Reading · Research
What we do not know yet.
Most of what we have written down is what we are confident enough to act on. This page is the other half: what we still do not know. We publish it because a firm that cannot say where its knowledge ends should not be trusted with your operation. It is written in our working vocabulary, for practitioners and for readers who want to know where the argument stops.
We have tried to obey one rule above all others: not to pretend certainty where there is none.
The Research Agenda · Vol. VI, Preface
Three questions that change what we build.
The full agenda below is written in our working vocabulary. These three are not, and they are the ones that decide what we are willing to claim.
Can we tell whether a system actually made decisions better?
Why it mattersIt is the difference between a system that is used and a system that works. Until it is answered, adoption is not evidence of value.
How it is being investigatedBy instrumenting one engagement against a baseline recorded before the work starts. No engagement carries that instrumentation today, which is why no outcome figure appears anywhere on this site.
Does what we learn in one company transfer to another?
Why it mattersIt decides whether reusable technology is real or a story we tell about ourselves. Client knowledge never travels; the question is whether structure does.
How it is being investigatedBy building the same structures in unrelated operations and recording where they hold and where they break. Three relationships is not enough to answer it.
Where should the boundary between a person and a system sit?
Why it mattersIt decides what may run on its own, and it is the difference between a system people trust and one they route around.
How it is being investigatedBy placing the boundary explicitly in every build, and recording every occasion it has had to move. So far it has moved toward the person more often than away.
Six questions, stated and left standing.
The largest unknowns, stated plainly, with no pretense of a path to an answer. Some may be unanswerable.
Can an organization become self-improving?
Able to redesign its own operating model without outside intervention — and if it can, is that desirable, or does it dissolve the human ends the discipline was built to protect?
Can organizational intelligence be transferred?
From one organization to another — or is it so bound to a particular history that transfer is impossible in principle, not merely difficult in practice?
Can deployment itself be automated?
Could the Method be run by the platform rather than by people — or does something in observation require a human observer, permanently?
Can doctrine emerge algorithmically?
Can the passage from evidence to abstraction be mechanized — or is the judgment of what generalizes, and what is mere coincidence, irreducibly human?
How should organizational memory forget?
Perfect memory may be as pathological as perfect forgetting. There is presumably a discipline of deliberate forgetting, and it has not begun to be built.
What is the limit of augmentation?
Is there a ceiling beyond which more intelligence in the loop yields nothing — or degrades the whole — and if there is, where does it lie?
Every claim in the Canon sits somewhere on this line.
- Settled doctrine
- Held in the Canon; survived recurrence and review. Provisional and demotable even so.
- Working hypothesis
- Falsifiable, with insufficient direct evidence. Where most of the Canon’s empirical claims sit today.
- Open question
- Deliberately unresolved. Closes only when evidence — not fatigue — closes it.
- Experiment
- A deployment designed to discover whether a design survives contact with the operation.
- Finding
- An observation with provenance: where, when, how many instances, what would have falsified it.
- Revision
- Doctrine narrowed, contradicted or retired by evidence. Prior wording is preserved, never deleted.
What we believe, what we have observed, and what reality has earned.
Bedrock keeps a ledger of every claim it makes and the evidence behind it, separate from the record of engagements. It was started in September 2026 by listing every material claim in the six volumes and assessing the evidence each one has. Most of that evidence is thin: the work is young, and no client has yet measured a dollar outcome. That is why the case studies describe what changed in how work gets done, and never quote a number that has not been measured.
118
material claims tracked
56
foundational commitments
27
empirical hypotheses
25
open research questions
0
field-evidence records
Qualifications
- 56
- chosen, not proven
- 27
- awaiting field evidence
- 25
- tracked; the volume itself poses twenty-eight
- 0
- against Canon claims at seeding; ten claims still unassessed
The rules the ledger is kept under
- A Canon citation is provenance for a claim’s existence, never evidence that it is true.
- Three cases inside one workflow are not three independent deployments. Brands under one parent are one organization.
- The reviewer’s task is refutation, not approval.
- Modeled value is analysis, not outcome evidence, and is labeled so wherever it appears.
- Contradictory evidence is never deleted because doctrine was revised.
Organizational intelligence
We named organizational intelligence and built a company around capturing it. We do not, in any deep sense, understand it. Naming located the phenomenon; it did not explain it.
Open questions
- Can it be represented formally, or only enumerated?
- What is its smallest unit — and does it quantize at all?
- By what law does it accumulate?
- By what law does it decay, and can decay halt?
- Is it a measurable quantity, or only an ordering without a scale?
- When, exactly, does an observation become an instance of it worth keeping?
Enterprise cognition
Six faculties were enough to build a platform that works. Enough to build is a low standard of truth. We sort decisions by frequency and consequence, but that is an engineering heuristic, not a law.
Open questions
- What faculties are missing from the model?
- Is there a principled machine/person boundary?
- Which parts of that boundary are permanent, and which move with capability?
- Do organizations reason collectively, with their own structure?
- What new faculties does abundant machine attention create?
- Is the single loop the topology, or only the simplest picture that fit?
Evaluation
Evaluation is the mechanism we trust to keep the discipline honest — and it rests on a capacity we have not established: to say, and to prove, what “better” means. A proxy is not a measure. We are asserting a durability we have not lived long enough to confirm.
Open questions
- Can decision quality be separated from outcome and luck?
- Can learning be measured, distinct from change or drift?
- Can the claim of durable, learned capability be verified?
- Can the surprises a system is not equipped to notice be measured?
- Can intelligence be benchmarked across organizations, or is that a category error?
The platform
We designed the platform to outlive its parts, and drew a boundary between what is permanent and what is replaceable. We drew it by judgment. Whether we drew it correctly is unknown.
Open questions
- Which abstractions survive model change, and which only seem to?
- Is there a correct partition of memory and ontology?
- Do reasoning structures generalize across domains?
- What is worth its modularity cost, and what is premature abstraction?
- Is any architecture invariant under a change in the nature of intelligence?
The discipline itself
The rule of three, the adversarial review, the earned abstraction — these are the conventions by which the discipline makes doctrine. They have held so far. That is not the same as knowing they are right. We may be the least reliable observers of ourselves, and we do not yet know how to correct for it.
Open questions
- Which regularities are lawlike, not coincidental?
- How much evidence is enough, given the stakes?
- What must never become doctrine, and how is it recognized early?
- How is disagreement preserved without paralysis?
- At what tempo should doctrine evolve?
- Can the discipline observe itself without bias?
Observations are many. Doctrine is few.
Most of what we notice on the job never becomes a rule. A rule has to repeat across unrelated organizations, then survive someone trying to knock it down. If new evidence contradicts it, it goes back to being a question. Even published doctrine is provisional.
Closing.
These questions are left open on purpose. We expect to answer some, to fail at others, and to be overturned on a few we thought were settled. We would not have written them down if we were sure. That is the point.
Vol. VI · Closing
If you run an operation where the knowledge sits in a few heads and decisions wait on them, you are where these questions get answered.