Meet something
you would not pick.
Use a real random draw from a declared population, an anomaly or an unfamiliar source. Preserve the actual inputs before interpreting them.
innoMOS / Research-informed invention with AI
What if
the question
is the idea?
Unlikely combinations. Childlike questions.
Borrowed mechanisms. Possibilities made
concrete enough to test.
01 / Wonder — before the answer
Keep the child's
permission to ask.
Why does it have to work that way? What if those things belonged together? Could the strange exception be the important part?
innoMOS is an AI invention method, delivered through a plugin. It treats finding a question as creative work in its own right. It helps you explore problems, anomalies and unexpected encounters — then turn a promising relationship into something you can actually investigate.
The ambition is more routes into invention, and a clearer path from curiosity to a useful test.
Models can generate promising ideas and still converge on familiar patterns or overlook a condition. innoMOS changes the starting material, representation, comparison and test around the model. It does not change the model’s weights. The research and its limits sit beside the test below.
02 / Explore — change the starting point
Same curiosity.
A different way in.
Choose one starting question. We’ll carry it through a relationship, a test and your own brief.
Change the representation.
What if a book could carry a memory of being read?
A book becomes a changing record, rather than only a container of content.
First ask what can be recorded, with consent, and what that memory lets someone do. A metaphor alone is not a mechanism.
These are authored questions. Choose one to follow through the page. The same question, its conditions and its limits will travel into your starting brief. No AI generation runs here.
Go deeper / The maths of an encounterHow can a few pieces open so many starting points?
A small change in inputs.
A much larger space.
Choose distinct pieces from a defined pool.
Possible groups grow quickly.
A mathematical illustration, not a live draw.
Uniform sampling gives each group an equal chance. Curating the population is still a human design choice.
Possible unordered groups2,024
C(24, 3) = 24! / (3! × 21!)C(n, k) = n! / [k! (n − k)!]
14.95
Expected distinct groups after 15 independent uniform draws, allowing repeats between draws.
Where the second number comes from
E[U] = N × [1 − (1 − 1/N)m], where N is the number of possible groups and m the number of independent draws. Groups can repeat between draws.
The calculation opens a space.
It does not count inventions.
Different groups can lead to the same familiar idea. A random input is not automatically a useful relationship. The creative work is to find what the pieces could do together, under which conditions, and what would show that the connection fails.
C(n, k) = n! / [k! (n − k)!]There are 2,024 unordered three-piece groups in a pool of 24 distinct pieces. A pool of 100 has 75,287,520 five-piece groups. These are counts of input sets, not original ideas.
E[Uₘ] = N [1 − (1 − 1/N)ᵐ]With N possible groups and m independent uniform draws, this is the expected number of distinct groups observed. For 15 draws from 220 groups it is about 14.53. Repeats are possible even with a sound random generator.
03 / Equations that have a job
Next / ConnectA stranger starting point is only a beginning. What could its relationship actually let you do?
03 / Connect — give the relation a job
Make the leap.
Keep a way back.
The relationship you are carrying
A book becomes a changing record, rather than only a container of content.
Meet something
Change what
the problem is.
Abstract the situation. Borrow a relationship from another context. Explore a different use for an existing function. The Ortho route provides one structured way to make that transfer.
Give the relation
a concrete job.
What action or inference becomes possible? Compare the proposal and its nearest known counterpart on the same valid input, with necessary properties stated.
Let reality
answer back.
Remove or replace the claimed coupling. Try a calculation, executable test or appropriate experiment. Keep the question and negative result when the mechanism does not hold.
Discovery preserves three outcomes through Explore → Realize. A mechanism, an open question and a failed match are legitimate results.
Next / TestKeep the relationship you chose. Now take away a condition it needs, and see which claim survives.
04 / Test — discover what holds
A possibility is allowed.
An explanation has to earn it.
What if a book could carry a memory of being read?
The familiar counterpart
An ordinary bookmark or reading-history feature may already preserve a stopping point. What extra action would the proposed relationship allow?
A necessary conditionA consenting reader supplies an identifiable, timestamped reading record.
With the declared condition
A recorded stopping point could be recovered. That alone is familiar functionality; the question still needs a more specific difference.
First ask what can be recorded, with consent, and what that memory lets someone do. A metaphor alone is not a mechanism.
An authored exploration you can operate. The control changes what may be inferred; it does not run an experiment or establish a new invention.
f_proposal(x) ≠ f_known(x)A test criterion, not a law of creativity: on the same valid input x, does the proposed relation enable a different relevant action or inference than the nearest known functional counterpart?
Then remove or replace the claimed coupling. If the difference survives unchanged, the proposed explanation may be wrong. A difference from a weak comparison does not establish a difference from the best known solution. None of these equations proves originality.
An exciting proposal and a working result are different things. Creativity research gives reasons to vary the search; computational discovery shows the value of an external evaluator. Neither establishes a benefit for this method by itself.
Go deeper / Research & limitsWhy explore widely, then test the explanation?
Five primary sources, with their boundaries.
The studies motivate parts of this approach. They do not establish that innoMOS as a whole outperforms a strong direct prompt.
Individual creativity and collective diversity
Doshi & Hauser — Generative AI enhances individual creativity but reduces the collective diversity of novel content.
In a short-story experiment, access to AI ideas improved evaluated individual stories while increasing similarity across AI-assisted stories.
- Design implication
- Keep different routes into the problem, rather than equating one polished answer with a varied search.
- Boundary
- A short-story result is not a universal result about invention, current models or innoMOS.
Ideas can be judged novel before they are built
Si, Yang & Hashimoto — Can LLMs Generate Novel Research Ideas? A Large-Scale Human Study with 100+ NLP Researchers.
The study found higher judged novelty for LLM-generated proposals, alongside feasibility and generation-diversity limitations.
- Design implication
- Take model-generated possibilities seriously. Preserve the conditions that would make them work.
- Boundary
- Reviewers' novelty judgements are not implemented discoveries or exhaustive prior-art checks.
A proposal's appeal can change after execution
Si, Hashimoto & Yang — The Ideation-Execution Gap: Execution Outcomes of LLM-Generated versus Human Research Ideas.
Experts implemented assigned research ideas. Earlier advantages for LLM proposals weakened after execution.
- Design implication
- Separate the exciting proposal from the result of a discriminating implementation or experiment.
- Boundary
- These tasks and implementations do not settle every domain or establish an innoMOS benefit.
Mathematical discovery needs an evaluator
Romera-Paredes and colleagues — Mathematical discoveries from program search with large language models.
FunSearch coupled model-generated programs with systematic evaluation and search, producing cap-set constructions and bin-packing heuristics.
- Design implication
- Use an external check that can distinguish a better proposal from a persuasive one.
- Boundary
- This is a particular evaluated search system, not a general theorem-solving guarantee. innoMOS is not FunSearch.
Look for a landscape of useful possibilities
Mouret & Clune — Illuminating search spaces by mapping elites.
MAP-Elites preserves high-performing solutions across predefined behavioural dimensions, rather than returning only one optimum.
- Design implication
- Diversity and intermediate stepping stones can be useful search objectives.
- Boundary
- innoMOS does not implement MAP-Elites by borrowing this idea; a curated operation catalogue is a different mechanism.
The project evidence, separately
Research-informed.
Still being tested.
What exists in the product
The inspected Discovery implementation offers problem, encounter, anomaly and exaptation entry profiles. It uses curated, source-linked operation hypotheses, a three-outcome Explore → Realize flow, and retention checks. Random and Ortho routes address different parts of exploration.
The host model supplies the creative proposals. Structural acceptance is not creative-quality acceptance.
What our pilots establish
Our bounded development pilots include actual paired runs and functional checks. They do not yet establish general superiority over a strong direct prompt, verified originality, first-mover status or human benefit. A three-pair core.4/core.5 pilot ended INCONCLUSIVE.
innoMOS 0.4.0 remains a preserved reference. A newer version's additional workflow features do not, by themselves, make it a better invention method.
Project basis: the 20-part discovery research synthesis (17 October 2026); inspected Discovery source; NEXT3 paired-pilot report (26 October 2026). The research loops are literature work, not 20 method-effect experiments. Source inspection here used paper abstracts, publisher excerpts and author project material; it was not a full-text systematic review.
Next / BeginA condition can fail without making the question worthless. Take the question, its comparison and its limits into the next exploration.
05 / Begin — take the question further
You have a thread.
Where will you take it?
A useful question can arrive before a mechanism. An old function can find a new purpose. A failed connection can leave the clue that matters later.
innoMOS gives those routes a deliberate place in the workflow — with the freedom to explore and the discipline to discover what holds.
The question you are taking with you
What if a book could carry a memory of being read?
Make room for your own question.
Use the thread above or begin with something entirely different. Edit the facts and conditions before preparing your brief. This page does not run AI; it gives your next chat a clear starting point.
Optional / Working examplesSee how the principles behave.
These digital examples make two principles tangible: change the available evidence, or change the starting ingredients. They illustrate parts of the practice; they do not measure its creative advantage.
Open the separate typography studio ↗Open the presentation / handout ↗Example / A claim depends on its evidence
Keep the endpoints.
Change what you know.
Now remove a necessary condition.
What survives? What becomes an open question?
A disappearing moment Fixed local example
Open in the morning.
Open again later.
What happened between?
The endpoints look identical. Could you still know what happened in between?
Northlight workshopSimulated · subject W02
The relationship makes this knowable
Yes. A cancellation happened.
The supplied event log contains v2. The workshop is currently open at v3; the cancellation is historical.
Current status stays open · v33 events available
Your counterfactual
Keep a starting point. Change one condition. Compare what it permits.
This example runs from a fixed, local dataset. It demonstrates evidence and inference; it does not generate an AI answer. A missing event record leaves history unknown.
Change the frame.
An idea becomes a relationship with conditions.
Keep a reference.
Compare a new possibility with what you actually started with.
Try to break it.
Remove what matters. See whether the function survives.
Example / An unexpected starting combination
Take an
unlikely turn.
Three things you might never have put together.
What relationship could make them useful?
This browser draws three distinct building blocks from a curated set of twelve using Web Crypto. Continue in your chat to explore the connection.
SVG paths
A path can represent a relationship with a measurable position along it.
B01Versioned snapshots
A saved state can carry a version without reconstructing missing history.
B02Pointer capture
A drag can keep receiving pointer events after leaving its original target.
B03