts are not the model. Amazon account management (ungating, brand and IP complaints, performance metrics, suspension and reinstatement) and wholesale supplier relationships (winning authorised distributor accounts at all, minimums, credit terms) are demanding operating problems that P34 does not solve. Enter it because you already run it, not because it is the developed one.
Menu: SKU × quantity × supplier offer, with lead times and fees · Data: Supplier price lists, catalogue and fee data, sales velocity, returns
Lend / no-lend calls where credit history is sparse or absent. About 3,000 loans issued in a live model test (2025). Regulated commerce: gated behind market-specific compliance review.
Menu: Applicant × offered principal × term · Data: Application attributes, repayment tape, collections outcomes
Routing and execution across venues in adversarial order flow. Deployed in research; a separate product and regulatory perimeter, not part of membership.
Menu: Route × size × venue, per transfer · Data: Venue depth, spread and fee telemetry, settlement latency
Which invoices to settle, in which currency and when, under liquidity and corridor constraints — a scheduling problem with a measurable cost of being wrong.
Menu: Invoice × corridor × settlement date · Data: Corridor rates and fees, liquidity positions, invoice ageing
Offer selection and sizing scored on lifetime player economics rather than redemption rate, with abstention when the economics do not clear the bar.
Menu: Player segment × offer × value · Data: Play history, redemption and margin tape
Allocation and limit decisions across traders and strategies. Financial perimeter: excluded from profit-share pricing and gated under the API Terms.
Menu: Trader/strategy × capital allocation × limit · Data: Realized P&L tape, drawdown and exposure history
Manager selection as a menu decision under partial observability: candidate managers, sparse comparable histories, a measurable recovery target. Enterprise discovery case.
Menu: Candidate manager × commitment size · Data: Historical manager performance, recovery and default records
Limit increases, overdraft and micro-loan offers treated as one portfolio decision scored on expected lifetime economics. Designed for shadow evaluation against the incumbent policy first.
Menu: Cardholder × offer type × limit · Data: Balance and utilisation history, loss tape, incumbent policy decisions
Onboarding incentives selected and sized on lifetime economics rather than conversion alone — refusing the offers that will not pay for themselves.
Menu: Prospect segment × incentive × value · Data: Onboarding cohorts, product uptake, lifetime margin
Retail and online arbitrage run as a portfolio on public telemetry, pointed at small-ticket buyers rather than wholesale lots.
Menu: Discounted listing × quantity · Data: Keepa-class price and rank history, fee and returns data
Structurally a menu problem, but a financial-market perimeter. Excluded from membership and from profit-share pricing.
Menu: Asset × position size × horizon · Data: Venue price history, liquidity and flow telemetry
Illiquid and non-standard asset selection. Same perimeter and same exclusion as the above.
Menu: Asset × allocation · Data: Sparse comparable sales, holding-cost and carry data
Advance / decline and sizing against future receivables, where the historical book is severely selection-biased — the failure mode P34 is built for.
Menu: Merchant × advance amount × factor rate · Data: Processing volume history, repayment and default tape
Which partial settlements to accept on delinquent balances, and at what discount — a recovery-maximisation menu, not a collections script.
Menu: Account × settlement offer × discount · Data: Delinquency ageing, historical recovery outcomes
Listed because it is repeatedly proposed, and repeatedly declined: prediction markets are a regulated-market use, excluded under the API Terms of Use and from profit-share pricing. Do not infer support from anything else on this page.
Menu: Contract × stake · Data: Public order books and settlement history
Supplier, quantity and timing selection against demand forecast and working-capital constraints — the buy side of the wholesale problem, inside an enterprise.
Menu: Line item × supplier × quantity × delivery window · Data: Quotation history, consumption records, supplier performance
Payment and plan-assignment decisions scored on realized cost rather than on rules written a year ago.
Menu: Claim/plan × payment decision · Data: Claims tape, plan cost history
Underwriting delay risk on shipments: enumerable consignments, measurable outcome, sharply biased history of what was previously covered.
Menu: Consignment × cover × premium · Data: Clearance timing records, route and broker history
Cover / decline and pricing on transaction fraud exposure, where refusal is most of the value.
Menu: Merchant/transaction class × cover × premium · Data: Chargeback and fraud-loss tape
Businesses conventional underwriting declines because the file is thin or the shape is unfamiliar, not because the economics fail. Regulated commerce: gated behind market-specific compliance review.
Menu: Business × principal × term × price · Data: Bank-transaction and accounting feeds, repayment tape, incumbent decline reasons
Invoices and firms left unfinanced because diligence per invoice costs more than the spread. The rejected set is observable where a factor logs what it turned down.
Menu: Invoice × advance rate × fee × term · Data: Invoice and ageing tape, debtor payment history, dilution and dispute records
POs declined because underwriting the buyer, the supplier and the shipment at once is expensive relative to ticket size.
Menu: PO × funded share × fee × expected settlement · Data: Buyer credit history, supplier delivery record, shipment and settlement timings
Transactions beneath a lender's attention threshold, where the asset is recoverable but the file is too small to justify manual review.
Menu: Applicant × asset × advance × term · Data: Asset resale comparables, obligor history, recovery and repossession outcomes
Applicant-vehicle-downpayment-term combinations declined by scorecards built on approved-applicant data only. Regulated: fair-lending review is a precondition, not a formality.
Hard to operate: Consumer credit. Adverse-action, fair-lending and disparate-impact obligations attach to every decline, including the ones a model makes. This is a compliance project with a model inside it.
Menu: Applicant × vehicle × advance × term × rate · Data: Application and decline records, vehicle valuation series, repayment and recovery tape
Customer, limit and payment-term combinations suppliers refuse — a decision usually made by policy table rather than by expected economics.
Menu: Customer × credit limit × payment term · Data: Order and payment history, ageing, write-off tape
Carrier invoices skipped because verification against the load and the broker costs more than the fee on a small haul.
Menu: Carrier × invoice × advance × expected payment date · Data: Load and rate confirmations, broker payment history, settlement timings
Small unusual risks declined or priced defensively because they do not fit a rating table. Regulated: rate filing and licensing sit outside the model.
Hard to operate: Insurance is licensed, filed and supervised per jurisdiction. A pricing model is not a rate filing, and binding authority is not something P34 confers.
Menu: Risk × coverage × limit × premium · Data: Submission and decline records, loss runs, exposure characteristics
The purest computable structure available to an individual operator: public comparable-sales data, enumerable listings, small units, fast settlement. The best place to start if you are claiming your first market.
The purest example on this list: about $10 units, 500-name portfolios, instant settlement, zero shipping.
Menu: Dropping name × bid · Data: GoDaddy / Dynadot auction and sales history
Arguably the finest public comparable-sales dataset in any collectibles market.
Menu: Pressing × condition × price · Data: Discogs sales history
Clean per-SKU time series with condition tiers already standardised by the market.
Menu: Title × condition tier × quantity · Data: PriceCharting time series
High spread, strong seasonality, and unusually forgiving buyers.
Menu: Model × condition × price · Data: Reverb price guide, completed sales
The appreciation curve after set retirement is remarkably modelable.
Menu: Set × sealed/used × quantity · Data: BrickLink price guide and sales
Grading arbitrage sits on top as a second, separate edge.
Menu: Card × grade × quantity · Data: TCGplayer, eBay sold, PSA population reports
Sealed appreciation follows print-run and rotation mechanics rather than taste — mechanical, therefore learnable.
Menu: Product × sealed lot size · Data: Print-run and rotation schedules, sold comps
Margins are thin now, but the telemetry is institutional-grade.
Menu: Style × size × quantity · Data: StockX / GOAT bid-ask and sales history
The existing wholesale pipeline pointed at a $500 buyer instead of a $500k one.
Menu: Listing × quantity · Data: Keepa price and rank history, fee schedules
Brutal, predictable seasonality. Abstention discipline matters more here than anywhere else in Tier 1.
Menu: ISBN × edition × quantity · Data: Sold comps, semester calendars, edition-change notices
Same menu structure, but the operator touches physical goods, estimates repair cost, or has to be somewhere. Spreads are wider because of it.
Enormous spread on local marketplaces, near-instant liquidity.
Data: Marketplace sold listings
Deep, well-catalogued comparable history and a stable collector base.
Data: KEH, eBay sold
Reference-level comps; the risk is authentication, not pricing.
Data: Chrono24 comps
Sharply seasonal with high local spread — a timing problem as much as a selection one.
Data: Local marketplace sold listings
Core charges create a price floor, which bounds the downside.
Data: Parts catalogues, core-charge schedules, sold comps
Very long tail and very cheap entry — a good place to learn refusal.
Data: AbeBooks, Amazon sold history
An ageing collector base is itself a modelable signal.
Data: Auction records, collector marketplace comps
The repair-cost estimate is the hard variable; everything else is well observed.
Data: Sold comps, parts pricing, repair-time records
Discontinuation announcements are a clean event trigger.
Data: Discontinuation notices, sold comps
Lot-level pricing against per-unit resale telemetry.
Data: Wholesale lot offers, resale sold comps
Lumpy outcomes; manifest quality is the whole game.
Data: Pallet manifests, historical recovery rates
High variance and thin telemetry, but fully enumerable and genuinely menu-shaped.
Data: Auction listings, realized resale outcomes
Badly catalogued — which is exactly where the mispricing lives.
Data: Auction catalogues, hammer prices
Per-item scanning: thousands of tiny decisions an hour, which is a volume the model is built for.
Data: Scanner lookups, sold comps
Larger tickets with excellent public telemetry and repair-cost history.
Data: Copart / IAAI auction results
Composition modelling from photographs against commodity prices.
Data: Commodity price feeds, recovery yields
The aroid boom showed how quickly telemetry-driven pricing forms in a new collectibles market.
Data: Marketplace sold history, auction results
Deeply under-served by software, with regular, enumerable sale events.
Data: Sale-barn and auction records
Supplier lots existing resellers pass on. Same menu shape as the Amazon deployment against a different fee, velocity and returns regime.
Menu: SKU × quantity × supplier offer × marketplace · Data: Supplier price lists, marketplace fee schedules, rank and velocity history
Store-level clearance other resellers ignore because it is scattered across locations and priced per store rather than per catalogue.
Menu: Store × SKU × price × quantity × exit channel · Data: Store-level clearance feeds, comparable sold prices, fee and shipping tables
Overstock lots rejected by normal distribution channels — the goods are fine, the channel is full.
Menu: Lot × price × quantity × onward channel · Data: Excess and overstock listings, comparable trade prices, carrying-cost assumptions
Brand, SKU and quantity deals buyers decline because the margin looks too thin before fees, returns and sell-through are modelled properly.
Menu: Deal × quantity × price × expected sell-through · Data: Closeout offer sheets, comparable sold prices, returns and fee history
Return lots rejected because condition and recoverable value are hard to estimate from a manifest. The estimation error is the market.
Menu: Lot × bid × grading assumption × exit channel · Data: Manifests, historical grade-out rates, per-condition sold comparables
Devices declined at particular acquisition and repair-cost combinations — a joint estimate of parts, labour, failure rate and resale.
Menu: Device × acquisition price × repair spend × exit channel · Data: Parts and labour costs, repair success rates, per-condition sold comparables
Low-volume parts nobody stocks because expected demand is hard to estimate — a portfolio decision across thousands of slow-moving SKUs.
Menu: Part × stocking quantity × reorder point · Data: Historical consumption, lead times, obsolescence and equipment-install base
No shipping, no storage. The telemetry is install counts, download history, traffic and expiry schedules. Fraud checking is the operator's main defence.
Sub-$5k tier: enough listings per month to be a menu, not a one-off deal.
Data: Flippa, Acquire.com listing and sale history
Install counts are the telemetry, and they are public.
Data: Store install counts, review velocity
Hundreds of tiny bets with marketplace search volume as the demand signal.
Data: Marketplace search volume, sales history
Download history is public on most platforms, which makes revenue projectable.
Data: Platform download and revenue history
Long-tail royalties against an acquisition price — a clean discounted-cash-flow menu.
Data: Sales rank history, royalty statements
High fraud rate — verify hard. Inflated metrics are the dominant failure mode, not mispricing.
Data: Public analytics, revenue verification
Thin, but genuinely computable and instantly settled.
Data: Raise / CardCash spreads and fill history
Devaluation risk is the modelable variable, and it is unusually well documented.
Data: Award charts, devaluation history
Recurring rather than one-shot: placement, utilization and routing decisions where the menu is rebuilt every period. Under-served by software, which is the opportunity.
Location selection is a textbook menu problem, and foot-traffic data is purchasable.
Data: Foot-traffic data, per-site takings history
Utilization telemetry, recurring rather than one-shot: the menu rebuilds every period.
Data: Utilization and booking history, replacement costs
Buy leads, resell to contractors. Menu, budget constraint, savagely biased history — an excellent fit, and nobody models it.
Data: Lead cost, contact and conversion outcomes
Pricing and acceptance decisions per slot per period.
Data: Occupancy and rate history
Windows, boards and community screens: enumerable inventory, measurable response.
Data: Placement inventory, response rates
Structured sub-tasks on freelance marketplaces have the properties of a computable market: an enumerable menu of jobs, a bid, and a realized margin per job.
Menu: Job posting × bid × delivery slot · Data: Platform job feeds, historical bid/win and delivery cost
Appliance repair, A/C installation and similar trades: accept, price and schedule jobs against crew capacity.
Menu: Job × price × scheduled slot · Data: Job history, crew cost and utilization
Buy the asset, rent it through its useful window, sell it at the right point — utilization and exit price decided jointly.
Menu: Asset × purchase price × rental period × exit date · Data: Rental rates and utilization, resale comps
The menu is a cell in a grid — audience × creative × offer × time × placement × bid — and a human instantiates a minute fraction of it. Everything never tested is the reject set, and the outcome is measurable per cell.
Keyword, bid, geography and time combinations campaigns never buy. The menu is enumerable and the outcome per cell is measurable, which is the whole requirement.
Menu: Keyword × match type × geo × time × bid · Data: Search-term and auction reports, conversion and margin tape, incumbent bid policy
Audience combinations excluded from existing campaigns — usually by habit or by a rule written once, not by measured economics.
Menu: Audience × creative × placement × budget · Data: Delivery and conversion reports, cohort margin, historical exclusions
Impressions declined below a bid threshold. Attribution is the hard part and it is the reason the reject set stays large.
Hard to operate: Attribution on display is contested and view-through effects are easy to overstate. Without a clean incrementality design the measured outcome can be an artefact of the measurement.
Menu: Placement × audience × time × bid · Data: Bid-stream and win/loss logs, post-click and incrementality tests, margin tape
Users not retargeted because simple recency-and-frequency rules rank them poorly, though their expected economics may clear.
Menu: Cohort × window × frequency cap × bid · Data: Session and cart telemetry, conversion lags, contribution margin per cohort
Publishers and traffic sources merchants decline wholesale, where the honest answer varies by source rather than by category.
Hard to operate: Affiliate fraud and incentivised traffic are endemic; a source that looks profitable on last-click can be buying credit for demand you already had.
Menu: Publisher × offer × commission × cap · Data: Per-source conversion and refund rates, chargeback and fraud history
Leads buyers refuse at a given price. Distinct from generating leads (see local lead arbitrage): here the menu is priced inventory someone else produced.
Menu: Lead source × attribute set × price × volume cap · Data: Contact-to-close rates by attribute, revenue per closed lead, refund and dispute records
Accounts a sales team never works because human capacity, not expected value, sets the list length.
Menu: Account × contact × sequence × timing · Data: Firmographic and intent signals, historical win rates by segment, deal margin
Customer, message and timing combinations never tested — a grid with millions of cells and a few dozen instantiated.
Menu: Segment × message × send time × frequency · Data: Send, open and conversion logs, unsubscribe cost, per-cohort margin
Discounts not offered to particular cohorts — and, just as often, discounts given to cohorts that would have bought anyway. Refusal is the valuable half.
Menu: Cohort × discount depth × channel × window · Data: Redemption and holdout results, baseline purchase rates, margin per order
Product, keyword and position bids sellers never place, on inventory whose own margin they already know exactly.
Menu: Product × keyword × position × bid · Data: Advertising reports, organic-rank series, unit economics per SKU
Rolling menus of loads, lanes, slots and spaces. What one operator must decline — wrong direction, wrong date, too small — is exactly what fits another, so the reject set is operator-dependent rather than uniformly bad.
Loads carriers decline. The classic operator-dependent reject: a load that is wrong for one truck is right for the truck that needs that lane today.
Menu: Load × lane × pickup window × rate · Data: Board postings and rate history, deadhead distance, hours-of-service and fuel cost
Small shipment combinations nobody assembles because the search over compatible pairs is combinatorial and manual.
Menu: Shipment set × route × sequence × price · Data: Shipment dimensions and windows, lane rates, terminal and handling costs
Return-leg loads missed or declined — value that exists only because the truck is already going that way.
Menu: Return leg × candidate load × rate × delay tolerance · Data: Fleet position and schedule, historical lane rates, empty-mile cost
Routes and jobs that look unattractive under simple dispatch rules but clear once sequencing and drop density are priced properly.
Menu: Job set × route × time window × pay · Data: Historical route times, drop density, failed-delivery and return rates
Time, location and delivery combinations left unassigned when demand and couriers are both moving.
Menu: Courier × job × time slot × price · Data: Job and completion logs, positioning history, cancellation rates
Empty moves operators decline, where the cost of being in the wrong place later is the real quantity being traded.
Menu: Container × origin/destination × date × cost · Data: Equipment positions, imbalance forecasts, per-lane repositioning cost
Small cargo blocks not accepted on particular flights because the handling overhead per block is human.
Menu: Flight × block size × rate × handling window · Data: Capacity and load-factor history, per-lane rate series, handling cost
Shipment, lane, date and rate combinations that go unbooked while equivalent capacity sails empty.
Menu: Shipment × lane × sailing date × rate · Data: Sailing schedules, historical spot and contract rates, rollover history
Short-term storage requests warehouses reject because the request does not fit the shape of a standard contract.
Menu: Site × pallet count × duration × rate · Data: Occupancy and turnover history, handling cost per pallet, seasonal demand series
Space, time and price combinations conventional pricing leaves unused. Operator-side pricing rather than leasing out your own space.
Menu: Space class × time block × price · Data: Occupancy telemetry, event and demand calendars, enforcement and turnover data
Perishable capacity priced by time. A GPU-hour, a curtailment window or a bandwidth block that goes unsold is gone, which makes the decline decision continuous and the outcome cleanly attributable.
Workload, hardware, duration and price combinations that never get matched. Capacity is perishable and the outcome per job is measurable, which is a clean menu.
Menu: Workload × GPU class × duration × price × SLA · Data: Spot price series, job runtime and failure history, energy and interconnect cost
Reservations owners do not reuse efficiently — committed spend that expires unconsumed while equivalent demand pays on-demand rates.
Hard to operate: Provider terms govern whether commitments may be transferred or resold at all, and they differ per provider. Read them before modelling anything.
Menu: Commitment × workload × window × internal price · Data: Commitment inventory and expiry, utilisation history, on-demand price series
Small compute jobs that cannot justify a human sales conversation, against hardware already bought and idle.
Menu: Cluster window × job × price × priority · Data: Utilisation telemetry, job queue history, marginal power cost
Low-priority jobs rejected by conventional scheduling, where deferral cost is small and capacity is cheap at the right hour.
Menu: Job × window × priority × price · Data: Queue and completion history, deadline sensitivity, time-of-day cost
Workloads declined at particular power and time combinations — a scheduling problem against a hard physical envelope.
Menu: Workload × power envelope × time block · Data: Power draw telemetry, tariff and demand-charge schedules, thermal headroom
Load-curtailment opportunities businesses leave unmonetised because the operational cost of curtailing is never quantified against the payment.
Menu: Site × load block × curtailment window × payment · Data: Interval meter data, programme price signals, production-loss cost
Time, location and charge/discharge actions conventional rules skip, where degradation is a real cost that a naive optimiser ignores.
Menu: Asset × time block × charge/discharge × market · Data: Price series, state-of-charge and degradation curves, connection limits
Local generation and time combinations poorly monetised under flat export tariffs.
Menu: Site × export block × price × storage decision · Data: Generation and consumption telemetry, tariff schedules, weather forecasts
Capacity, route and duration deals carriers do not assemble because the search across routes and terms is manual.
Menu: Route × capacity × term × price · Data: Route inventory and utilisation, historical transit pricing, latency and SLA records
Temporary capacity that is never dynamically marketed, so it is held idle against an outage that mostly does not happen.
Menu: Site × capacity × standby window × price · Data: Link utilisation, outage history, contractual burst allowances
B2B goods and slots where matching is the hard part: surplus, by-products, off-spec lots and idle production time that conventional channels skip because valuation and counterparty search cost more than the ticket.
Vehicle and bid combinations dealers decline. Distinct from salvage: these run, and the estimate is reconditioning plus retail days-to-turn.
Menu: Vehicle × bid × reconditioning spend × exit channel · Data: Auction result history, condition reports, retail comparable prices and turn times
Machines rejected because repair and resale economics are uncertain and inspection is expensive.
Menu: Machine × bid × repair spend × resale channel · Data: Auction comparables, hours and service records, parts and transport costs
Factory equipment no conventional reseller wants to underwrite, where the buyer set is small, findable and specific.
Menu: Asset × price × removal cost × buyer segment · Data: Comparable sales, rigging and transport quotes, install-base and demand signals
Excess steel, lumber and fixtures ignored because matching a specific lot to a specific job is the expensive part.
Menu: Lot × spec × location × price · Data: Project and tender pipelines, commodity price series, haulage cost
Crop lots conventional channels skip on grade, size or location — perishable, seasonal, and priced by whoever shows up.
Hard to operate: Perishability and food-safety rules are unforgiving: a modelling error here spoils rather than sits, and handling is licensed in most jurisdictions.
Menu: Lot × grade × location × price × exit window · Data: Grade and yield records, regional price series, storage and haulage cost
Inventory rejected because remaining shelf life against sell-through rate is a calculation nobody does per lot.
Hard to operate: Date-code, labelling and cold-chain rules govern what may be resold and where. Treat them as constraints, not as friction.
Menu: Lot × remaining shelf life × price × channel · Data: Sell-through rates by channel, date-code data, storage and disposal costs
Secondary outputs firms currently pay to dispose of. The interesting case: a search over combinations where negative-value disposal becomes positive-value supply for someone else.
Hard to operate: Waste classification, transport and permitting law decides whether a stream is a product or a regulated waste — and getting that wrong is a serious matter, not a paperwork slip.
Menu: Stream × specification × counterparty × price · Data: Composition and volume records, disposal cost, buyer specifications and permits
Production slots factories do not monetise, where the marginal cost of a slot is known and the search for a fitting job is not.
Menu: Machine group × slot × job type × price · Data: Utilisation and changeover history, marginal cost per slot, quote win/loss records
Custom jobs factories reject because quotation and setup overhead dominates the ticket — a human transaction cost, not a negative margin.
Menu: Enquiry × batch size × setup × quoted price · Data: Historical quotes and win rates, setup and run times, scrap and rework rates
Jobs declined for human transaction cost rather than negative value — reading the tender, estimating, scheduling, chasing. The reject frontier here moves when the cost of evaluating a job falls.
Plumbing, electrical and HVAC jobs declined because the scheduling is awkward rather than because the job is unprofitable.
Menu: Job × technician × time slot × quoted price · Data: Job duration and callback history, travel times, per-job margin
Sites and bids operators never pursue, where the estimate is area, frequency, staffing and travel.
Menu: Site × scope × frequency × bid · Data: Historical bids and win rates, labour hours per site, retention history
Small local contracts beneath the cost of a manual sales visit — a route-density problem as much as a pricing one.
Menu: Property × service plan × season × bid · Data: Route density and travel time, crew hours per property, renewal rates
Origin, destination and date combinations movers reject, where the return leg and crew utilisation decide the answer.
Menu: Job × crew × date × route × price · Data: Historical job times, damage-claim rates, backhaul and utilisation data
Low-probability or geographically inconvenient jobs, priced without knowing whether the part will be in the van.
Menu: Call × technician × parts kit × price · Data: First-time-fix rates, parts consumption, travel and diagnostic times
Underpaid and denied claims providers do not pursue because the expected recovery does not cover the human follow-up.
Hard to operate: Patient data is regulated health information. Access, retention and business-associate obligations are the design constraint here, and they come before any modelling.
Menu: Claim × appeal path × effort × expected recovery · Data: Remittance and denial-code history, payer-specific appeal outcomes, effort per claim
Small overdue invoices not worth human collection effort — first-party follow-up on your own ledger, not buying someone else's claims.
Hard to operate: Chasing third-party consumer debt is a regulated activity in most jurisdictions and is a different market from managing your own ledger; see judgment and receivables purchasing under Flagged.
Menu: Invoice × contact sequence × timing × settlement offer · Data: Ageing and payment behaviour, dispute reasons, per-action recovery rates
Opportunities firms never analyse because reading the tender costs more than the option is worth. The reject set here is created almost entirely by evaluation cost.
Menu: Tender × bid/no-bid × price × scope · Data: Published tenders and awards, historical win rates by category, delivery cost records
These have textbook menu structure, which is exactly why they are listed. They also carry legal, regulatory or platform-rule exposure. HyperC declines them under the API Terms; they are documented so nobody rediscovers them the expensive way.
Perfect menu structure — and state-regulated, with long resolution horizons. Counsel first.
Collections law applies to the operator, not only to the seller.
Resale is restricted or criminal depending on state and venue. Declined by policy.
Outright banned in several cities. Declined by policy.
Terms-of-service violation with account-death risk on both sides of the trade.
UDRP proceedings will eat the portfolio. Distinct from ordinary expiring-domain drops.
Claims too small for conventional collection economics. Debt collection is a licensed, heavily supervised activity and HyperC declines it under the API Terms; listed so the structure is not rediscovered the expensive way.
Menu: Claim × action × cost × expected recovery · Data: Claim and judgment records, recovery outcomes
Small positions institutional investors cannot economically analyse. Securities: a separate regulatory perimeter, excluded from profit-share pricing and gated under the API Terms.
Menu: Security × position size × entry/exit · Data: Price and volume series, filings, liquidity measures
Small temporary pricing discrepancies participants leave. Securities perimeter; listed as structure, not as an offer.
Menu: Fund × size × entry/exit window · Data: NAV and market price series, creation/redemption data, spread history
Executable states that disappear before slower participants act. Quoted price alone is not the opportunity — size, liquidity, exit and impact usually dominate it. Financial perimeter.
Menu: Venue pair × size × latency budget · Data: Order-book snapshots, fee and settlement terms, realised slippage
Six criteria from the published market-fit framework. Answer honestly — P34's whole discipline is refusing what doesn't clear the bar.
This check reflects structural fit only. Reference-deployment status additionally requires data validation, model coverage, execution understanding and risk controls under a defined methodology.
Each reference deployment ships with a capability matrix: supported data, action types, automation level, known failure modes, capital assumptions, execution dependencies and model version. If we can't publish that matrix, the market isn't a deployment — it's an experiment or a blueprint, and it says so above.
Members discover markets we haven't listed, and several entries here started as one member's experiment. New market experiments and the skills to operate them are published to members weekly, and land in this catalogue as they are confirmed. Bring your market: the 135 entries here exist to teach you how to formulate the 136th.
Having found your market above, these are the two ways down into it.
Named deployments and proposals from this catalogue: the workflow before P34, its role, the human controls, the stage reached and the limitations.
See the industry cases →The governed rollout for a market you already operate at scale: shadow test, capped pilot, then scale with controls and audit.
Explore Enterprise →