# FindOpera editorial policy How to settle the judgment calls that come up when editing the FindOpera catalog, so that edits are consistent across contributors. The preamble states the principles, the procedures (`P`) apply to every edit, and the rules (`R`) cover specific trade-offs. A number means the same thing forever. --- ## Preamble FindOpera records what is true about opera recordings, not what a release printed about itself. Where a booklet, sleeve or label listing is wrong, the catalog is corrected and the booklet's version remains in the record's history. Edits are cheap and reversible; silence is not. If you have good reason to believe something, make the edit and say what you relied on. Do not abstain merely because you are not certain. That is about correcting what is already here. It is not license to add a record, merge two, or align a value that a rule leaves alone. Leaving something as it stands, and saying what you found, is an outcome you can be right about. Every edit carries a justification (P3), citing by number the rule applied when one applies. If no rule covers your case, file feedback whose first line begins **no rule covered this**. That is a first-class use of the feedback tool, not a last resort. A rule that is unclear or wrong is worth feedback too. --- ## Procedures These are preconditions, not judgment calls. **P1. Read history before any delete, merge or correction of an existing value.** History is not an audit log you consult after the fact; it is a source. It may hold sourcing you do not have, or show that your intended edit was already tried and rejected for a cited reason. Read it and apply the policy to what you find. If your source is new, or stronger than what history considered, proceed and say in the justification what is new. If history already weighed your source and rejected it, skip the edit. A disagreement that history does not resolve is filed as feedback whose first line begins **contested edit**, rather than re-litigated by edit. **P2. Search before creating, and check that it belongs.** A singer, opera or character you cannot find under one spelling may exist under another, and creating a duplicate costs more to undo than searching costs to do. Being asked to catalog something is not a finding that it belongs in the catalog: the scope rules decide that. **P3. Never leave a justification empty or generic.** "Correcting error" is not a justification. Name the source you read, specifically enough that another editor knows which object you mean. The test is whether there was an artifact in front of you: a booklet you are holding is a source even where no one else can reach it, and a claim resting on memory is not one, however confident it feels. That holds for your knowledge of opera history exactly as it holds for anyone's. State your confidence when it is less than firm, since recording what you were unsure about is part of the edit rather than a reason to withhold it. --- ## How to use the rules Rules run in stages, and an earlier stage is a gate rather than a weight: a work ruled out of scope is not argued back in by a strong source, and the sourcing question is never reached. Only titles are given here: read a rule's full text before applying it. **Scope** decides whether the thing belongs in the catalog at all. - **R1** Opera, operetta and zarzuela are in scope, and a musical on its opera-house history - **R2** A recording is a complete performance, so a highlights disc is not one - **R3** A box set is several recordings, not one **Identity** decides whether two records are one thing, and which record a recording belongs to. - **R4** Merge only on identity, not similarity - **R5** Recording identity weighs the facts together, and one conflict need not defeat it - **R16** Default to attaching the standard version of an opera when multiple exist **Fields** decides what form a value takes. - **R6** One spelling per person, not one record per name - **R7** One canonical title per work, in the language its libretto was written in - **R8** One canonical name per character - **R9** Billing flags are editorial, and mark three singers rather than three portrayals - **R10** A recording's date is when it was recorded, with a flagged fallback - **R11** Orchestra and chorus names follow each recording's own source, not other recordings - **R17** A credited musical director goes in the conductor field **Sourcing** decides which evidence wins. R12 is the general case; R13 and R14 narrow it for one class of fact each. - **R12** Conflicts resolve by source strength: the source closest to the fact wins - **R13** Correct a performer credit only with a source about the performance - **R14** Correct a name only with a source about the person **Conduct** governs how far an edit reaches. - **R15** An edit improves the records it names, and files feedback on the rest --- ## Stage 1: Scope _What the catalog holds. These are gates. A question settled here is not reopened by a later stage._ ### R1. Opera, operetta and zarzuela are in scope, and a musical on its opera-house history, while a whole new category is parked rather than decided The catalog covers the operatic repertoire, understood to include operetta and zarzuela, which are stage works sharing the same performance tradition, performers and houses. A musical comes in on that same ground and no other: a substantial opera-house performance history, programmed as repertoire. A cast decides nothing either way. - _Candide_ (Bernstein) is **in**, on that history. - _West Side Story_ (Bernstein) is **in** on the same test, and the catalog holds it (opera 674). The 1984 Te Kanawa and Carreras recording is not what places it; that is a cast. - _Mamma Mia!_ is **out**. A commercial run is not a repertoire history. The boundary is genuinely unclear in places. For an individual work, placement is the editor's judgment: decide, say in the justification why you placed the work where you did, and file feedback if the case seems likely to recur. Do not stall an edit waiting for the boundary to be settled. But whether a whole _category_ is in scope, meaning a new medium, genre, or release type, is a scope decision rather than an editor's call: park it and file feedback so the decision is made deliberately and applied consistently. **Worked example.** _Candide_ and _West Side Story_ are two individual cases, each decided on its own history, and coming out the same way settles nothing about musicals as a class. The lip-synced opera feature films (Powell and Pressburger's 1951 _Hoffmann_, Pabst's 1931 _Dreigroschenoper_, the 1939 _Mikado_, Karajan's 1967 _Carmen_, Fracassi's 1953 _Aida_) are a category question, "are opera feature films in scope at all?", and stay parked until that scope decision is made. ### R2. A recording is a complete performance, so a highlights disc is not one The test is whether the work is present start to finish, not whether anything else shares the disc. **In.** A disc carrying a complete opera followed by unrelated arias is a release of that opera, and its UPC attaches normally. Filler material does not disqualify a release. State the boundary in the justification, naming which tracks are the opera and which are filler, so the next editor can check the claim without replaying the disc. **Out.** Highlights discs, under any circumstances. A highlights disc is not a recording of its own: its cast and date belong to the complete performance it was cut from. This holds even where no complete recording of those sessions was ever issued, so an orphan highlights disc is simply not recorded rather than added as a partial one. Do not attach a highlights UPC to the complete recording either, because the product is not that recording. Completeness is the whole test, so how much of the work an extract covers changes nothing. ### R3. A box set is several recordings, not one A box containing three complete operas is three recordings. Do not create a single record for the box, and do not attach one opera's cast to another's because they shipped in the same package. Each recording in the box keeps its own facts: its own date under R10, its own cast, its own conductor. These frequently differ across a box, since a label will collect sessions spanning a decade under one cover. Link the box's UPC to each recording in it. The UPC then lists all of them, and that list is where the catalog records that they were issued together. The box's title, catalog number and packaging date describe the product, not any recording in it, so none of them goes on a recording. --- ## Stage 2: Identity _Whether two records are one thing. Also gates: an identity question is settled before the surviving record's fields are argued over._ ### R4. Merge only on identity, not similarity Two records merge when they are the same thing, not when they look alike. Same name and same era is not identity. State in the justification what establishes identity: a shared external ID, a matching catalog number, documented biography. That is a bar to clear, not a box to fill: where nothing clears it, the records stay separate and you say what you looked for. Two people can share a name. The costs are not symmetrical, which is why the default is to leave them apart. A merge deletes one record and sends its id to the other, so a wrong one hides a real person behind somebody else's page, where nobody thinks to look. A duplicate left in place is still there for a later editor to find. A merge copies no fields. Before merging, copy to the survivor anything only the losing record holds, such as an external ID, a date or a link. Otherwise it survives only in the losing record's history. Applies to people, works, characters and recordings alike. For recordings, see R5. ### R5. Recording identity weighs the facts together, and one conflict need not defeat it Establish it from the performance's identifying facts, which are opera, date, conductor, principal cast, and label or catalog number, weighed together against provenance. No single element is sufficient on its own. Where one attribute conflicts, apply R12: if the conflict is plausibly an error in the weaker-sourced record, identity can still hold, but the justification must say which record is wrong and why. A conflict with no plausible explanation is not identity. **Worked example.** Two _I vespri siciliani_ records, same conductor (Santi) and cast, one dated 1974-11 and unsourced, the other dated 1975 and sourced to the discography documenting the performance. The date conflict is plausibly an error in the unsourced record: merged, with the conflict explained in the justification. _Related: R4, R12, P1._ ### R16. Default to attaching the standard version of an opera when multiple exist Some operas exist in more than one version: revised by the composer, or rewritten for a different house. Verdi's _Macbeth_ has its 1847 and 1865 forms, _Don Carlos_ its French and Italian ones, _Tannhäuser_ its Dresden and Paris texts. The catalog may hold a separate opera record for each, and usually does not. Where it does, a recording belongs to the record for the version it documents. Where it does not, that is a gap in the catalog and not an invitation to close it: leave the recording on the opera record it already has, say so, and file feedback under R15. Creating the version record, and the characters and portrayals that follow from it, settles a modeling question for the whole catalog from inside a single edit. Among the records that do exist, an explicit version mark, printed on the release, in the booklet or in a discography, routes the recording to that opera record. Absent a mark, and equally where the evidence is hybrid or conflicting, the recording routes to the version current performance tradition treats as standard. Nothing is parked on this question. A routed recording carrying a described conflict stays findable and is cheap to split out later; a parked one is a record nobody looks at again. The justification records which of the three branches applied and why, and for the conflicting case what the conflict was. **Worked example.** A _Macbeth_ release marked "1865 version" routes to the 1865 record where the catalog holds one. An unmarked _Macbeth_ routes to the standard 1865 text. A recording mixing 1847 and 1865 material also routes to the standard text, with the mixture described in the justification. _Related: R5, R12, R15, P1, P2._ --- ## Stage 3: Fields _What form a value takes. Each rule owns a field, so these rarely collide with each other._ Two of these point opposite ways on what looks like the same problem. A person gets one canonical name (R6); an orchestra keeps whatever the source printed (R11). The difference is not naming philosophy, it is that people are entities and ensembles are free text. Normalizing an ensemble name loses the historical form and buys nothing, because nothing joins on it. If ensembles become entities, R11 is the rule that changes. ### R6. One spelling per person, not one record per name A singer has exactly one first name and one last name in the catalog. Per-release spellings are not stored; they live in history. Prefer the form used by the person themselves, or failing that, the form dominant in current scholarship. Where the person used different forms at different points in their career, prefer the later form. _Diacritics are part of the name and are retained._ **Worked example.** Tenor Giorgio Lamberti (b. 1938) performed early as Giorgio Casellato, then as Giorgio Casellato-Lamberti, then as Giorgio Lamberti (documented name history; fr.wikipedia discography). The canonical record is **Giorgio Lamberti**, the later form. Variant spellings that neither the person nor scholarship uses ("Castellato" for "Casellato") are not candidates at all. _Related: R14, R4, R7, R8._ ### R7. One canonical title per work, in the language its libretto was written in A work has exactly one canonical title. Prefer the language the work was written in, meaning the language of the libretto the composer set. That outranks the language of the premiere, and it outranks how a general reference work files the piece, which R12 ranks as a finding aid rather than a source. Where the libretto language does not settle it, prefer the form used by the current critical edition and by commercial releases. `englishTitle` holds a genuine English translation or nothing. An English title character-for-character identical to the canonical title carries no information and is rejected at write time. **Worked example.** Offenbach's _Les Fées du Rhin_ against _Die Rheinnixen_. The record under the German title was merged into the record under the French. Nuitter's libretto is French, the Keck critical edition and both commercial releases use the French, and the surviving record already carried the librettist and language. The work premiered in Vienna in German and general reference works file it under the German, and neither outweighs the libretto. **Known loss.** One canonical title is lossy for a work carrying a second legitimate title. The merge above made "Die Rheinnixen" unfindable: a search for "Rhein" now returns only _Das Rheingold_, and `englishTitle` cannot hold the displaced string because that title is not English. The catalog has nowhere else to put it yet. Until it does, say in the justification which title you displaced, so the string survives in history and the merge is recoverable rather than lossy. _Related: R6, R4, R12, R8._ ### R8. One canonical name per character A character has one name, in the language of the work's canonical title (R7). Prefer the character's name over a title or epithet they are billed under. Per-release and per-translation variants are not stored; they live in history. **Worked example.** In _Don Carlos_, the characters "Posa" and "Marquis Posa" were merged into "Rodrigue". The work is French grand opera, so the French form governs under R7; and Rodrigue is the character's name where Posa is his title, Marquis of Posa, which is why both epithet forms gave way rather than one of them winning. **Note on strength.** The name-over-epithet preference is drawn from that one adjudicated case rather than from a body of practice. A character known overwhelmingly by an epithet may deserve an exception. File feedback if you meet one rather than deciding it silently. _Related: R7, R6, R4, P2._ ### R9. Billing flags are editorial, and mark three singers rather than three portrayals `noted` marks the three top-billed singers on a recording, or every singer when the recording has fewer than three of them. The editor chooses the three using best judgment from the information at hand: the release's billing order where it is known, otherwise cast-table order, role prominence, or singer notoriety. Where the choice was not straight billing order, the justification says what it rested on. The flag is carried by portrayals but read as a set of singers, and everything that consumes it takes the distinct singers. So a singer who doubles roles may be flagged on one of them or on all of them, and neither is a mistake worth an edit to correct. What must be three is the singers. ### R10. A recording's date is when it was recorded, with a flagged fallback Live recordings take the performance date. Studio recordings take the date recording was completed. Where neither is recoverable, fall back to the first release date, and say in the justification that the stored date is a release date rather than a recording date, so that it is not later mistaken for one. No field carries that distinction, so the justification is the only place it can live, and the next editor depends on your saying it. The fallback is to the _first_ release only. A reissue, remaster or streaming-service release date is never a recording date. This is the most common error available: a streaming page will happily show a 2015 release date for a 1962 recording, and taking it at face value misdates the record by decades. Where a candidate date postdates the singers' careers, or sits implausibly far from the rest of the cast's activity, treat it as a reissue date and look further. R12 ranks streaming and retailer metadata as public but unattributed, below anything that states its basis, for exactly this reason. ### R11. Orchestra and chorus names follow each recording's own source, not other recordings `orchestra` and `chorus` are free text, not canonical entities. Record the name as the source material gives it, in English or the group's native language, whichever the source uses. Do not translate, normalize, or align it with other recordings of the same group. Because the source records the name in use at the time, a group that later renamed itself keeps its historical name on older recordings. _Consequence: the same group may appear under several spellings, and these strings are not reliable for finding every recording by a group._ ### R17. A credited musical director goes in the conductor field `conductorId` is required, and some performances have no conductor: a chamber opera led from the piano, a Baroque opera directed from the harpsichord. Enter whoever the release credits with musical direction, and quote that credit in the justification. No column tells direction from conducting, so the justification is the only place it can live, the same way R10's release-date fallback depends on being said out loud. Do not leave the recording uncreated over it. Where nobody is credited with direction at all, file feedback rather than promoting a performer into the field. _Related: R10, R13, P3._ --- ## Stage 4: Sourcing _Which evidence wins when sources disagree. R12 gives the ranking. R13 and R14 narrow it for two classes of fact where the release looks like the proximate source and is not. Reached from inside another rule, not on its own._ ### R12. Conflicts resolve by source strength: the source closest to the fact wins The strongest source for a fact is the one closest to the fact's origin: the release for release facts, the person for name facts, the presenting house's archive for performance facts. Where proximity does not separate two claims, provenance does. Strongest first: 1. **Attributable.** An identifiable authority stands behind it: the label, the publisher, a named author, an institution. Liner notes, dedicated discographies and scholarship, and official and institutional pages sit here. Having no URL makes a source harder to reach, not unverifiable: anyone holding that release can check it, and anyone else can obtain one. 2. **Public but unattributed.** Findable again, with no stated basis: retailer and streaming metadata, a non-canonical site, a blog citing nothing. Reproducing it gets you no closer to the evidence. General reference works top this tier and are finding aids: follow one to its source rather than citing it. 3. **Uncheckable artifact.** A document or file you are reading that no one else can reach: a text file bundled with a rip, a private letter. A last resort. Say what it is and that no one else can check it. Below all three, and never sources at all, are memory, intuition and guess (P3). Two qualifications. First, R13 and R14 narrow this ranking for two classes of fact, and where they apply they say what counts as proximate. Second, provenance cuts both ways, so check history (P1) for what established the current value. A value resting on a weak or unrecorded source yields to a stronger challenger, but a lone weak source never overturns a well-sourced value. Cite both sources and the ranking in the justification. **Worked example.** A conductor record reads "Emilio Tieri"; a 1957 _Ballo in maschera_ booklet credits "Emidio Tieri". Read the record's history. If "Emilio" traces to a weak or unrecorded source, the contemporaneous booklet, being the proximate source, wins and the record is corrected to Emidio with both cited. If "Emilio" is well-sourced, the booklet spelling is recorded as a history note and the canonical name stands. _Related: R13, R14, R11, R7, P1._ ### R13. Correct a performer credit only with a source about the performance, not the release Where scholarship and a release's printed credits disagree about who sang, conducted or played, record what scholarship establishes, most often a reissue documented by a named researcher who corrected the credits the original printed. Cite it in the justification. The printed credit is preserved in history and is not separately recorded. This narrows R12 for one class of fact: proximity is measured to the performance, not to the release that printed the credit. A release is a report about a performance, and a report can be mistaken or motivated. _Applies to: substitutions concealed at the time, uncredited replacements in reissues, misattributions carried forward from earlier releases._ ### R14. Correct a name only with a source about the person, not about a release A booklet misspelling is evidence about that booklet. Changing a canonical name requires evidence about the person. This narrows R12 for one class of fact: proximity is measured to the person, not to the release that printed the name. **Where no source about the person exists**, three or more independent releases agreeing on a spelling stand in for one. This is the comprimario case: below a certain rank a singer is witnessed only by cast lists, and the strict reading leaves the catalog holding a spelling nothing prints. Independent means separate witnesses, so a reissue, a box collecting an earlier release and a licensed transfer are one witness between them. Name the releases you counted, and say what you searched for and did not find. Accents are asymmetric. Agreement that a diacritic is **absent** is weak, because stripping accents is a typesetting habit and a shared habit is not independent observation. Agreement that one is **present** is strong, since nobody adds an accent by accident. This does not reopen R13: a release still cannot settle who sang. _Related: R6, R12, R13, P3._ --- ## Stage 5: Conduct _How far an edit reaches. Cross-cutting, and reached on every edit rather than at a particular point._ ### R15. An edit improves the records it names, and files feedback on the rest Editors are expected to be active, not narrow. Adding or updating a recording means revisiting the records the edit directly names, which are its singers, conductor, opera and characters, and any discrepancy noticed there is investigated and, where appropriate, resolved. Noting a discrepancy in a justification and moving on is not sufficient: if a source contradicts a stored value, that contradiction is itself a finding. Investigate it, and either correct the record or say in the justification why the stored value stands. Two bounds keep this from expanding without limit. First, investigate cheaply: if a reasonable check does not resolve the discrepancy, file feedback rather than turning the edit into an open-ended investigation. Second, scope: discrepancies in records the edit directly names are investigated now; discrepancies noticed further afield are feedback filings, not same-session detours. In both cases the finding lands where it will be seen, instead of surviving only inside the prose of an unrelated edit. This rule pushes against the minimal-diff instinct deliberately. The catalog's cost model is asymmetric: an editor already looking at a record is far cheaper than a future editor rediscovering the same problem, and most discrepancies are never seen twice. (R14 still governs throughout. "The release spells it differently" is a reason to investigate, not by itself a reason to change.) **Worked example.** The Tieri discrepancy in R12 reached the catalog as a passing note in an edit about a barcode. The conductor is a record that edit directly names, so it is investigated now: R6 frames it, R14 sets the bar, R12 and P1 decide it, and R4 governs the merge, since a second conductor record exists under the other spelling. Stopping at "noted in passing" skips all four rules. _Related: R6, R14, R4, R12, P1, P3._ --- ## Citing rules Cite the rule you applied by number, as "applied R12", in the edit justification. A number means the same thing forever, so the log of justifications stays readable against a fixed vocabulary. That permanence is the reason for the rules below. It also means the numbers cannot be tidied later. - **A number is never reused or redefined.** New rules take the next free number. Nothing is renumbered. - **Rules retire by being superseded**, with a pointer to the replacement, rather than being deleted. A reader of an old justification can always resolve the number they find. - **Procedures work the same way.** `P` numbers are cited exactly as `R` numbers are and carry the same guarantee. - **The preamble is versioned whole** rather than per-paragraph, since justifications do not cite it. Superseded versions stay retrievable, so a reviewer can tell which principles an old edit was made under.