Indexing every PMID in every entry - CORRECTED to 4,000 distinct papers across 2,411 entries after the original regex was found to miss 73.7% of them by matching prose but not the schema's own citation field, an undercount that made this entry declare a gap the corpus refuted 14 times over

**SUPERSEDED IN PART, 2026-09-23 (cycle 120). THE HEADLINE NUMBER IS A 52% UNDERCOUNT AND EVERY TOPIC COUNT BELOW IS COMPUTED ON A QUARTER OF THE CORPUS.** The regex required the literal string `PMID`, so it matched prose (`PMID 35288693`) and could not match this schema's own citation field (`"pmid": "35288693"`). Re-run with both paths unioned: | | cycle 118 | corrected | |---|---:|---:| | distinct PMIDs | 1,908 | **4,000** | | entries citing one | 1,230 | **2,411** | | resolving to a real paper | 100% | **100%** | **The prose path alone missed 2,946 of the 4,000 - 73.7%.** The 100% resolution rate that made this entry sound solid was a PRECISION measure, and the script had no recall measure at all; cycle 119's follow-up then inherited the same half-corpus. **The entry's own caveat named this exact failure** - *"any citation written without that prefix is missed entirely and the 1,908 is itself a lower bound"* - and it was written as a hedge rather than acted on as a task. **The cost was not abstract: this entry's own gap selection was wrong.** It reported single-cell as the thinnest topic at 8 papers, *"none of them a KMT2A::AFF1 or infant-ALL study"*, and that selected the next cycle's beat. The corrected index holds **19** single-cell papers, and the most-cited of them is **PMID 35288693, Khabirova 2022 Nat Med, *Single-cell transcriptomics reveals a distinct developmental state of KMT2A-rearranged infant B-cell acute lymphoblastic leukemia*, cited by 14 entries here** - precisely the study declared absent. Two more follow: Candelli 2022 *Leukemia* on relapse-initiating cells in MLL-rearranged infant ALL (4 entries) and Chen 2022 *Blood* on KMT2A-r single-cell multiomics (5 entries). **Corrected topic counts** - every one of them higher, none by the same factor, so the ranking itself was unreliable and not merely the level: relapse/MRD **91** (was 26), metabolism **65** (39), condensates **51** (22), splicing **48** (15), CAR-T **38** (14), 3D genome **30** (10), degrons **25** (14), single-cell **19** (8). **What survives.** The mechanism this entry diagnosed is real and is now better evidenced, not worse: a citation is not prose, and searching entry text cannot find it. The rule stands. What fails is this entry's execution of it - and the failure is the fourth consecutive instance of the pattern it was built to stop, *measuring an index instead of the thing*, this time in the instrument built to stop it. See `ana-the-index-built-to-stop-gap-claims-made-one` for the full accounting. --- **Statement.** Three consecutive dispatch briefs asserted a gap that did not exist: | cycle | the brief claimed | what was actually there | |---|---|---| | 113 | *"the analogy axis is cold"* — 17 entries by `type` | a 68-entry `comparative/` folder; 458 entries touch another fusion family | | 115 | the menin-vs-DOT1L comparison exists in the index fusion | **RS4;11 has zero MEN1 or VTP samples** — two true facts joined into a false one | | 117 | *"the surfaceome screen has never been run"* | **PMID 33910331 ran it in 2022 and nine node files cite it** | **The third is the instructive one.** I searched entry **ids** and **prose** for subject words and found nothing, because the refutation was a **PMID inside entries filed under other topics** — the LAMP5 paper lives in a `cell-state-` entry. **A citation is not prose.** It is the corpus's link to the literature, and it is where *has this been done?* is actually answered. **Built:** every PMID in every node file, with titles fetched from PubMed. - **1,908 distinct PMIDs**, **3,440 distinct DOIs**, across **1,230 of 2,602 entries** - **1,908 of 1,908 resolve to a real paper — 100%** **That 100% is worth stating on its own.** This project has had two PMIDs resolve to entirely unrelated papers and, in the last three cycles, agents reached from memory for identifiers that turned out to be a *Trypanosoma brucei* paper, an asciminib paper and a cheese-whey biohydrogen paper. **Every one of those was caught before it was written.** The committed citation record has no unresolvable identifier in it. **And the index immediately paid for itself.** Searching it by topic rather than counting entries gives a coverage profile the entry counts could not: metabolism **39** papers, relapse/MRD **26**, condensates **22**, splicing **15**, CAR-T **14**, degrons **14**, 3D genome **10** — and **single-cell just 8**, each cited once or twice, **none of them a KMT2A::AFF1 or infant-ALL study**. That is a real gap in the instrument best suited to the PI's *cell type* and *cell state* axes, and it selected this cycle's second beat.