Most sales tools answer every question. Ours answers the ones it read an answer to, and says so about the rest. This page is the long version of that: where a fact comes from, what happens to one that fails a check, and where the software is least reliable.
Two kinds of fact, kept in different tables
A fair's dates, its exhibitor list, what each company sells, who is speaking on the second morning. All of that is the same for every customer, and expensive to establish, so Pengo establishes it once and shares it. How well any of it fits your buyer is a different kind of statement. It belongs to one account and never reaches another.
The split is structural rather than a promise. Shared facts live in catalog tables that carry no account id and no score column at all, so there is nowhere for one customer's opinion to be written into another customer's directory. A score in a shared table would be one seller's judgment served to their competitor as a fact.
Every fact carries the page it came from
Pengo reads published pages at the time you press the button: the fair's own exhibitor directory, the organiser's programme, and the companies' own sites. A booth row carries the hall and stand number as the directory printed them. A person carries the page their name was read on.
Four things it will not invent
- An email address
- A person is kept with an address only when the research states it read one. Asked for a name and a working address in one breath, a model answers the second by pattern, usually
firstname.lastname@at whatever domain it saw last. A guess is indistinguishable from a finding to whoever sends to it, so the guess is dropped. - A score with no reason under it
- Every fit score quotes the sentence it was decided on. A row Pengo cannot justify is dropped rather than filled in with a default rationale or a zero, and the number of rows dropped is reported on screen instead of quietly rounded away.
- A LinkedIn profile built from a name
- Most people arrive off an agenda that named them and linked nothing. Pengo will look a profile up, but it keeps the address only when the handle begins with the person's full name and a profile actually turned up at it. A wrong profile link is worse than an absent one, because somebody opens it and addresses a stranger.
- A stand count nobody published
- A fair that states how many exhibitors it has is priced against that number. A fair that states nothing is sold at a rate per company and charged against the list when the list arrives. The alternative is an invented total on the screen where somebody decides what to spend.
What complete means for a directory
Exhibitor portals paginate, and the failure they cause is specific: a pass answers page one, comes back with eighteen booths of a stated 2,200, fills no cap, and reads exactly like a small fair. Eighteen booths is a stable, plausible, completely wrong answer.
Three signals have to agree before a directory is stamped complete. The pass reports the pages it saw and did not read, so a continuation starts at page two. The fair's own claim about its size is checked against what came back. And the run answers, in a required field, whether it reached the last entry, because a response that did not run out of room proves nothing about whether it stopped in the H's. Silence on that field is read as unfinished, never as done.
A continuation is priced as a continuation. It inherits the first pass's crawl and carries the names already held, so it is quoted at the remainder rather than at a fresh read of page one.
When a pass finds nothing
It settles as failed and releases the credits in full. A pass that established nothing has nothing to sell you, and a completed run reporting zero rows against a charge is the single outcome that makes a research product impossible to trust. The same rule covers a website read: a 404, a timeout or a PDF costs no tokens, so it costs no credits.
Where it is weakest
Four places, and they are the ones worth knowing before you plan a trip around the output.
- 1Session times are held as the programme printed them. A fair publishes in its own local zone and usually does not say which, so Pengo does not convert. Read the agenda tab as the venue's clock.
- 2A company with a thin website gets a short profile or none. Those rows keep a stamp saying Pengo looked, rather than going blank as though nobody had, and they are held back from scoring. A company name with nothing behind it cannot be judged, and judging it anyway is how a scoring pass returns two embassies and a drone company.
- 3A listing that a clean pass no longer finds is retired rather than deleted, and only when the whole pass finished cleanly. A vendor's bad afternoon must never read as a fair that emptied out, and your notes on a stand survive a refresh that lost it.
- 4Scores answer the description of your buyer that you had when they were written. Edit that description and every score in the account is answering a question nobody is asking any more. Pengo stamps each score with the description it was an answer to and tells you they are stale, rather than letting them look current.
How to check any of this
Open a booth and read the rationale under the score, then open the company page it links to. The sentence Pengo judged on is on screen, next to the page it was read from, and if it is wrong the row is wrong in a way you can see in a few seconds. Everything above is written so that a claim we make here is one you can disprove there.
The rest of the argument is on what Pengo does with a fair, and what each pass costs is on pricing.