What this doesn’t do yet
Open work, not a roadmap and not a request. None of it is built. It is here so the limits of this contribution are legible, and so that if any of it is worth doing the case and the cost are already stated.
Four of the nine come from this augmentation failing its own audit rather than from wishing it were bigger. That is the useful kind of future work, and it is why the two lists below are separated: the first three would add something, the rest mostly fix something. Ordered by what would add the most, not by what would be easiest.
The three worth doing first
1. Critical paths across the whole map, not two
There are two chains here, built by hand, against 103 gaps. They exist to show something a catalogue structurally cannot — that two gaps in different fields can share one binding link, so bottlenecks recur across fields and can be counted. Two chains prove the shape exists. They say nothing about where else it occurs.
One thing blocks it. Capability edges are untyped. Nothing in the source data marks a capability as necessary, sufficient, or partial for the gap it hangs under, and a chain needs to know which links are load-bearing. Both existing chains had their link semantics reconstructed by hand, which is exactly why there are two.
Type the 389 capability edges. One pass, one column, no schema redesign — and chains stop being illustrations somebody authored and start being something the data yields.
2. A 5-10 year value, and splitting the two questions maturity conflates
Maturity has three values and the middle one is doing most of the work: 62 of 103 gaps sit in 2-5 years. A bucket holding 60% of the map is barely a label.
And the three values are not on one axis. “Speculative” means no clear path from here, which is a claim about kind. “2-5 years” is a claim about time. A gap with a perfectly clear path that simply takes fifteen years has nowhere to go, and lands in the middle bucket next to things that are nearly here.
Not fixed here, and the reason is not caution. Adding a fourth value means re-reviewing every gap already in the middle bucket, because a value nobody has applied to the whole set is worse than three honest ones — those rows would silently mean “2-5 or 5-10, unexamined”. That is a full labelling pass with a blind second reader.
Probably two fields rather than one more bucket: how long, and whether a path is known. Different questions, currently answered at once.
3. Blocked on adoption, or blocked on capability
Repairing the maturity dimension turned up eleven gaps where the reason the label is not “working now” has nothing to do with whether the technique exists. DNA-synthesis screening works and is not adopted. Web-scale archiving works and permission is withheld. Adaptive platform trials work, have run since 2010, and the field has not taken them up.
Maturity absorbs all of that and reports it as not ready — and not ready sends a funder toward more research. For an unadopted capability that is the wrong intervention. The answer is procurement, a standard, a mandate, or somebody paying for a thing nobody is billed for.
One flag distinguishing the two would change what the map recommends for a fifth of its rows. It is the only item on this page that would add to the map rather than correct this contribution, and it needs its own definition, pass and audit.
The rest
Smaller, or already written up elsewhere, or larger than an attribute and properly theirs to decide.
Indicators as coverage, and as a series
A sample of 8 gaps across all four tiers, one of them honest nulls, explicitly not to be extrapolated to the other 95.
Coverage is the smaller half. A one-time reading says where a gap is; a funder needs the slope, and a slope needs the same quantity read off the same source on a schedule. That is a maintained system rather than a research output.
Three measurability tiers, not four
“Proxy only” failed its own blind audit at 78% disagreement, against 0% for directly measurable.
Every auditor independently reported it was the nearest alternative and almost never the winner. A correction rather than an expansion, and the cheapest thing on this page.
Outcome sentences and progress indicators, built and withdrawn
A one-sentence outcome on every gap, and a progress indicator on eight of them. Both were built, both are in the CSV and JSON, neither is proposed.
The critical paths do the same two jobs better. A chain has to state the axis it runs on, which is what the outcome sentence was for, and it carries a sourced quantity on every step, which is what the indicator was for. Two attributes doing a job a third does better is a worse proposal than one, so they came out.
Decomposing composite gaps
Some gap statements bundle several unrelated research programmes under one heading, and one tier and one AI type cannot represent them.
Written up in full, with both patterns and a worked proposal. Not applied, because splitting them means authoring gap records Convergent did not write.
Urgency and impact, if they ever want them
There is no score column anywhere in this augmentation, by design.
Convergent deferred prioritisation deliberately and a stranger ranking their map would be presumptuous. But the reason to build a gap map is eventually to choose. Listed because refusing to build it was a decision, not an oversight, and because the call is theirs.
Coverage
103 gaps is not the field, and Convergent say so plainly.
The two proposed gaps here demonstrate a method for adding one — near-duplicate checked against every existing gap and capability, funding-checked, written to their format. They are not a survey. This is the largest item here and the least suited to being done from outside.
What is deliberately not on this list
Anything that ranks. No numeric score column exists in the schema and adding one is not future work, it is a different project.
Any redesign of the interface or the schema. Two of the items above imply new columns and say so. Proposing a redesign by shipping one is how a contribution turns into a rewrite of somebody else’s map.