Checked · The Architect Question
The recall argument is available to Hancock, and here is exactly where it stops
The claim
If archaeological survey methods only find a fraction of what is there, then the absence of evidence for a lost civilisation proves nothing.
Our verdict
Under reviewThe first half is true and we can prove it with our own numbers. Our detector recovers about a third of known monuments and is blind to entire classes. What the argument cannot do is get from a measured blind spot to a specific thing inside it, particularly one that would also leave evidence outside.
This is the best argument on that side of the table, almost nobody makes it properly, and we can hand them the numbers for it.
The argument, at full strength
Archaeological survey does not see everything. It sees what its methods are built to see, over the ground it happens to cover, at the resolution it happens to have. Reasoning from "we have not found it" to "it was not there" therefore requires knowing how good the looking was, and in most cases nobody has measured that.
That is not a rhetorical move. It is the standard structure of an inference from absence, and it is exactly the inference professional archaeology makes when it says complex societies did not exist before a certain date.
Our own numbers, which support it
Over 16 square kilometres of Salisbury Plain containing 121 scheduled monuments, our detector recovered 40. That is 33.1%. Over 16 square kilometres of Bodmin Moor containing 94, it recovered 27, which is 28.7%. On Bodmin, 443 of its 470 candidates are not within tolerance of any scheduled monument.
Runs 005 and 006, scored against the National Heritage List, 6 August 2026. Tolerance 35 metres, or the monument span capped at 120 metres.
Worse, and more useful: the aggregate hides structural blindness. Split by monument class and our detector scored zero on disc barrows, zero on henges, zero on cursus monuments, and zero on pond barrows. The pond barrow zero is permanent, because a pond barrow is a hollow and the detector only accepts positive relief. No setting fixes it.
So: a survey can be blind to an entire class of monument, report a respectable overall figure, and give the reader no way to notice. We publish that about our own work. It is straightforwardly true, and anyone arguing that the record is incomplete can cite us.
It gets worse for us, and the worse version is the more useful one to hand over.
On Bodmin Moor, 470 randomly scattered points score 28.2% recall against the scheduled monuments. Our detector scored 28.7% with 470 candidates. On that block our recall figure is indistinguishable from noise, and we published it for three days before computing the baseline that says so.
Run 007, 9 August 2026. 400 rounds of uniformly random points matched to the real candidate count, same monuments, same tolerance rule.
Anyone making the recall argument should take that and use it. It is a stronger version of their point than the one they usually make, and it comes with a seed and a run log.
Where it stops, in one sentence
A measured blind spot licenses uncertainty about what is inside it. It never licenses a claim about what is inside it.
We know our detector cannot see pond barrows. That is a reason to withhold judgement on how many pond barrows are in our survey area. It is not a reason to assert that the area is full of them, and if we did assert that, the correct response would be to build the hollow finder and measure, not to argue.
What the recall argument earns
Confident claims of absence from unmeasured surveys should be discounted. Ours included.
Whole categories of evidence can be invisible by construction, and usually nobody has checked which.
The burden on anyone asserting absence is to state their recall, and almost nobody does.
Funding replication and blind-spot measurement is a legitimate demand, and cheap.
What it does not reach
A blind spot is symmetric. It conceals the ordinary far more often than the extraordinary, because the ordinary is far more common.
It predicts a gap, not a content. Nothing about the gap tells you what fills it.
The claim in question predicts evidence outside the blind spot too, and that evidence is absent from a dense, well-covered record.
The correct response to a known blind spot is to measure it, which is available and is not being done by the people invoking it.
The test we would accept
We built the hollow finder, because complaining about other people's blind spots while keeping our own would be indefensible. It recovered a pond barrow, which closes the structural zero. Then the null showed that almost all of the headline recall improvement was the arithmetic of having 762 more candidates rather than better detection, and we published that too.
That is the whole move, and it is available to anyone. If you think the record is missing something, specify what class of evidence would be invisible, build the instrument that would see it, run it over ground with known answers, and report how often it is right. Then the argument is a measurement.
Where this leaves things
The five labels, applied to this question. Nothing is left unlabelled.
Our own recall is 33.1% and 28.7% on two 16 square kilometre blocks, with precision of 23.8% and 5.7%.
Our detector is blind by construction to pond barrows and scored zero on four monument classes.
How much of the prehistoric record is missing for methodological rather than historical reasons. Nobody has a general figure.
The inference from a blind spot to a specific claim about its contents.
Our own Bodmin recall figure as evidence about the detector. 470 random points score 28.2% on that ground against our 28.7%.
Sources
- Sea level and global ice volumes from the Last Glacial Maximum to the Holocene. Lambeck, K. et al., PNAS 111, pages 15296 to 15303, 2014.
Related
Where this is argued in full
Answering the Architect Question
Both cases at the same length, measured and printed at the back. The argument runs across four papers in one journal and you can read all of it.
Both cases at the same length, measured and printed at the back. It is written and not on sale, and advance readers can read it first.