124 entries already write NOT APPLICABLE as a category statement on the localization dimension, and the epigenetics dimension has no equivalent - so a correctly-annotated Golgi enzyme and an unwritten chromatin protein are indistinguishable in the epigenetics figure
also called q181 the missing third state · NOT APPLICABLE convention
When a protein sits in the Golgi and has no address inside the nucleus, this project's localization entries say so explicitly - 'NOT APPLICABLE, and this is a category statement, not a gap'. Nothing like that exists for the chromatin topic, so a protein that correctly has no chromatin biology looks exactly like one nobody has got round to.
The same thing, for a scientist
This corpus has evolved an undocumented convention on the localization dimension: an explicit NOT APPLICABLE, frequently with the clarifying phrase 'this is a category statement, not a gap' or 'a bounded absence, not a NOT APPLICABLE'. 124 nodes carry it inside data{}. gene-fut4's subnuclear field reads 'NOT APPLICABLE - AND THIS IS A CATEGORY STATEMENT, NOT A GAP'; gene-cd19's reads 'not applicable - CD19 is a single-pass plasma membrane protein'; gene-smarcd1's reads 'NONE ASSIGNED - A BOUNDED ABSENCE, not a NOT APPLICABLE', which distinguishes the two states explicitly. No equivalent exists for epigenetics: of the 24 epigenetics key spellings in use, exactly one - epigenetic_bounded_absence, on 10 entries - encodes a negative, and it encodes the unmeasured state, not the inapplicable one. The consequence is measurable: 25 of the 195 entries the epigenetics line scores as empty already declare NOT APPLICABLE somewhere in data{} for the localization dimension, including every one of gene-cd19, gene-cd33, gene-cd34, gene-fut4, gene-fut9, gene-elovl1, gene-kras, gene-nras, gene-vim and gene-mapre1.
The convention, quoted
Verbatim from data{} fields in entries this beat did not write 1:
| entry | field | text |
|---|---|---|
gene-fut4 | subnuclear | "NOT APPLICABLE - AND THIS IS A CATEGORY STATEMENT, NOT A GAP" |
gene-fut9 | subnuclear | "NOT APPLICABLE - FUT9 is a type II single-pass Golgi membrane..." |
gene-cd19 | subnuclear | "not applicable - CD19 is a single-pass plasma membrane protein" |
gene-elovl1 | subnuclear | "NOT APPLICABLE. ELOVL1 is a seven-transmembrane integral ER..." |
gene-kras | subnuclear | "NOT APPLICABLE, AND THIS IS A BOUNDED REJECTION RATHER THAN..." |
gene-mapre1 | subnuclear | "NOT APPLICABLE for wild-type MAPRE1 - a protein with no nuclear..." |
gene-smarcd1 | subnuclear | "NONE ASSIGNED - A BOUNDED ABSENCE, not a NOT APPLICABLE." |
gene-bicra | subnuclear | "UNPUBLISHED - a bounded absence, not a NOT APPLICABLE." |
The last two are the ones that matter most. Two entries, written by different beats, explicitly contrast the two negative states -- nothing to say versus nobody has looked. The vocabulary is already three-valued in practice and has never been written down as a rule.
Epigenetics has one negative spelling and it is the wrong one
Of the 24 epigenetics key spellings in use across the 541-entry population, the counts are led by epigenetic_readout (182), epigenetic_mark_named (23), mark_written (22), epigenetic_role (19), non_histone_substrate (16) 1.
Exactly one encodes a negative: epigenetic_bounded_absence, on 10 entries. A bounded absence is "the measurement has not been made". It is not "there is no measurement to make". So the dimension can currently express covered and unmeasured, and has no way at all to say inapplicable -- which is the state this beat estimates applies to roughly a third of the shortfall, ~108 of 329 entries.
And note non_histone_substrate on 16 entries: that spelling exists because authors needed to say "this enzyme acetylates something that is not a histone". That is a category statement in disguise, invented ad hoc because there was no field for it.
The 25 entries that already answered the question on the other dimension
25 of the 195 entries scored empty on epigenetics already carry a NOT APPLICABLE declaration in data{} for localization 1: comut-kras-kmt2a-r, gene-bicra, gene-ccnt2, gene-cd19, gene-cd33, gene-cd34, gene-cdk12, gene-elovl1, gene-fut4, gene-fut9, gene-hexim2, gene-igf2bp2, gene-kras, gene-larp7, gene-luzp1, gene-map3k14, gene-map7d1, gene-mapre1, gene-mepce, gene-nras, gene-prkar1a, gene-septin2, gene-skp2, gene-smarcd1, gene-vim.
For a Golgi fucosyltransferase or a lipid-anchored GTPase, an author who has already written "not applicable, this is a category statement" about the subnuclear compartment has, in substance, answered the chromatin question too. The epigenetics figure counts all 25 as uncovered.
Three of them -- gene-ccnt2, gene-hexim2, gene-larp7, gene-mepce, gene-cdk12 -- are P-TEFb-axis proteins where the category call is genuinely arguable, and those are the interesting cases, not the fucosyltransferases.
| script | analyses/q181-is-the-epigenetics-gap-real/{baserate,keys}.py |
|---|---|
| status | run |
| generated | 2026-09-24 |
| na convention nodes | 124 |
epigenetics negative spellings
{
"epigenetic_bounded_absence": 10
}na declared but epigenetics empty
[ "comut-kras-kmt2a-r", "gene-bicra", "gene-ccnt2", "gene-cd19", "gene-cd33", "gene-cd34", "gene-cdk12", "gene-elovl1", "gene-fut4", "gene-fut9", "gene-hexim2", "gene-igf2bp2", "gene-kras", "gene-larp7", "gene-luzp1", "gene-map3k14", "gene-map7d1", "gene-mapre1", "gene-mepce", "gene-nras", "gene-prkar1a", "gene-septin2", "gene-skp2", "gene-smarcd1", "gene-vim" ]
measurements — the numbers behind the chart
[
{
"quantity": "nodes carrying an explicit NOT APPLICABLE / category statement in data{}",
"value": 124,
"units": "nodes",
"n": 3032,
"subject": "the undocumented three-valued convention on the localization dimension",
"universe": "all 3,032 node files at commit ac2aeab, 2026-09-24",
"null": null,
"note": "4.1% of nodes; concentrated on subnuclear_localization"
},
{
"quantity": "epigenetics data-key spellings that can express a negative",
"value": 1,
"units": "of 24 spellings",
"n": 24,
"subject": "epigenetic_bounded_absence, on 10 entries - the UNMEASURED state, not the INAPPLICABLE one",
"universe": "24 distinct _EPI_KEY-matching data keys across the 541-entry population",
"null": null,
"note": "there is no spelling at all for 'this protein has no chromatin context'"
},
{
"quantity": "epigenetics-empty entries that already declare NOT APPLICABLE for localization",
"value": 25,
"units": "entries",
"n": 195,
"subject": "entries where an author has already made the category call on a neighbouring dimension",
"universe": "195 gene/protein entries with no epigenetics field and no mark token",
"null": null,
"note": "12.8% - including every one of CD19, CD33, CD34, FUT4, FUT9, ELOVL1, KRAS, NRAS, VIM"
}
]- Are all 124 NOT APPLICABLE hits localization statements? Eight were read; 116 were not, and the phrase 'not a gap' is loose enough to match incidental prose.
- Does a NOT APPLICABLE on subnuclear localization license one on epigenetics? IMPDH2 and CDK12 are counter-examples in this corpus - no subnuclear address, but a real chromatin or CTD readout - so the inference is not safe in general and the 25-entry overlap should not be treated as 25 answered questions.
- Should the convention be written into SCHEMA.md as a rule, given that 124 of 3,032 nodes is a minority practice? Rule 96 records what happens when a convention only some entries follow: the detector built for it undercounted by half.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
124 nodes in this corpus carry the phrase 'NOT APPLICABLE', 'CATEGORY STATEMENT' or 'not a gap' inside their data{} payload, overwhelmingly on the subnuclear_localization field, and two entries (gene-smarcd1, gene-bicra) explicitly contrast 'a bounded absence' with 'a NOT APPLICABLE'; of the 24 epigenetics key spellings in use across the 541-entry population only one, epigenetic_bounded_absence on 10 entries, encodes any negative, and it encodes the unmeasured state rather than the inapplicable one; 25 of the 195 entries scored empty on epigenetics already carry a NOT APPLICABLE declaration for localization.
What this evidence cannot show
The phrase scan is case-insensitive over the serialised data{} and does not verify that each of the 124 hits is a localization statement rather than a NOT APPLICABLE about something else - the 8 quoted in the table were read, the other 116 were not. 'not a gap' as a search phrase is loose and will match prose using those three words incidentally. Key enumeration covers top-level data keys only.
Evidence that argues against this (2)
- Calling this a convention overstates it. 124 of 3,032 nodes is 4.1%, and the localization dimension covers 419 entries by field - so even on its own dimension the NOT APPLICABLE form appears on a minority of covered entries, and most localization records simply state a compartment without ever addressing whether a subnuclear answer applies. A minority practice on one dimension is thin ground for asserting that the corpus 'already has' a three-valued vocabulary; Rule 96 in this project's brief records an emergent ALL-CAPS convention whose detector found 8 outcomes where 17 existed, precisely because a convention only some entries follow is indistinguishable from none.
- The 25-entry overlap is presented as evidence that these entries have effectively answered the epigenetics question, but a NOT APPLICABLE on subnuclear localization does not entail one on epigenetics. A protein can have no subnuclear compartment and still carry a chromatin readout - IMPDH2 is in this corpus with a damage-induced chromatin association and no subnuclear address, and CDK12 phosphorylates the Pol II CTD while having no named nuclear body. The inference from one dimension to the other is the same shape as Rule 89's defect: crediting a dimension on evidence about a neighbouring one.
the corpus already writes the third state on a neighbouring dimension, which is why the category-error share is invisible here
the recommendation for a third state rests on the 124-node localization convention already in use