Archives of California: A Documentary Calendar

Resolve a Bancroft footnote

Since 1906, the manuscript citations in Hubert Howe Bancroft’s History of California (Prov. St. Pap., MS., xvi. 123 and their kin) have pointed at documents destroyed in the San Francisco fire. Paste one here and it resolves to the surviving Savage copy in this catalog, and from there to the manuscript image.

How it works, and what it cannot do

• Coverage. A footnote resolves when its volume is in this catalog. A miss is honest: it names the series and tomo it is waiting on, and each newly cataloged volume makes more of Bancroft’s footnotes resolve retroactively.
• Accuracy. The resolver is run against a census of 8,839 Bancroft footnotes to already-cataloged records (rebuilt 2026-09-09). This is an internal consistency test: it shows that the citation parser and the page index agree with each other, not that the underlying page readings are correct. It is an automated internal round-trip. It proves the index is self-consistent, not that every underlying page number was read correctly from the manuscript; an independent, manuscript-verified sample of resolved footnotes is planned. Where several Savage abstracts share a cited page, all candidates are shown; that is the correct answer, not a failure. Volumes whose page systems carry a caveat are flagged on each result.
• Id. / ibid. chains carry no series of their own. Paste the last full citation before them.
• Non–C-A citations (the Vallejo Documentos, dictations, mission registers, county archives…) are other collections and will not resolve here.
• Legislative Records. Bancroft does not cite these volumes by a single page system. Leg. Rec. I (C-A 59) he cites by the Savage copy’s ink foliation, which this catalog records, so those footnotes resolve normally. For Leg. Rec. III–IV (C-A 61) he appears at times to cite the original Diputación libro pagination, which Savage’s copy preserves only sporadically; check a candidate’s date against the footnote, or search by content.

Resolution matches the record’s page numbers against the cite, with tomo boundaries pinned from Savage’s own title leaves and an alias table for Bancroft’s abbreviations. Bancroft’s footnotes cite different page systems in different volumes: usually the original tomo pages, but in some volumes the copy’s own leaf numbers (leaf-verified on C-A 25, 2026-07-12); where both systems are indexed, matches are labeled, and candidates show their dates so you can check them against the footnote. Method & verification: see About.

The tag vocabulary

Tags are compiler-assigned working labels, not a controlled archival thesaurus: places, persons, document genres, and research themes mix freely, and coverage deepens in later-read volumes. There are 14,571 distinct tag strings, but that figure overstates the vocabulary: case and punctuation variants (monterey and Monterey, baja-california and Baja California) are counted separately, and merging them gives about 12,600. Search treats those variants as one tag; the counts below do the same. The 150 most frequent are listed here (scope notes where a tag isn’t self-evident; tags marked as research threads reflect this compiler’s projects). The complete list ships in the CSV download.

People: a prosopographical register

The figures who recur most often in the catalog’s author and recipient fields, each given a canonical identity: the offices they held, the span of dated records that name them (fl.), the name-forms Savage used, and (where two people share a surname) a note telling them apart. Click Find in the catalog to search every record that names a person.

Offices and dates come from reading the manuscripts, not invented to fill a field; bare family surnames (Argüello, Carrillo, Castro…) are deliberately left unattributed rather than guessed. fl. is the span of dated records in which the name appears as author or recipient. An outlier year usually means a later copy, a retrospective mention, or a homonym, so check the records, not this line. The catalog search matches names in summaries too, so a search often returns more than a person’s own correspondence.

About this catalog

A searchable, document-level calendar of the Archives of California (BANC MSS C-A 1–63), Bancroft Library, University of California, Berkeley. The collection is the surviving record of the Spanish- and Mexican-era provincial government of Alta California, c. 1767–1850.

What this is, and what it is not

These are not the original documents. The original provincial archive, held at the U.S. Surveyor-General’s Office in San Francisco, was destroyed in the 1906 earthquake and fire. What survives, and what this catalog indexes, are the copies and abstracts made in the 1870s by Thomas Savage and his team for the historian Hubert Howe Bancroft. That means:

• Most entries are Savage’s abstracts: his 19th-century summaries and paraphrases, not the documents’ original wording.
• Some are verbatim transcriptions (marked “verbatim” in the Source column); a direct quotation from one of these is far more reliable than from an abstract.
• Savage’s choices are why anything survives at all. What he chose not to copy was lost in 1906, unless the other end of the correspondence survives in Mexico (Archivo General de la Nación) or Spain (Archivo General de Indias).
• Dates, names, and details are as Savage recorded them, including his occasional slips. Always check the linked manuscript image before quoting or citing.

This is a finding aid, not an edition: it tells you what exists and where, and links each entry to the manuscript page so you can read it yourself.

A note on citing. Since 1906, anyone citing these series (Bancroft’s readers included) has in fact been citing Savage’s copy of a destroyed original. Cite accordingly: treat an abstract as an abstract (Savage’s summary, not the document’s words), and quote verbatim text only after checking it against the linked manuscript image. Each record’s “Cite as” string names the Savage transcript page for exactly this reason.

Where this is going

I am extending it beyond Savage to the other nineteenth-century compilers who worked from the same records, among them Hittell and Eldredge, and cross-referencing it with the pueblo archives of San José and Los Angeles, with further collections to follow. The aim is a cross-archive index of early California, one that lets a person or a document be followed across collections.

So this calendar is one collection of several, and it is not finished. The companion projects listed at the top of this section are the parts of it that are already published.

Thomas Savage and his method

Thomas Savage (b. 1823, Havana) was Bancroft’s principal assistant for Spanish-American materials and fluent in Spanish. Bancroft called him “my main reliance on Spanish-American affairs.” Beginning 15 May 1876, he and a team of ten to twelve assistants worked through the provincial archive (then “nearly 300 volumes of MSS, ranging from 700 to 1900 pages each,” held at the U.S. Surveyor-General’s Office) and by the first weeks of 1877 had produced the 63 bound volumes cataloged here. He recorded the whole project in his “Report of Labors” (BANC MSS C-E 191, 1879), published in Rose Marie Beebe & Robert M. Senkewicz, Testimonios (Heyday, 2006).

In his own words, how he decided what to preserve:

“Of the useful portion the substance had to be carefully extracted, and written down concisely tho’ clearly. But if the document or any part of it was important enough to be copied, it was taken down verbatim et literatim. What was of no practical use was left out.”Thomas Savage, Report of Labors, 1879

That sentence is the key to this catalog’s Source column: an entry is either an abstract (Savage’s concise extract of the substance), a verbatim transcription (copied word for word), or, for what he judged of no practical use, nothing at all. What he chose to leave out was lost in 1906.

The originals had been gathered by U.S. government order in 1851 and bound in haste “with no attempt at chronological arrangement.” On their state, Savage was blunt:

“The whole is a mere jumble from beginning to end.”Thomas Savage, Report of Labors, 1879

That disorder (dates and subjects mixed freely across the volumes) is exactly why a catalog like this one is needed to navigate them.

The 63 volumes (Savage’s copies)

The Archives of California run to 63 bound volumes (BANC MSS C-A 1–63). All 63 volumes are now in the catalog, at varying depth. The … rows highlighted below (“✓ searchable”) are already in the catalog above; every volume links to its digitized manuscript on the Internet Archive where one is available. Most are cataloged leaf-by-leaf; some remain at a partial or targeted grain. The per-volume coverage is shown in the Volume Guide. Coverage grades: complete: cataloged leaf-by-leaf; partial: the significant documents are cataloged, routine runs bounded; targeted: a single dossier extracted; a few are complete by design or at index / acta grain (register or duplicate-legislation volumes). The ✓ beside a volume marks it as searchable. An empty search result within a partial volume means “not yet cataloged,” not “no documents.”

How to use it

• Search across summaries, authors, recipients, places and themes; filter by volume, source type, grain, or decade; sort any column.
• Click a row for the full record; click a scan number (↗) to open that page’s manuscript image on the Internet Archive.
• Field notes: Source = Savage abstract vs. verbatim transcription (or “unspecified” where his catalog didn’t say); Grain = a single document (item) vs. a leaf- or section-level survey entry; dates carry a confidence flag (exact / month / year / inferred).

How this catalog was made

Entries were produced by reading the digitized Savage volumes on the Internet Archive. Most volumes have been read leaf by leaf; a few are catalogued at a coarser grain, and the Volume Guide states which. I did this reading with the help of a large-language-model assistant (Anthropic’s Claude), which I directed leaf-by-leaf: it drafted transcriptions and abstracts from the manuscript images, and I set the cataloging rules, adjudicated uncertainties, and reviewed the output. Machine assistance is a tool here in the same sense Savage’s ten assistants were his. The judgments, and the errors, are mine.

Four safeguards stand behind the data. First, transcription rules: Savage’s text is recorded as written, uncertainty is marked ([?], [illeg.]), and inferred dates are flagged rather than silently asserted. Second, verification passes: selected high-value documents are re-checked against the manuscript image, and a growing register of verified quotations records exactly what has been confirmed at the leaf; corrections found in these passes are logged, not overwritten. Third, automated integrity checks run on every publication (link targets, date ranges, duplicate identifiers, export consistency, and a sampled link-rot sentinel), and the catalog publishes only when these checks report zero blocking errors (the checks themselves are public in the repository).

Fourth, a blind sample audit, completed 2026-09-16, superseding the audit of 2026-07-19: 200 records were drawn from everything this site serves (stratified across all 63 volumes; the sample committed by hash before any checking began; no substitutions: an unreadable target is graded unverifiable, never swapped), and each was re-read against the manuscript leaf and graded on nine dimensions, 1,742 field judgements in all. Counting only exact matches, pooled accuracy is 81.1% (79.2–82.9); counting logged partial imprecisions as acceptable, 93.3% (92.0–94.4). On the dimensions a citation depends on: dates wrong in 10.1% of records, the abstract’s substance in 12.5%, the original-page citation in 9.3%, the scan pin in 3.0%. 21.0% of records were flawless on all nine dimensions; 64.0% carried no outright wrong field. A correction: this page previously reported the July 2026 audit’s rates (dates 6.5%, substance 5.5%). A September reconciliation regraded a slice of that audit’s own sample at higher magnification and found it had undercounted those two rates by roughly half; the figures above replace it, and the reconciliation worksheet ships in the repository alongside both audits. The catalog’s halves differ: records added since July 2026 measure about twice as clean as the older half on every dimension. Every wrong field the audit found is registered in the errata file with verbatim leaf evidence, and the 16 leaves known to be physically absent from the scans are registered there too. Practical meaning unchanged: a record is a reliable guide to what a document is and where it sits; verify exact dates, spellings, figures and quoted words at the linked leaf before citing. The committed sample, both grading worksheets, and the reconciliation ship in the repository’s audit/ directory.

What the provenance labels on each record mean. Every record card states its tier. “✓ Leaf-verified” (green) means the record carries a dated note recording exactly what was re-checked against the manuscript image and when: a dateline, a quotation, an author line. The note’s scope is the claim, nothing broader. “Provenance: AI-drafted calendar entry, human-reviewed” is the baseline tier for everything else: drafted from the manuscript image by the assistant under my direction and reviewed as described above, but not yet individually re-checked at the leaf. The blind sample audit measured this tier’s field-level error rates, and those numbers are this tier’s honest warranty. Records corrected through the errata register additionally carry the correction box with its proving crop. The absence of a verification note means not yet re-checked, not wrong, and verification accretes: sweeps and volume-end passes move records from the baseline tier to leaf-verified continuously, with each note dated.

A further limit applies to the scans themselves: the Internet Archive surrogate does not in every case reproduce every leaf of Savage’s volume. Where a leaf is known to be absent from the scan it is recorded in the corrections file, but the catalogue cannot describe what was never photographed. None of this replaces the scan. Every record links to its manuscript page precisely so you can, and should, read Savage yourself before citing. This catalog is not affiliated with or endorsed by the Bancroft Library; the manuscripts are theirs (BANC MSS C-A), the digitizations are the Internet Archive’s, and the mistakes are mine.

Errata: readings corrected against the manuscript

Nothing here is trusted blind: not Savage, not the published editions, not this catalog. The register below records every reading so far corrected (or flagged) by re-reading the manuscript image; each entry links to the proving scan from its record card. It will grow.

See something wrong on a record? Every record page carries a “Suggest a correction” button; accepted corrections are published on the record itself, with credit.

Limits

Thematic tagging varies by volume and is a compiler’s working vocabulary, not a controlled thesaurus; each record carries the place it was written where the manuscript gives one (73% of records). Duplicate and cross-reference flags are candidates for verification, not confirmed. This is one scholar’s working reading of Savage’s catalogs; corrections are welcome.

Compiled by Aodhan Coyne ORCID 0009-0002-8630-3768 · MMXXVI
Corrections & questions: aodhancoyne@gmail.com · source, data downloads & issue tracker: github.com/Aodhanm/archives-of-california

Privacy. This site sets no cookies and runs no invasive tracking. If analytics are enabled they are GoatCounter, an open-source, no-cookie counter that records page views and referrers only, no personal data.