Mine did too. So I built the layer I run everything on.
the decision learning layer
It watches the outside world, holds what each company knows inside, keeps the evidence behind every call, and learns from what happened next.
I built it because I got tired of teaching software the same thing twice. Everything I work out running it ends up here.
The model is rented. The record is not.
RSLVD is where I show the work. Ghost is what the work runs on.
Companies already have the tools. What they do not have is the connective intelligence between them. Ghost holds what a company knows, what its market shows, what its people decide, and what happened next, connected and dated, as one record.
Models reason across it. People gate anything consequential. Outcomes flow back in, so the next decision inherits what the last one taught.
The record gives it structure. The models provide the reasoning. The loops are what make it compound.
holds the company, the market, the evidence, and how all of it relates, as a connected graph
read, reason, compare, explain, and make the work
approves, corrects, and supplies what only somebody in the business knows
return the action and the outcome to the fabric
Ghost does not have a public home yet. It is running with a small group of partners across very different markets before we open it wider.
Model-agnostic on purpose. Claude runs here, OpenAI models run here, and others are being graded against the work rather than argued about in the abstract.
No. A database keeps the last value. This keeps the decision.
What was believed, the evidence for it, who agreed or overruled it and why, what was done, and what happened after. Dated, so you can ask what was believed in March rather than only what is believed now. The reasoning on top is swappable. The record underneath is yours.
Can it see a fragmented market?
Thousands of organizations spread across registration systems, governing bodies, social accounts and local directories that rarely agree with each other. Ghost has to work out who exists, what just changed, and where a buying window is opening.
A weak signal here wastes a call.
Can it tell momentum from evidence?
A loud market where being early is not the same as being right. Ghost has to score what is spreading separately from what is supported, and know which claims it must never repeat no matter how well they are traveling.
A weak signal here becomes a bad claim.
Can it preserve expert judgment?
Every answer grounded in the record before an attorney reads it. When one of them changes it, Ghost keeps the change and the reason, so the next piece of work inherits what the last one taught.
A weak signal here gets caught, and the catch is the asset.
Can it learn from the physical world?
Barrels lose product to leakage and nobody knows until the loss is already counted. A sensor on the barrel reads that in real time, off a signal no public source has. Ghost has to treat a hardware stream as evidence, with the same provenance and grading as anything it reads off a page.
A weak signal here is measured, not argued.
These are not four markets I am selling software into. They are four very different places I am teaching the same layer what it means to learn.
Two live environments. One question: does the next decision get better because of what Ghost kept from the last one?
Every consequential call carries the evidence behind it.
source + date + methodNothing consequential moves without a person.
enforced in codeOne company record never enters another.
a wall, not a settingwhat moved
evidence and contradictions
confidence and relevance
what matters now
a person, when it counts
the work goes out
correction and outcome
The numbers still forming are the ones that matter. They are the only ones that can show whether Ghost is actually learning.
how often the operator takes the call as delivered
whether Ghost does better after it has been corrected
how much of what we acted on eventually gets a result attached
web, public data, research, news, and market signal
what appeared, closed, changed hands, or changed meaning
source, method, date observed, date true, confidence, contradictions
every important output graded against a written standard
different models put against the same work
consequential action stops where judgment still matters
an override is stored as evidence, not discarded
what happened comes back to the recommendation that caused it
We are not optimizing for how much Ghost can read. We are measuring how much it can learn.
I stopped optimizing for a smarter model and started keeping the receipts. That is the bet.
We are all renting increasingly capable intelligence, and that gain arrives for everyone. The private record of your evidence, your decisions, your corrections and your outcomes is the part that compounds for you alone.
The model is rented. The record is not.
Where I show the work. The field notes, the ventures, and the parts I got wrong in public.
What the work runs on. The record underneath every venture, and the reason the next decision starts from something.
Almost everything I take on arrives unresolved. A market I do not understand yet, a signal I do not trust, a decision nobody can quite explain. Ghost is how those get resolved without losing what they taught. This is where that gets written down.
Share what you are building, and we can see if there is an opportunity to discuss further.
dave@rslvd.ai