# A MAP OF UNKNOWNS *Why this folder contains generators and not just more hypotheses* The five expeditions produced 1,500 hypotheses. Almost all of them were **claims** — things that might be true. That is one kind of output, and it has a ceiling: every claim you can state is, by the act of stating it, already inside the space of things you can imagine. This folder tries something else. It collects **operations**, **questions**, **candidates for the invisible**, and **inventions** — the machinery that produces claims rather than the claims. --- ## The taxonomy Ignorance is not one thing. Fourteen kinds are distinguishable, and they fail differently. **1. Unmeasured.** Measurable with existing instruments, simply not measured. *Detection:* look for quantities that are always inferred from something else. *Fix:* measure it directly. Most of the v1–v5 "existing-data" entries live here. **2. Unrewarded.** Measurable and uninteresting to whoever would have to do it — no paper, no product, no grant. *Detection:* ask what measurement would settle a dispute but produce a boring result either way. *Fix:* framing it as a methods result. This is where Program F of Expedition V lived. **3. Resolution-walled.** Below an instrument's reach, for a reason that is architectural rather than temporary. *Detection:* the quantity scales with a parameter you cannot change — the electron wavelength, the diffraction limit, the thermal noise floor. *Fix:* change the *observable*, not the instrument. (The NV rosette in Expedition IV is an example: the split between four orientations gives a tensor without improving the sensor.) **4. Conceptually unavailable.** The phenomenon is measured but has no name, so it is reported as noise, as part of another quantity, or not at all. *Detection:* look for things that are reproducible and unnamed. *Fix:* name it, then look for its scaling law. This is the most under-exploited category in the collection. **5. Assumption-hidden.** The thing is invisible because an assumption is in the methods section and not in the results. *Detection:* read the methods of a field you do not belong to and ask which step is a choice. *Fix:* vary the choice. (Expedition V Program C is entirely this.) **6. Boundary-invisible.** Falls between two fields, so no journal, no reviewer and no grant panel owns it. *Detection:* the two halves are each well studied and their interface is not. *Fix:* a review paper that is actually a research programme. **7. Unit-incommensurable.** Two fields measure "the same" quantity in units that differ by a factor that everybody absorbs silently — a per-mass versus per-volume, a per-atom versus per-mole, a full-width versus half-width. *Detection:* compare two standard values that should agree and do not. *Fix:* dimension analysis across the boundary. Some of the best findings in the corpus came from this (e.g. the 4πG versus 2πG coefficient). **8. Averaged away.** The mean is reported; the mechanism is in the variance, the skew, or the tails. *Detection:* anything reported as a single number with a standard deviation. *Fix:* report the distribution. Expedition IV's single-cell batch is largely this. **9. Subtracted as background.** The quantity is measured, then removed by an automated correction, and never inspected. *Detection:* every pipeline has a step labelled "background subtraction". *Fix:* run it backwards and look at what was removed. **10. Rare but real.** Real, large when it happens, and too infrequent to study. *Detection:* tail events that are individually dismissed. *Fix:* stack them, or find a proxy that fires more often. **11. Forbidden by prior.** Would contradict a load-bearing assumption, so it is not looked for — not because anyone is suppressing it, but because the search would look incompetent. *Detection:* ask what a field would do with a positive result. *Fix:* design the test so that a null is also publishable. **12. Systematically non-reproducible.** Either noise or real-but-uncontrolled; the field has stopped asking which. *Detection:* a well-known effect everyone privately doubts. *Fix:* a pre-registered multi-lab trial — expensive, and the only honest route. **13. Unfalsifiable-in-practice.** Testable in principle only on a timescale or budget nobody will spend. *Detection:* "it would take a thousand years". *Fix:* find an analogue system or a shorter proxy. Expedition III's mantle entries struggled here. **14. Genuinely unknown.** No instrument, no concept, no handle. **This category cannot be enumerated.** Anything listed under it would, by the act of listing, cease to be in it. --- ## The one honest handle on category 14 You cannot list unknown unknowns. You can, however, identify **places where the machinery of knowing is structurally blind** — and those are the only places where a category-14 item could be hiding in plain sight. Five such places recur: - **Where a correction is applied automatically.** Whatever the correction removes is, by construction, not in the data anybody looks at. - **Where two conventions meet.** Any factor silently absorbed at a boundary is invisible on both sides. - **Where a measurement is a proxy for a mechanism.** The proxy's own behaviour becomes the thing studied. - **Where an effect is defined as an error.** Noise, drift, artefact, leakage — each is a label that ends inquiry. - **Where the incentive is absent.** No career, no product, no paper. Notice that all five are *properties of the knowledge machine*, not of nature. That is the useful consequence: the unknown unknown, whatever it is, is more likely to be sitting behind a structural blind spot than in an exotic physical regime. --- ## What this folder therefore contains - **Generators** — operations that take a problem and produce an idea you would not have reached by staring at the problem. These are the most transferable output. - **Inventions** — things with a mechanism and a first test, where the obstacle is technical. - **Discoveries** — questions whose blocker is identified, not merely their subject. - **Unknown territory** — twenty candidates for the invisible, each with the reason it is probably nothing. - **Unasked questions** — questions whose absence has a structural cause. --- ## What it deliberately does not contain - **Another 300 hypotheses.** The five expeditions already did that, and a sixth would add volume rather than reach. - **Claims that the unknown has been mapped.** The map is of *how* things go missing, not of *what* is missing. - **Novelty claims.** No literature search backs anything here. A reader who finds one of these already published has found a defect in the folder, and that is the expected outcome for some fraction of it. --- ## How to use it If you have a problem: read `01_GENERATORS` first and apply three operations to it. If you have a laboratory: read `02_INVENTIONS` and pick the one whose first test is cheapest. If you have a dataset: read `05_UNASKED_QUESTIONS`. If you want to know where to look for something nobody has seen: read `04_UNKNOWN_TERRITORY`, and read its honest-status lines first.