Backfill to 1985: construction and coverage report

This extension searches 1985-01-01 through 1999-12-31, separately from the previous 2000 backfill and the frozen ten-year collection. Main coverage now runs from 1985-01-01 through 2026-09-09. The extension was acquired and reviewed on 14 September 2026. It is a screened OpenAlex backfill with explicit provider and metadata limitations, not an exhaustive bibliography.

Result and units

Counts are bibliographic canonical records, not necessarily distinct intellectual works. Preprints, journal versions, corrections, and similar titles are not merged without identifier evidence. The 1,980-record historical view includes nine records already present before this extension. Five earlier accepted records newly qualify for the expanded main window. The original 3,076-record decade and 59-record foundation views are preserved; the combined backfill and decade overlap by 47 canonical IDs with dated source manifestations in both windows.

Discovery requests and stopping evidence

The search plan defines substantive Andreev mechanisms and directly related superconducting proximity, Josephson spectra/transport/dynamics, YSR, and hybrid-device literature. A surname, unrelated Andreev eponym, or passing mention is insufficient. Exact provider-native requests are retained in searches/2026-09-14-backfill-1985/. The ledger records timestamps, returned counts, new canonical counts, continuation specs, and the stopping status of each batch.

Twelve OpenAlex families were run serially in batches of at most 200 records. The core used title search; the larger Andreev query used title/abstract search and cursor continuation. Related families used the literal query strings shown in their first request; they are discovery heuristics, not inclusion rules.

All twelve successful families ended with provider exhaustion and no next spec. Across them, 3,671 returned records represent 3,096 distinct OpenAlex IDs. There are four repeated IDs within individual families: two in proximity, one in current-phase, and one in Shiba. Their unique-ID counts, rather than returned row counts, match the corresponding provider totals. Cross-family overlap is also retained. The native-ID manifest and validation make these comparisons inspectable. Ending pagination does not establish literature coverage.

arXiv is incomplete. The core query failed twice and the related query once. Sanitized diagnostics matched rate-limit/gateway error text; the failures are recorded rather than treated as empty or exhausted results. Neither arXiv family produced a successful backfill batch. No claim of native arXiv coverage is made, especially for works not represented in OpenAlex. The user-designated PatchScholar development thread received a request for structured, credential-safe provider errors and bounded timeout/retry diagnostics.

Known-paper recovery and search expansion

The DOI audit found two relevant papers missing from the keyword results: Universal limit of critical-current fluctuations in mesoscopic Josephson junctions and Quantum transport in semiconductor-superconductor microjunctions. They were acquired, screened, and included as records 11312 and 11313. This gap triggered the additional mesoscopic Josephson and current-phase search families. The audit also recovered existing records for electron-pair tunneling, the 1997 quantum-transport review, and ferromagnet/superconductor Andreev reflection. One proposed seed DOI resolved to unrelated exciton spectroscopy (11314); it was explicitly excluded. The audit therefore checked metadata rather than assuming that every supplied identifier was a relevant seed.

Citation links are cached provider references resolved to corpus IDs. The expanded graph contains 58,634 directed links, including 55,305 between accepted records, and 5,660 globally unlinked records. No recursive citation crawl was run for this extension; missing links do not establish absence of citations. The seed check is a bounded recovery audit, not a comprehensive citation audit.

Screening, uncertainty, and retained history

Three Luna agents screened disjoint, frozen input batches using titles and cached abstracts. Every new ID has an include, exclude, or defer judgment, a reason, specific evidence, assessment=agent, and human_confirmed=false. Relevant non-article types may be included when their subject is substantive; standalone datasets, software, and supplements are excluded as literature entries. Topic tags can be absent when no particular physics category is justified.

The review passes and overrides are preserved in reviews/backfill-1985/. Audits revisited deferred cases and 278 physics-adjacent exclusions. Compared with the first-pass decisions, 44 exclusions and 44 deferrals became inclusions, 49 exclusions became deferrals, and eight deferrals became exclusions. Specific false exclusions of mesoscopic NS-junction and phase-controlled interferometer papers were corrected. The original judgments remain in their first-pass CSVs; the expanded register names the final evidence-bearing review register for each record.

287 new records remain deferred because the available evidence does not establish a substantive connection or needs fuller reading. Their list is explicit; they are not counted as accepted. This was title/abstract screening and auditing, not comprehensive full-text review. The published snapshot has abstracts for 9,678 of its 12,218 records. It preserves 403 shared-title groups for identity follow-up; title similarity alone does not establish identity.

No prior decisions, review evidence, or frozen window assignments were changed. The new register refreshes cached source identifiers/types/dates and computes eligibility from all dated manifestations. Four previously present records gained an additional OpenAlex manifestation during this backfill. The old 2000 and decade registers remain byte-for-byte preserved, while the new register records current source provenance. Database checks confirm 13,814 prior review events preserved and 3,071 new final review events. Intermediate agent judgments are recorded in CSVs; they were not separately replayed into database history.

Historical distribution

The year below is the earliest qualifying 1985–1999 source date. It can differ from the frozen original window date or journal publication date.

Reproduction and validation

The database was backed up before ingestion. Reusable acquisition, projection, review application, and BibTeX serialization use PatchScholar; the scripts in here are collection-specific recipes. Requests/results and screening inputs are bounded and frozen locally. Raw caches, downloads, and database state are ignored by Git and are not published. Canonical IDs depend on this database; a fresh acquisition elsewhere may assign different local IDs. Provider results may drift.

# Discovery resumes successful pages and reports failed families.
pixi run python -m scripts.continue_1985
# Freeze new disjoint batches, then screen them into matching review CSVs.
pixi run python -m scripts.prepare_1985 BATCH_NAME
pixi run python -m scripts.consolidate_1985
# Apply final decisions once; review-event validation detects extra replays.
pixi run scholar run-pack data/batches/2026-09-14-backfill-1985/final-review.yaml
pixi run scholar corpus export andreev-literature > exports/andreev-all-2026-09-14.bib
pixi run python -m scripts.export_1985
pixi run python -m scripts.validate_1985
pixi run explorer-build
pixi run explorer-test
npm test

Pybtex is a declared project dependency. It parses every export and compares its exact key set with the corresponding accepted register subset. SQLite integrity and foreign-key checks pass. Database decisions/reasons/labels match every register row. Previous events and frozen review fields are checked against the pre-run database/register. The browser suite checks actual hiding of excluded records, search highlighting, paper cards, date views, exports, and desktop/mobile flows. The graph test compares the exact exported directed edge set with the snapshot.

Validation artifacts: counts, local/database/provider audit, bibliography audit.

Explorer and downloadable data

The default highlights 7,784 accepted articles/preprints in 1985–2026. Review decision now determines actual graph visibility: Included hides excluded and deferred records. Search and the other filters highlight matches within that visible network. All decisions exposes the complete screened corpus. Recent papers have no special rings or provisional labels. Their citation rates remain descriptive and sensitive to short observation times.

• Expanded 1985–2026 bibliography: 8,623 entries.
• 1985–1999 extension bibliography: 1,980 entries.
• Combined 1985–2016 backfill: 5,594 entries.
• All accepted including older foundations: 8,629 entries.
• Expanded review register.
• 1985–1999 review view.

The original ten-year and 2000 exports and review registers remain available. Completion here means that this recorded OpenAlex extension has been screened, audited, and exported with explicit deferrals and failed arXiv requests. It does not mean that every relevant 1985–1999 publication has been found or read.