The Institute's briefs are written one topic at a time, and each one answers its own question: where does this stand, what would move it, what is pretended about it. This essay asks the question none of them can — what do they say together? It is a synthesis, not new research. Every claim in it generalises over the corpus and is checkable against it: the constraint figures against the Constraint Atlas, the experiment figures against the Decisive Experiments Index, and everything else against the briefs themselves.
The shape of the frontier, by its flags
The corpus carries 37,384 claims under the four-flag discipline: 39 percent established, 37 percent frontier, 17 percent speculative, and 6 percent flagged honestly as handwave — asserted somewhere in the literature, supported nowhere in it. That distribution is the site's one-line answer to "is this stuff real?": about two-fifths of what can be said about the frontier is established science, about a third is live research with real publications and real disagreement, and about a quarter is speculation that earns its place only by being labelled.
The distribution is not uniform, and where it varies is itself a finding. The two most established categories on the site are Energy Systems and Governance & Institutions, both at 53 percent established: these are frontiers limited by deployment rather than by knowledge, and both are areas where the Institute has pushed hardest on the measured record. At the other pole, Civilization-Scale Infrastructure runs 33 percent established — the field where the load calculations are textbook and almost everything above them is extrapolation. The largest share of live, contested claim — frontier and speculative together — sits in Civilization-Scale Infrastructure at 62 percent, which is exactly where a reader should expect the flags to be doing the most work.
What actually blocks the frontier
Every workback plan on this site ends with the constraints that are not research problems, in a closed vocabulary of five kinds. Turned around and counted, those 1003 constraints are the year's clearest single result: 43 percent of them are institutional or regulatory — a licensing regime, a consent framework, an evaluation office, a statute with teeth — against 17 percent that are a scientific result nobody is producing. The popular picture of the frontier as a wall of unsolved science is, on this corpus's evidence, backwards for most topics. The wall is made of institutions that do not exist yet.
Three constraint themes recur across almost the whole map. A first buyer, or capital patient enough to wait, is named 57 times, and it appears in every category the Institute covers — antimatter propulsion and civic technology have, on this axis, the same missing ingredient. Supply or capacity at a scale nobody has built recurs 79 times, in every category. And in all ten the briefs call for an independent referee — verification, audit, adversarial review — some institution whose job is to check the claim that a thing works rather than to make it. That last one is the quiet result of the whole atlas: the frontier asks for umpires more often than it asks for discoveries.
The institutional reading has a limit worth stating, because the 2026 expansion found it. 5 of the 10 categories — Physics & Propulsion, Space & Astronomy, Biology & Human Enhancement, Energy Systems, Civilization-Scale Infrastructure — carry more industrial, supply and market constraints than institutional ones, and the pattern is sharpest in Energy Systems, at 79 against 27. These are what a field looks like when its institutions already exist and its bottlenecks are factories, feedstocks and capital. The institutional majority is real and it is a majority, not a law — the hard-engineering categories are precisely the ones where it does not hold.
The decisive tests are mostly unfunded, not impossible
For every brief, the index records the single result its assessment most turns on. The horizon column is the headline: 43 of those decisive tests are already running — natural experiments and instruments whose data is arriving now — and 21 more are tied to stated windows within a few years. But 214 of 295 are unscheduled: tests the briefs judge runnable that nobody has scheduled or funded. Read against the constraint finding, the two numbers tell one story. The frontier is not mostly waiting on breakthroughs. It is waiting on buyers, referees and appropriations — and, for 73 percent of its decisive experiments, on anyone at all deciding to run them.
This is also where the corpus is most honest about its own genre. The index's extraction rules allow a brief to state that no decisive test exists — that nothing observable currently separates the hypothesis from its absence — and record that rather than inventing one. In this year's extraction, no brief needed to: every one of the 295 names a result it treats as decisive, however remote its horizon. Workback planning for the impossible begins with exactly this distinction: between the untested, which is a funding problem, and the untestable, which is a research programme's job to convert.
What moved this year
The largest single change in the corpus's assessments came not from the world but from the Institute's own discipline, and it came in two movements. The first was a deepening round that rewrote 28 briefs — all of energy, half of governance — from research packs whose contents had been gathered but never written down: at the worst, a published brief carried one source from a pack citing sixty-one. Writing the research down moved those categories' established share to the top of the site, and it surfaced a rule now enforced by machine: a rewrite that adds evidence but drops a speculative or handwave claim is a regression, because the labelled uncertainty is the product. Eight rewrites were caught doing exactly that, and restored.
The second was the widest single expansion the map has had. An external review of where the corpus was silent produced a longlist of candidate topics; 83 of them were commissioned and written this cycle, taking the site from 212 briefs to 295 and the map from 235 slots to 318, while 36 further findings were folded into briefs that already existed. The additions are deliberately unglamorous — screening infrastructure, qualification queues, metrology, repair fleets, chokepoints — because that is where the review found the silence, and because the atlas above is what those topics turn into when you count them. The same non-regression rule governed every one of the 36 expansions: they could add and sharpen, and they could not lose a flagged claim or a cited source.
The evidence base underneath was rebuilt the same way. The citation record now holds 6,374 distinct sources, each with its own page; the rebuild found and unfolded a defect that had merged five unrelated papers into one record, and it publishes its own verification state plainly — 3,902 sources (61.2 percent) have had their links re-checked live, a number this site prints rather than implies. The full release-by-release account, defects included, is on the Assessment Changes page.
What this essay cannot tell you
Everything above is derived from the corpus, which means it inherits the corpus's limits. The briefs are written reviews, current as of their last revision, and the world does not wait for revisions; a decisive experiment recorded here as unscheduled may have been funded since its brief was written, and the fix belongs in the brief, never smuggled into a synthesis. The constraint counts measure what the briefs name, not what exists — a category whose briefs name few constraints may simply name them less systematically. And the flag distribution measures claims, not truth: it says how much of what the corpus asserts is contested, not how much of the contested part will win. Those caveats are not throat-clearing. They are the method — uncertainty is not a problem where it is acknowledged, and the same rule that governs every brief governs this page.