An invitation to
build together.

02 / A hypothetical example

What did the action change?

Imagine a robot picking up a banana.

It can adjust its grip using pressure or slip feedback and eventually place the banana in the required location.

But “the banana was successfully moved” and “the banana was not damaged” are not the same judgement.

If the record tells us only that the object reached its destination, we still do not know what changed during contact, which of those changes were observable, or whether they entered the conditions of the next grasp.

The question is therefore not only whether an action can be adjusted. It is also:

• What did the action change?
• On what basis can the system distinguish that change?
• Does the next action begin from the changed conditions?

The same problem appears in batteries. A charge or discharge task may be completed and the delivered result recorded, while the record may or may not preserve how individual members responded, what conditions they experienced, or which processes were never measured.

We cannot infer that damage occurred simply because the record is incomplete. But neither can we treat “no change was recorded” as equivalent to “nothing changed.”

When a later judgement depends on that distinction, making the distinction observable is not an optional annotation. It becomes part of building the capacity for the next judgement.

03 / A correction in our own work

More indicators.
More independent evidence?

In an analysis of public battery-test records from NASA PCoE, we examined several impedance indicators that changed over experimental history.

It is easy to look at several changing indicators and feel that multiple pieces of evidence are converging on the same conclusion.

When we traced their origins more carefully, however, the question changed. These outputs were derived from the same underlying impedance-test records. Independence between their measurement and calculation chains had not been established. Different indicators and agreeing curves therefore could not automatically be counted as multiple independent pieces of physical evidence.

We revised our own interpretation.

We retained the changes that the records supported, including differences in their magnitude, but stopped treating the number of indicators as a count of independent evidence.

This does not mean that the indicators are useless, or that the observed changes are false. Different expressions of the same underlying records may still contain complementary information. What changed was the evidential weight we were entitled to give them.

This was a correction to LFR’s own interpretation. It was not an allegation that the data provider had made an error, and we did not use it to diagnose battery damage.

It made a simple question concrete:

When a conclusion appears to have gained more support, have we gained another observation — or another expression of an observation we already had?

Answering that question further would require additional information about the measurement chain, examination of possible shared influences, and experiments capable of distinguishing competing explanations.

Drawing more curves cannot substitute for that work.

04 / Different places to look

One inquiry.
Different physical settings.

These areas are at different stages. They do not yet constitute a single product or unified method.

We are interested in discovering capacities that can genuinely be reused across different objects, while preserving the differences that cannot.

This includes analysis carried out jointly by people and AI. How a question was framed, why particular evidence was selected, and why a method was considered applicable should remain open to examination alongside the final answer.

05 / A possible starting point

Begin with something
we can actually work on.

We would like to begin with a concrete problem and develop it together into a piece of work that can be inspected, challenged and continued by someone else.

One possible starting point is:

Choose a real process in which an important distinction needs to be made, then build a small observational or analytical prototype to test what helps us understand or judge it: making better use of existing observations, seeking additional evidence, or revising the representation or task.

A collaboration could begin with two situations that appear similar in the available records, and a question about whether their differences matter for a particular task. Together, we would examine the objective, the consequences being considered, and what remains outside scope—not assume in advance what an appropriate response should be.

Where existing observations are insufficient, a bounded experiment might help. Two objects with similar static readings, for example, might respond differently to a carefully chosen interaction, where that interaction is appropriate and authorised. Changing when or how we observe may help distinguish explanations that the existing records leave unresolved.

The result might support the current approach, justify a change, or remain inconclusive. We would retain what became distinguishable, which interpretations need revising, and which conditions remain unresolved, so that the next participant can understand why the work continues as it does—and can question that direction.

That work might involve:

• defining what needs to be distinguished, and why that distinction matters in practice;
• examining what existing records actually preserve and, where necessary, designing a limited additional observation or experiment;
• building a working and modifiable tool that retains the necessary history, member identity and temporal relationships;
• asking another participant to retrace the path from input through processing to output and, where justified, testing which parts remain usable in another context.

Intermediate outputs might include prototypes, experimental records, methodological notes and explicit boundaries of use.

They may also include hypotheses that did not survive testing and designs that had to change because of what was observed.

The purpose is not only to find errors.

We also want to make previously invisible distinctions observable, allow new questions to become possible, and turn experiments or tools that previously lacked sufficient grounds into things that can actually be built and tested.

06 / Ways to take part

Bring more than agreement.

Participants do not bring only resources. They may introduce distinctions that LFR has not previously noticed, changing both the question we ask and the method we thought we would use.

Specific roles, project periods, funding amounts and compensation arrangements have not yet been fixed.

Employment, project collaboration, research support and equity investment create different relationships, responsibilities and expectations. They should be discussed separately rather than being blurred together under the phrase “supporting LFR.”

07 / Working together

Leave room for
the work to change our minds.

We do not ask participants to begin by believing that LFR’s interpretation is correct.

We would instead agree to preserve enough of the work for it to be inspected, take disagreement seriously, and allow new evidence to change a judgement.

When a result does not match an expectation, we should be able to say which step no longer holds, which outputs are affected, and whether there is still a justified path forward.

Not every failure should be renamed as success.

But neither should a rejected hypothesis erase findings that remain supported.

Before work begins, responsibilities, resources, compensation or expected returns, access and disclosure conditions, use of resulting work, and conditions for revision or ending the collaboration should be made explicit.

Useful capabilities may develop into commercial applications and may also help sustain further research.

Commercial expectations, however, cannot determine in advance what an observation must find.