corbett-heights

Imported from nolastan/know-this-place (san-francisco/corbett-heights/AGENTS.md). Install upstream with npx skills add nolastan/know-this-place --skill corbett-heights. Copyright stays with the author.

Corbett Heights — neighborhood guidance

Applies to everything under san-francisco/corbett-heights/, in addition to the root AGENTS.md.

Scope

Corbett Heights is the hillside neighborhood around Corbett Avenue, between Market Street and Twin Peaks, north-west of the Castro. Its streets include Corbett Avenue, Ord, Hattie, Danvers, Mars, Romain, Levant, Museum Way and the upper reaches of Clayton and Ashbury.

It is filed separately from the Castro even though the city's addressing data labels both Castro/Upper Market. Local sources — the Corbett Heights Neighbors association and Michael Corbett's 2017 Corbett Heights: San Francisco, An Historic Context Statement — treat it as a distinct neighborhood with its own subdivisions and development history, and its streets don't overlap the Castro tree.

• The .sub locality line is Corbett Heights · San Francisco, CA <zip>.
• If a source calls a Corbett Heights address "Castro/Upper Market," that's the city's analysis-neighborhood label, not a reason to refile the page.
• The separation is enforced in one place: AREA_EXCLUDE_STREETS in seed_pages.py keeps seed castro off these streets, so a parcel here is never filed twice. Seeding new parcels on them means seeding --area corbett-heights and pointing it at the same analysis neighborhood.

Three pages here are filed under the address the building is known by rather than the lowest number EAS holds for the parcel, which is what the seeder would pick: clayton-street/1349 (EAS also has 5 Villa Terrace), corbett-avenue/127 (EAS has only 9 Hattie Street) and museum-way/199, the Randall Museum (EAS also has 122 States Street). Each says so in also_addressed. Don't "correct" them to the mechanical answer.

Presenting history from newsletters and context statements

The sources themselves — the Corbett Heights Neighbors newsletter archive and the 2017 Context Statement — are cataloged in the research module: corbett-heights-neighbors and sf-context-statements. Go there for access notes, coverage and the cautions learned so far; what follows is how to put what they say onto a page.

Local history writing is a rich source, but it is prose written by someone else. Two hard rules when using it:

• Extract facts; never paraphrase the source's prose. A dated event becomes a .vtl timeline item; a discrete fact (contractor, architect, build cost, first owner) becomes a .speclist row or a stat tile; a category becomes building.style or building.subdivision, which are .speclist rows too. Do not reproduce the source's sentences or their structure — that reads as an article and risks copying someone else's expression. Facts are not copyrightable; wording is.
• Never name the newsletter in the page body. "The newsletter says…" is meaningless to a reader with no context for it. The citation goes in the Sources footer only, naming the specific issue (title, volume, number, date) and linking it, and repeating any original source the newsletter itself names (a permit date, a census year, a dated newspaper item).

Orientation (context, not citable facts)

Verify before asserting any of this about a specific building.

• The land was subdivided in stages. The McKee Subdivision (forty numbered lots) and the Park Lane Tract (from 1885) both recur in the record. McKee's central street, Clara, was renamed Ord in 1909; Hattie Street was named for his daughter.

• Street numbers changed in 1909, and some streets were renamed. An address given in a pre-1909 source often does not correspond to today's number. Always check today's number against the EAS dataset before creating a page. The pairs the record states for itself, so far:

  Then	Now
  1200 Ashbury	1200 Clayton Street
  34 Clara Avenue	18 Ord Street
  56 Clara Avenue	80 Ord Street
  77 Clara Avenue	81 Ord Street
  110 Clara Avenue	110 Ord — no such address today
  104–106–108 Clara	104–106–108 Ord Street

• Hattie Street was renumbered separately, between 1905 and 1913–14, and not in 1909: 7 became 17, 11 became 25–27, 15–15½ became 29–31, and 17 became 35. A pre-1913 Hattie number is a different building from the one that carries it now — the 1897 chimney fire "at 17 Hattie" is on the parcel the city calls 35 Hattie today.

• Two stairways were streets first. Jupiter Street became the Vulcan Stairway, Serpentine Place became Saturn Street and then the Saturn Stairway, and Mono Street is now Al's Park. A number on any of them is a number on ground the city took: none of the three can carry a page.

• Much of the housing is early-20th-century, with post-war infill on the upper slopes.

The assessor's year built disagrees with the photographs — say so

The neighborhood's history writing dates buildings from photographs, and the assessor's year_property_built frequently contradicts them. Both are on the record; neither gets quietly dropped. Put the assessor's date in the Built … tag (it is the structured field the rest of the page is built from), put the photograph in the timeline, and name the disagreement plainly in unknowns, which renders on the line closing that timeline. Do not average them, pick a favourite, or write "about". Worked examples in this tree:

Round years (1900, 1890) in the assessor roll are placeholders as often as they are facts — treat them as the weaker evidence, and say in unknowns that the date has not been checked against a permit.

Buildings here were moved

A construction date does not always belong to the site. 11–23 Mars Street was moved to Mars from Falcon Street — the assessor's 1909 is the date it was built somewhere else, and photographs of about 1921 show its Mars lot still empty. When a source says a building "moved here from" somewhere, that is a Relocated building tag plus separate timeline entries for the construction date and the move; never fold them into one "built" claim.

Historical addresses that no longer exist

The neighborhood's history writing frequently discusses buildings that have been demolished or addresses that were renumbered out of existence. Do not create a page for an address that isn't in the EAS dataset — record the history on the surviving building's page or the street hub instead. Two examples found while seeding:

• 110 Ord Street — the Cassin cottage. Ord runs 7–91 today; no 110 exists.
• 1200 Clayton Street — the 1909 building was replaced; the address is now thirteen condominium parcels, and EAS carries 1200 CLAYTON ST with no parcel number at all. It is the most documented corner in the newsletter's whole archive — a 1906 contract, a 1909 permit by Rainey and Phillips, four alterations, the Georgian Apartments, demolition before 1950, a filling station of 1953 and the condominiums of 1985 — and none of it can go on a page. Eight findings sit unresolved for it in research/findings/corbett-heights-neighbors/issues-2026-02-to-2026-07.json.
• Anything on Falcon Street — Falcon was expunged by the Market Street extension. Fred G. Horner's grocery and saloon at 2 Falcon Street recurs in the record; its story belongs on the surviving buildings nearby (danvers-street/56/, danvers-street/60/) and the Danvers Street hub.
• 2 Mars Street — appears in a 1925 photograph, but no such address exists in EAS today; Mars begins at 4. The site is the corner parcel 2654001, which the city addresses 4465 17th Street: the Belle-V Apartments of 1961, whose build year on the roll matches the one the newsletter gives. The newsletter itself hedges the corner as "2 (?) Mars or 4465 17th".
• The 3000 block of Market Street was Merritt Street. Upper Market Street was cut through in the 1920s and Merritt was eliminated; the north side of the block, 3000 to 3094, carries the numbers that replaced it. Two pairs are stated in the record: 2 Merritt is 3000–3002 Market (the corner at Hattie, built 1890) and 4 Merritt is 3004 Market — except that no 3004 exists in EAS, which runs 3000, 3006, 3008, 3012, so that one resolves to nothing.

Dated events go in historical_record; standing facts go on building

Several pages in this tree once carried a building_history object instead — a dated events list plus scalars (architect, contractor, first_owner, estimated_cost_usd, relocated_from, conflict...) the flat historical_record shape can't hold. Issue #148 migrated all of them: the events are historical_record entries now, and the scalars are building.architect/building.builder/building.first_owner/ building.cost_usd/building.relocated_from/building.conflict (with architect_note/builder_note riding beside the field they qualify — see with_note in seed_pages.py). Don't reintroduce building_history: a dated fact from a newsletter or context statement goes in historical_record, and a standing fact about the building goes on building, the same as everywhere else in the corpus.

parcel is the assessor's roll and nothing else. These pages were written before the seeder was, and carried six keys under parcel that no other page in the corpus has and the renderer never read — the assessed values and sale date (they belong in assessment, which is what draws the value chart and the "Last sale" row), a spelled-out construction_type (the renderer maps construction_type_code), and style, subdivision and a rear unit's number. Those three are now building.style, building.subdivision and an also_addressed entry. A fact the assessor doesn't state doesn't go in parcel, however much it looks like parcel data.

Contributed memoirs: take the building, leave the people

Several issues are first-person accounts by living residents (the Shiloh's Way piece in January, the Corbett Avenue childhood memoir in July). They are good sources for building facts — a contractor's name, a stair, a rebuild — and are not licence to name the writer, their family, or who lives where. Attribute them in the footer as an account published in the issue, not by name. Deceased figures already published with dates (first owners, builders, nineteenth-century grocers) may be named, as elsewhere.