Closure — evidence census

How it works

Three steps — and no model in the part that decides what exists.

Define a field, not a paper. A set of queries, written down and version-controlled, pulls every indexed work matching the method. Then anything cited by enough of those works is pulled in too — which catches the foundational papers no query string names.

Follow who cites what. For every work, walk its references outward and inward. Papers citing the same literature are in the same conversation. This never compares anything to a search term, which is exactly why it finds work you did not know to look for — retrieval returns what resembles your query, and absence resembles nothing.

Classify, then count. Each work is typed by which estimator it uses and what kind of contribution it makes. Counting typed things gives a table where a zero means something: nobody has done this — against a population you can see.

Why classify instead of rank? Nobody can check whether a work is correctly ranked 7th. Anyone can check whether a paper is about distance sampling. We built ranking, measured it twice, found no signal — so the output is a table you can audit rather than a list you have to trust.

The census

Bounded by a method, not by one paper's bibliography.

A per-paper roster moves when you change the seed or the data provider — ours reordered when the forward-edge source changed. A census is bounded by camera-trap animal density and abundance estimation, so it holds still, and corrections accumulate into it instead of evaporating on the next run.

1,003

works in the census

912

found by definition query

added by citation closure — foundations no query names

version-controlled definition queries

The boundary is reviewable. The queries that define this census live in the repository, so a change of scope is a diff someone can argue with — not a silent drift in what "complete" meant.

Evidence grid

What this whole field has done, by method and by contribution type.

method develops or formalises an estimator

validation tests one against a known or independent population

application applies an existing estimator to a study system

review synthesis or commentary, implementing nothing

automated the detection or measurement step is done by machine

A · is a genuine gap — no work of that kind was found inside the denominator. The reliability column is not decoration: it is the measured precision of the classifier for that row, so you know which counts to trust before you use them.

Method	method	validation	application	review	total	automated	reliability
SCR	66	28	59	6	160	0	unmeasured
not-stated	19	15	29	6	69	1	P 17% · R 82%
REM	14	6	10	3	33	0	P 100% · R 32%
CT-DS	16	5	11	·	32	5	P 100% · R 56%
index-RAI	4	7	10	5	27	1	P 100% · R 33%
N-mixture	11	6	1	·	18	0	P 100% · R 25%
REST	7	2	4	·	13	0	P 100% · R 40%
spatial-count	6	3	1	·	10	0	unmeasured
TTE	5	2	·	·	7	0	P 100% · R 10%

6 empty cells 9 method families 2 replication candidates

The not-stated row is a feature. Those works estimate density but never name the model, even in their abstract. An earlier version of this vocabulary forced them into whichever class shared a word with the title, which is how its largest row reached 31% precision. Declaring them is the honest alternative.

Precision before recall. Every estimator class scores 100% precision at the title tier — a named method is a correct one. Recall is low there by design, because the classifier abstains instead of guessing, and the abstract tier recovers what it left.

Reliability

The classifier was measured before its output was published.

Classification runs in two tiers. Titles first, then works the title cannot decide are escalated to their abstract. To measure the title tier, two independent labellers were given strictly more information — title plus abstract — and ground truth is where those two agree. Asking a second model whether the first was right would have been the same correlated-error trap that already produced one false result here.

The category surgery worked. An earlier vocabulary put identification status ("unmarked") on the estimator axis, producing a sink class that absorbed anything with the word in its title — 31% precision on the largest row in the table. Splitting the axes and permitting the classifier to answer not-stated moved inter-labeller agreement from κ 0.73 to κ 0.95, and every estimator class to 100% precision: when a method is named here, it is right.

What the percentages below do and do not cover. They measure the title tier only. Its recall is deliberately low because it abstains rather than guesses; the abstract tier then resolved 84 of the 101 works it could not decide. The two-tier pipeline as a whole has not been measured end to end — that is the next evaluation, not a claim being made here.

0.95

Cohen's κ between independent labellers — is the task well defined?

60%

title-tier accuracy — abstentions count as misses

129

works with solid ground truth

134

abstracts read — none stored, none served

κ = 0.95 is moderate, not strong. Two labellers holding the abstract disagreed on 4% of works. That is a finding about the categories, not the classifier: some method families overlap in practice and need tightening before their counts can be sharpened.

Position

Where one manuscript sits inside the field.

Manuscript

Automated Parameter Estimation for Camera Trap Density Models Using Computer Vision-Enhanced Distance Sampling

10.64898/2026.06.14.732225 · cites 56 DOI-bearing works

Of the 298 works in this field using a named estimator, this manuscript cites 14 (5%)

Method	in field	cited here	engagement	automated in field
SCR	159	1	1%	0 — open
REM	33	4	12%	0 — open
CT-DS	32	4	13%	5
index-RAI	26	0	0%	1
N-mixture	18	0	0%	0 — open
REST	13	2	15%	0 — open
spatial-count	10	1	10%	0 — open
TTE	7	2	29%	0 — open

This is the differentiation question, answered against a denominator. 6 of the field's method families have no automated implementation at all — REM, REST, TTE, SCR, spatial-count, N-mixture. That is not a hunch about novelty; it is a count over a population you can inspect.

Low engagement is not an accusation. A methods paper is not obliged to cite a whole field. The number is here so a reviewer's "you have not engaged with the SCR literature" becomes a checkable claim rather than an impression.

Gaps

Two kinds, from the same set operation with the sign reversed.

What your peers cite and you do not

Of the works this conversation treats as core — cited by more than half its members — 8 of 8 appear in the seed's own bibliography. The rest are below, as rates rather than ranks: "17 of 40 peers" is checkable against your own reference list in a minute.

Work	peers citing	rate	year
Applying a random encounter model to estimate lion density from camera traps in Serenget	17/40	43%	2015
Spatially explicit models for inference about density in unmarked or partially marked po	13/40	33%	2013
An invasive‐native mammalian species replacement process captured by camera trap survey	12/40	30%	2016
Validating camera trap distance sampling for chimpanzees	11/40	28%	2019
Random encounter model is a reliable method for estimating population density of multipl	11/40	28%	2022
Camera trapping photographic rate as an index of density in forest ungulates	11/40	28%	2009
Framing pictures: A conceptual framework to identify and correct for biases in detection	10/40	25%	2019
Estimating animal abundance and effort–precision relationship with camera trap distance	10/40	25%	2021
Application of the Random Encounter Model in citizen science projects to monitor animal	10/40	25%	2020
Risky business or simple solution – Relative abundance indices from camera-trapping	10/40	25%	2013

Grouped, not ranked. The order of this table carries no demonstrated signal — see Validation. Read it as a set.

Replication candidates

Cells holding exactly one study — a result nobody has independently reproduced. Only 11–18% of ecological research reaches its full informative value, so a tool that surfaced novelty alone would be pointed the wrong way.

spatial-count · application 1N-mixture · application 1

Validation

Measured against what real authors cited — not against another model's opinion.

Every published bibliography is an expert relevance judgement made before this existed. So: hide a quarter of it, rebuild the pool from the rest, and count what returns. Across 10 seed papers and 64 held-out real citations.

79%

of works the authors actually cited were recoverable in the pool

0.00

score for off-field and same-field-different-topic controls, from three expert graders

p 0.66

semantic reranking vs graph order — no effect

κ 0.95

agreement between independent labellers after the category surgery

A result we published and then withdrew. Semantic reranking scored r = +0.83 against expert personas — model graders judging a model reranker. Against real author citations it showed no effect, and gained on pre-2022 papers while losing on recent ones: the signature of pretraining recall, not judgement. The ordering claim was removed.

Rights

Every document, and every source we read them from.

398

documents classified

34%

servable — open licence or public domain

104

publisher TDM terms declared

policies to read to clear them all

explicit_commercial 134silent 110tdm_declared 104explicit_noncommercial 50

Our own supply chain

A rights ledger that exempts its own sources is decoration. Each data dependency carries a verdict, and the build refuses to publish an artifact derived from a non-commercial source.

crossref CC0 / PDopenalex CC0 / PDs2 restrictedunpaywall CC0 / PDanthropic-claude CC0 / PD

Reading is not serving. A bibliographic reference is a fact and may always be named; reproducing expression from a restricted source may not. Abstracts were read to build the reliability measurement above and are stored nowhere.

What this does not claim

The limits are the product too.

The ordering carries no demonstrated signal. Citation frequency scored r = +0.10 against expert relevance. Semantic reranking scored p = 0.66 against real author citation decisions — no effect — and gained on pre-2022 papers while losing on recent ones, the signature of pretraining recall rather than judgement. Rows are grouped and measured, never ranked.

What is validated. Across 10 seed papers and 140 held-out real citations, 79% of works the authors actually cited were recoverable in the pool. Same-field and off-field controls scored 0.00 with three independent expert graders.

Classification is from titles. Method family and contribution type are measured above. Sample sizes, effective sampled area and validation targets require full text and are not asserted anywhere on this page.

What has this field
actually done?