Runs
runsrelease v0.7.1CC0 1.0
About
What made each table: rules, hand work, or a model run (model, prompt).
- How far to trust it
- The release passed its automatic checks before it was published (what is checked). A hand check by the editor of qipu data is under way: 36 of 38 items checked so far were right, on votes, money, places and the map. A row says what the record states; the scanned page is the authority. Details
- Licence
- CC0 1.0: use it for anything, no permission needed. The scans and other sources keep their own rights.
- Version
- v0.7.1, published 21 October 2026. A published release is never changed; corrections come as a new release.
Technical details
- Table
- runs
- Rows
- 43 (San Francisco 42; Vancouver 1)
- Columns
- 15
- Parquet
- runs.parquet, < 0.1 MB
- Version
- v0.7.1, 21 October 2026
- Licence
- CC0 1.0. Sources keep their own rights.
Rows
Rows by kind
| Value | Rows | Share | |
|---|---|---|---|
| llm | 15 | 34.9% | |
| rules | 9 | 20.9% | |
| record | 5 | 11.6% | |
| llm+hand | 4 | 9.3% | |
| hand | 3 | 7.0% | |
| search | 2 | 4.7% | |
| model | 2 | 4.7% | |
| llm+rules | 1 | 2.3% | |
| model+hand | 1 | 2.3% | |
| rules+hand | 1 | 2.3% |
Rows by model
| Value | Rows | Share | |
|---|---|---|---|
| (empty) | 26 | 60.5% | |
| claude-haiku-5-5 | 15 | 34.9% | |
| claude-sonnet-5-5 | 2 | 4.7% |
Rows by prompt_version
Method and accuracy
No accuracy measure for this table yet.
The release passed its automatic checks before it was published (what is checked). A hand check by the editor of qipu data is under way: 36 of 38 items checked so far were right, on votes, money, places and the map. A row says what the record states; the scanned page is the authority.
Columns 15
| Column | Type | Can be empty | Privacy | Public copy | Description |
|---|---|---|---|---|---|
| run_id | VARCHAR | no | none | as is | The run's id, as used in run_id elsewhere. |
| kind | VARCHAR | no | none | as is | How the run worked: llm, rules, llm+hand, llm+rules, rules+hand, hand, search.Values hand, llm, llm+hand, llm+rules, rules, rules+hand, search |
| model | VARCHAR | yes | none | as is | Model used (claude-haiku-5-5, claude-sonnet-5-5); empty for rule, hand and search runs.Values claude-haiku-5-5, claude-sonnet-5-5 |
| prompt_version | VARCHAR | yes | none | as is | Prompt file or script the run used. Values city_run.py, decisions_bench/v2/v2_prompt.py (schema v2.1), ia_records/scripts/dl_tag.py, ia_records/scripts/sfra_extract.py, street_decisions PROMPT.md / UNPLACED_PROMPT.md / IV_PROMPT.md, street_decisions/RESWEEP_PROMPT.md |
| schema_version | VARCHAR | yes | none | as is | Output schema version (1, v2.1) where set.Values 1, v2.1 |
| settings | VARCHAR | yes | none | as is | Empty in v0.4. |
| started | VARCHAR | yes | none | as is | When the run started (YYYY-MM-DD or YYYY-MM). Values 2026-07-28, 2026-09-02, 2026-10, 2026-10-01, 2026-10-06, 2026-10-12, 2026-10-13, 2026-10-14 |
| pages | INTEGER | yes | none | as is | Empty in v0.4. |
| cost_usd | DOUBLE | yes | none | as is | Empty in v0.4. |
| gold_scores | VARCHAR | yes | none | as is | Empty in v0.4. |
| notes | VARCHAR | yes | none | as is | What the run made, in a line; model runs with no gold end with [unmeasured: no gold for this run].Values Board items the location gate dropped, re-swept (resweep.py) and read one batch of 60 at a time [unmeasured: no gold for this run], Democracy's Library (IGS) SF items: the hoods, streets and places each is about, read from metadata and the start of the OCR [unmeasured: no gold for this run], Redevelopment Agency property files (Western Addition A-2), one row per parcel page; matched to EAS by sfra_build.py, blocks/build_index.py: decision -> assessor blocks (tiers street/corner/segment/addr/stated); accuracy.json: 88.9% exact on 3,195 checks, decisions v2 for Vancouver council minutes (decisions_runs/van_v21_haiku), decisions v2: one call per page, previous page as context, Message Batches (decisions_runs/sf_v21_haiku), open-public-data pipeline/minutes extract.py (location gate) + blocks/build_core_table.py classification (form/rule/title rules, some LLM labels), scripts/11_stage_extra.py: every photo, film, recording, clip and document the map shows, with credit, licence and a rights note, scripts/20_rollcalls.py: printed roll calls -> votes; names matched to the Wikipedia roster, scripts/30_build.py: "<number> <street>" texts matched exactly to DataSF EAS addresses, scripts/80_showcase_fetch.py: archive.org advancedsearch (title/subject/description), Open Library search and Wayback CDX for the six SF showcase places; metadata only, no model, site rows (su:/sd:) of the map: plain line, outcome, importance, accuracy fixes [unmeasured: no gold for this run] ... |
| gold_id | VARCHAR | yes | none | as is | The gold set the run was scored against (see field_scores); empty when unmeasured.Values decisions_bench Vancouver gold (30 pages, 48 decisions: Opus-written, 21 decisions hand-checked), decisions_bench/gold.jsonl (SF, 40 pages: an Opus-5.5-written key, no record of a hand check) + v2 Opus key, decisions_bench/gold.jsonl roll calls (Opus-written key), scripts/22_rollcall_gold.py, ia_records/sfra/bench (18 pages) |
| prompt_id | VARCHAR | yes | none | as is | Same as prompt_version.Values city_run.py, decisions_bench/v2/v2_prompt.py (schema v2.1), ia_records/scripts/dl_tag.py, ia_records/scripts/sfra_extract.py, street_decisions PROMPT.md / UNPLACED_PROMPT.md / IV_PROMPT.md, street_decisions/RESWEEP_PROMPT.md |
| extraction_table | VARCHAR | yes | none | as is | Empty in v0.4. |
| city | VARCHAR | no | none | as is | sf or van.Values sf, van |
Privacy: what kind of text the column holds. Public copy: what the published file does with it; the embargo covers records after 1951.
Preview
| run_id | kind | model | prompt_version | schema_version | settings | started | pages |
|---|---|---|---|---|---|---|---|
| site-street-layer-2026-10 | llm+rules | claude-haiku-5-5 | street_decisions PROMPT.md / UNPLACED_PROMPT.md / IV_PROMP… | — | — | 2026-10-13 | — |
| sf-v21-haiku-1014 | llm | claude-haiku-5-5 | decisions_bench/v2/v2_prompt.py (schema v2.1) | v2.1 | — | 2026-10-14 | — |
| van-v21-haiku-1014 | llm | claude-haiku-5-5 | decisions_bench/v2/v2_prompt.py (schema v2.1) | v2.1 | — | 2026-10-14 | — |
| sfra-appraisa…iku55-2026-10 | llm | claude-haiku-5-5 | ia_records/scripts/sfra_extract.py | — | — | 2026-10-14 | — |
| board-permits-haiku55-2026-10 | llm | claude-haiku-5-5 | enrich_1410/permits/prompt.md | — | — | 2026-10-14 | — |
| census-tracts-2026-10 | record | — | — | — | — | 2026-10-14 | — |
| hubs-v061 | rules | — | — | 1 | — | 2026-10-16 | — |
| hand-v061 | hand | — | — | 1 | — | 2026-10-17 | — |
8 of 15 columns shown. Eight rows sampled at build time. Hover a shortened value for all of it.
Download
| Format | File | Size |
|---|---|---|
| Parquet | runs.parquet ↗ | < 0.1 MB |
| SQLite (all tables) | qipu_sf.db ↗ | 6.44 GB |
SQL
SELECT *
FROM runs
LIMIT 100
Run in the SQL console (DuckDB in the browser, over the Parquet file).
Python
import duckdb
duckdb.sql("SELECT * FROM 'https://tiles.qipu.org/data/releases/v0.7.1/parquet/runs.parquet' LIMIT 5").show()