Hesia

Hesia · Method

How a file is made, and why you should be able to argue with it.

We publish critical analysis of real companies. That is only worth reading if you can see how we got there — so the scoring rule is on this page, including the arithmetic. If you think a file is wrong, this is the page that tells you where to push.

What we commit to

A person writes it

Not generated. We use models the way an analyst uses a search engine — to find things faster — and a human decides what the file says and signs their name to it.

The subject reads it first

Every company we write about gets the draft and 5 working days before anything is published. No exceptions, including for companies we are critical of. Especially those. The signed record sits at the foot of each file, with the window counted from the dates rather than restated — if a subject got less than we promise, the file says so.

Their reply is printed next to ours

Where a company disagrees, their answer appears in the file itself — same page, same weight, not a footnote. We do not get the last word by default.

We are never paid on the outcome

No success fee, no percentage of a raise, no introduction commission. The moment our income depends on a deal closing, our analysis of that deal is worth nothing to the person reading it.

The shape of every file

  1. 01

    Diagnose

    Decompose the decision into the claims it actually rests on. Not a summary of the deck — the list of things that would each have to be true.

  2. 02

    Attribute

    Attach every piece of evidence to the specific claim it bears on, score the source, and mark whether it supports that claim or cuts against it.

  3. 03

    Price the gap

    What is left unresolved, what it would cost to resolve, how long it would take, and how much it would move the picture if you did.

How a source is scored

Every piece of evidence in a file carries a credibility score from 1 to 5, and where it came from sets the ceiling. A founder’s deck and an audited statement are not the same object and must never be able to score the same, however confidently the deck is written.

  1. 5/5AuditedSigned off by someone with a licence to lose.
  2. 4/5Third-party dataA source with no stake in the outcome.
  3. 4/5Customer referenceSomeone who actually paid, reached without the founder in the room.
  4. 4/5Public filingOn the record, with a penalty for being wrong.
  5. 3/5Management callFirst-hand and specific, but from an interested party.
  6. 2/5Founder suppliedA claim, not a fact. Capped — no deck is ever worth more than this.

A source can always score below its ceiling — a vague management call is worth less than a specific one. It can never score above it.

How the scores combine

An assumption usually has several artefacts behind it. They do not get averaged — averaging lets a pile of weak material outvote one strong document, which is backwards. The rule is how a diligence analyst actually reads a file:

The best source sets the ceiling

One audited statement is worth more than five founder slides. The strongest supporting artefact decides how high the assumption can go.

Breadth adds at most one point

Corroboration counts, with heavy diminishing returns — so restating the same claim from the same mouth earns nothing.

Contradiction discounts everything

A contradicting artefact does not merely fail to help; it drags the whole stack down in proportion to how credible the contradicting source is.

Concretely: a mixed reading counts for half of its source’s credibility, breadth contributes 0.15 per additional supporting point up to a cap of one, and each contradicting artefact removes 16% of the total per point of its own credibility. An audited document pointing the other way takes the assumption to zero. That is the intended behaviour, not an edge case.

What the verdict is about

The headline on a file describes the file, never the investment. We can defend a statement about whether a set of documents is sufficient to price a round. We could not defend telling you to pass.

Under 40
Not enough on file to price the roundNot ready
40 – 69
Priceable, with named conditions outstandingProvisional
70 and over
The file supports the askReady

The number is a count of what is on file out of 100 — not a probability that the investment works. Nobody can calculate that, and a figure that looks like one would be the most misleading thing on the page.

Still being written

How we choose which companies to write about, and how a claim gets decomposed in the first place, is being written down properly rather than reconstructed after the fact.

It will be published here when it is finished. We would rather leave a gap on this page than fill it with something that sounds like a method and isn’t one — which is, more or less, the entire product.

What we are not

  • We are not licensed advisers, and nothing we publish is investment advice or a recommendation to buy or sell anything.
  • We do not arrange investment, introduce for a fee, or take any payment contingent on a financing outcome.
  • We do not publish an analysis of a company that has not read it and agreed to publication.
  • We do not claim a file is complete. Every file states what is still unresolved, because that is the point of it.

What we count while you read this

Five things, to our own server, with no cookie and no advertising tag. Nothing third-party loads while you’re reading a file. The list is enforced in code on both ends — an event not named below cannot be sent by the page and is rejected by the endpoint — which is the only reason it is worth printing here. (The one exception on the whole site: signing in and connecting a data source under Settings — an action you take, not one anything here does to you — opens Nango’s own hosted connect screen.)

  • file_openedA file page was opened, whether it was behind the wall for you, and how much of it your screen already showed.
  • preview_depthHow far down you chose to go, rounded to a quarter, and only counted once you scrolled — arriving on a short page is not reading it. Four numbers, each sent at most once, never a scroll trace.
  • gate_shownYou reached the point where a file stops and asks for an email.
  • gate_passedYou gave one and the rest of the file opened.
  • ask_submittedAn enquiry was sent from /ask. The event records that one was sent, not what it said.

Not recorded: your IP address, your browser, where you came from, anything you type into a form, and any identifier that outlives the tab. The key that ties two of these events together is random, made in your browser, and gone when you close the tab — so a visit next week is a stranger to us. That costs us the ability to measure returning readers, and we would rather lose that than ask you to trust a paragraph about privacy on a site whose argument is that you shouldn’t have to.

If a file gets something wrong, tell us. Corrections are published in the file, dated, with what changed.

Read a file →