Behind the data

How the archive behind the rankings was built, how much of each regatta it covers, which consistency checks it fails and where, and the assumptions it rests on. The methodology page explains how ratings are computed from it.

From web pages to results

Results are collected only from original sources: organizers, timing vendors, governing bodies and clubs. 1,615 URLs were found; 607 were catalogued as results pages, of which 589 could be fetched and 14 could not (4 skipped as out of scope). They yielded 539 distinct files (74 were copies of another file), and 477 of those parsed into 18,625 races and 100,595 crew results, 97,707 of them with a finish time.

3 of the results files were downloaded by hand from pages the fetcher cannot reach (live-results sheets, file viewers) and are recorded with the page they came from; they hold 453 crew results and appear as "manual" in the diagram.

Left to right: how each URL was found, what the catalogue made of it, whether it was fetched, which parser read the file, and the coverage status of the edition it belongs to. A fetched sheet can hold several tabs and a file can be found at two URLs, so 589 fetched URLs became 613 files. Hover a bar or a flow for its count.

Funnel table (85 flows)

Every flow of the diagram, from reports/funnel.json
From	To	Count
Parser family: alkame	Edition status: parsed	57
Parser family: csv blocks	Edition status: parsed	20
Parser family: dbc htm	Edition status: parsed	9
Parser family: csv cards	Edition status: parsed	4
Parser family: event blocks	Edition status: parsed	13
Parser family: dbbc grid	Edition status: parsed	103
Parser family: team place html	Edition status: parsed	3
Parser family: jy html	Edition status: parsed	13
Parser family: lane grid	Edition status: parsed	5
Parser family: lane grid	Edition status: partial	5
Parser family: ordered list	Edition status: parsed	80
Parser family: ordered list	Edition status: partial	2
Parser family: placings sheet	Edition status: parsed	25
Parser family: club pdf heats	Edition status: parsed	40
Parser family: dbc pdf	Edition status: parsed	7
Parser family: gwn pdf	Edition status: parsed	50
Parser family: gwn pdf	Edition status: partial	3
Parser family: race cards	Edition status: parsed	12
Parser family: race rows	Edition status: parsed	1
Parser family: racegrid html	Edition status: parsed	12
Parser family: results of racing	Edition status: parsed	1
Parser family: standings	Edition status: found	1
Parser family: standings	Edition status: partial	2
Parser family: alkame	Edition status: parsed	4
Parser family: crash	Edition status: found	3
Parser family: not results	Edition status: parsed	12
Parser family: ocr low quality	Edition status: ocr low quality	2
Parser family: ocr needed	Edition status: found	1
Parser family: unknown	Edition status: found	31
Parser family: unknown	Edition status: partial	6
Parser family: titled list	Edition status: parsed	6
Parser family: titled list	Edition status: partial	2
Parser family: two lane	Edition status: parsed	3
Parser family: two lane	Edition status: partial	1
Fetched: ok	Parser family: alkame	57
Fetched: ok	Parser family: csv blocks	20
Fetched: ok	Parser family: dbc htm	9
Fetched: ok	Parser family: csv cards	4
Fetched: ok	Parser family: duplicate	74
Fetched: ok	Parser family: event blocks	13
Fetched: ok	Parser family: dbbc grid	103
Fetched: ok	Parser family: team place html	3
Fetched: ok	Parser family: jy html	13
Fetched: ok	Parser family: lane grid	10
Fetched: ok	Parser family: ordered list	82
Fetched: ok	Parser family: placings sheet	25
Fetched: ok	Parser family: club pdf heats	40
Fetched: ok	Parser family: dbc pdf	7
Fetched: ok	Parser family: gwn pdf	53
Fetched: ok	Parser family: race cards	12
Fetched: ok	Parser family: race rows	1
Fetched: ok	Parser family: racegrid html	12
Fetched: ok	Parser family: results of racing	1
Fetched: ok	Parser family: standings	3
Fetched: ok	Parser family: alkame	4
Fetched: ok	Parser family: crash	3
Fetched: ok	Parser family: not results	12
Fetched: ok	Parser family: ocr low quality	2
Fetched: ok	Parser family: ocr needed	1
Fetched: ok	Parser family: unknown	37
Fetched: ok	Parser family: titled list	8
Fetched: ok	Parser family: two lane	4
Catalogued as: results	Fetched: http 4xx	14
Catalogued as: results	Fetched: ok	589
Catalogued as: results	Fetched: skipped	4
Found by: catalog override	Catalogued as: results	3
Found by: catalog override	Catalogued as: review	1
Found by: manual	Catalogued as: results	3
Found by: seed crawl	Catalogued as: not results	52
Found by: seed crawl	Catalogued as: results	243
Found by: seed crawl offsite	Catalogued as: index	88
Found by: seed crawl offsite	Catalogued as: not results	75
Found by: seed crawl offsite	Catalogued as: results	11
Found by: seed crawl page	Catalogued as: index	95
Found by: seed crawl page	Catalogued as: not results	77
Found by: seed crawl page	Catalogued as: results	108
Found by: seed crawl page	Catalogued as: review	16
Found by: seed known	Catalogued as: results	7
Found by: wayback cdx	Catalogued as: index	64
Found by: wayback cdx	Catalogued as: not results	364
Found by: wayback cdx	Catalogued as: results	144
Found by: wayback cdx	Catalogued as: review	147
Found by: wp media	Catalogued as: not results	28
Found by: wp media	Catalogued as: results	88
Found by: wp media	Catalogued as: review	1

Coverage by event and year

The coverage register lists every edition of every event the project knows of, with how far its results got: 331 of 566 editions across 62 events are parsed. Events are sorted by parsed editions. A blank cell is a year with no known edition; hover or focus a cell for its detail.

PA: parsed: results in the archive (331)
PT: partial: some of the results parsed (6)
FD: found: results located, not parsed yet (30)
SU: summary only: no race-level results published (1)
OC: scanned, OCR too poor to parse (2)
NF: searched, no results found (89)
NS: not yet searched (105)
NH: not held (2)

566 editions of 62 events, 1989–2026
Event	1989	1990	1991	1992	1993	1994	1995	1996	1997	1998	1999	2000	2001	2002	2003	2004	2005	2006	2007	2008	2009	2010	2011	2012	2013	2014	2015	2016	2017	2018	2019	2020	2021	2022	2023	2024	2025	2026	no year
concord-pacific	NF	NF	NF	NF	NF	NF	NF	NF	NF	NF	NF	NF	NF	NF	PA	PA	PA	PA	PA	PA	PA	PA	PA	PA	PA	PA	PA	PA	PA	PA	PA	NS	PA	PA	PA	PA	PA	PA
ottawa-dbf																PA	PA	PT	PA	PA	PA	PA	NS	PA	PA	PA	PA	PA	PA	PA	PA	NS	NS	PA	PA	PA	PA	PA
tidbrf	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	PA	NF	PA	PA	PA	PA	NF	PA	PA	PA	PA	PA	PA	NF	PA	PA	NS	NS	PA	PA	PA	PA	PA
halifax-manulife												PA	PA	PA			PA	PA	PA	PA	PA		PA	PA	PA	PA	PA	PA	PA	PA	PA
penticton												NS	NS	NS	NS	NS	NS	NS	NF	PA	PA	PA	PA	PA	PA	PA	PA	PA	PA	PA	PA	NS	NS	PA	PA	PA	PA	PA
dbc-nationals																			PA	PA	PA	PA	PA	PA	PA	PA	PA	PA	PA	PA	PA			PA	PA		PA
pickering																		PA	NF		PA	PA	PA	PA	PA	PA	NF	PA	PA	PA	PA			PA	PA	PA	PA	PA
lachine																				PA	PA		PA	PA	PA	PA	PA	PA	PA	PA	PA				PA	PA
coupe-du-quebec																				PA	PA	PA	PA	PA	PA				PA	PA	PA				PA	PA	PA
pickering-youth																					PA	PA	PA	PA	PA		NF	PA	PA	PA	PA			PA	PA
dz-500																									PA			PA	PA	PA	PA			PA	PA	PA	PA	PA
dz-youth-regatta																									PA		PA		PA	PA	PA			PA	PA	PA	PA	PA
richmond-steveston																									PA	PA	PA	PA	PA	PA	PA			PA	PA	PA
calgary														FD	NS	FD	FD	PA	NF	NF	NF	NF	NF	NF	NF	NF	NF	NF	PA	PA	PA	NS	NS	PA	PA	PA	PA	PA
montreal-challenge																								PA	PA	PA	PA	PA	PA	PA					PA	PA
gwn-challenge								NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NF	NF	PA	PA	NF	NF	NF	NF	NF	NF	PA	PA	NF	NS	NS	PA	PA	NF	PA	PA
gwn-chicago																								PA		PA	PA		PA						PA	PA	PA	PA
gwn-mercer																				PA				FD	FD	FD	PA		PA	PA					PA	PA	PA	PA
dz-spring-sprint																									PA		PA		PA	PA					FD	PA	PA	PA
christie-lake																															PA			PA	PA	PA	PA	PA
gwn-sarasota																												PA	PA	PA					PA	FD	PA	PA
gwn-sport-regatta																					PA								PA		PA				PA	PA	PA
hamilton-waterfest																														FD	PA			PA	PA	PA	PA	PA
milton																														NS	PA			PA	PA	PA	PA	PA
port-perry																															PA			PA	PA	PA	PA	PA
dz-fall-classic																																		PA	PA	PA	PA	PA
gwn-orlando																					PT	PA									PA				PA	PA	PA
edmonton																				NS	PA	NF	NF	NF	NF	NF	NF	NF	NF	SU	NF	NF	NF	NF	NF	PA	PA	PA
gwn-ridbf																										PA				PA						PA		PA
hope-floats																																			PA	PA	PA	PA
regina																									PA	PT	PA	FD	PA	PT				PA	OC	FD
h2o-open																														PA				PA		PA
pan-am-ccc																											PA		PA		PA						NS
victoria																PA	PA	PA	NF	NS	NF	NF	NF	NF	NF	NF	NF	NF	NF	NF	NF	NF	NF	NF	NF	NF	NF	NF
can-am-ccc																																				PA		PA
gwn-kcdbf																																				PA	PA
gwn-nyc																																					PA	PA
lethbridge																																				NF	PA	PA
okanagan-super-sprints																				FD	NS	FD	FD	NS	FD	FD	FD	FD	FD	FD	NS	NS	NS	NS	NS	FD	PA	PA
vernon																							NS	PA	PA	FD	NS	FD	NF	NF	FD	NS	NS	FD	FD	NF	NS	NS
dz-oddball-2k																																		PA
harrison																	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NS	NF	NF	NF	PA
kamloops																																					PA
kelowna																													PA							OC	PT
osoyoos-flag-race																																					PA
ottawa-ice-dragon																																PA
winnipeg-september																																			PT	PA		NS
alberta-long-tail																																							NS
atlantic-long-tail																																							NS
barrie																																			NS
dbc-race-series																																							NS
gwn-other																																							NS
idbf-champs																																							NS
manitoba-school																																				NS
montreal-festival-historic																																							NS
ontario-long-tail																																							NS
shuswap																								FD	NF	FD	FD	FD	NH	NH	NF	NF	NF	NF	NF	NF	NF	NF
sudbury																								NS	NS	NS	NS	NS	NS		NS
sydenham-challenge																																				NS	NS	NS
triple-crown-qc																																							NS
winnipeg																																			NS	NS
woodstock																																							NS

Consistency checks

Every build counts the rows that break one of these invariants. Each count has a ceiling it may not rise above without a logged decision, so the archive can only get cleaner. Races that fail I3, I6 or I12 are kept on their event page but left out of the ratings (see which races count).

16 invariants, from reports/quality.md
ID	Violation	Rows	Editions	Top editions
I1_time_in_name	team_name holds a finish time anywhere (H:MM:SS / M:SS.ff / M:SS.fff)	0	0	-
I2_rule_as_crew	team_key is progression or heading text (fastest, from race, advances, ...) or a bare division word	25	5	gwn-sarasota-2023 (8), dbc-nationals-2007 (7), pickering-2013 (5)
I3_empty_result	race_level entry with no time, no place and nothing printed in the time column (status NULL, or NO_RESULT from a blank time_raw)	1,409	66	concord-pacific-2023 (254), concord-pacific-2025 (214), coupe-du-quebec-2024 (167)
I4_edition_id	edition_id is not <event_id>-<4-digit year>, or event_id itself contains a 4-digit year	0	0	-
I5_time_range	time_ms outside the plausible band for the race distance (200 m 0:30-4:00, 250 m 0:35-5:00, 500 m 1:20-7:00, 1000 m 3:00-12:00, 2000 m 7:00-25:00)	106	27	gwn-challenge-2026 (16), dz-oddball-2k-2022 (11), montreal-challenge-2024 (11)
I6_place_order	within a race, a place repeated more than twice, or a place greater than the number of entries	1,412	58	ottawa-dbf-2006 (157), ottawa-dbf-2005 (140), ottawa-dbf-2018 (120)
I7_blank_name	team_key blank or NULL	21	9	montreal-challenge-2013 (7), dbc-nationals-2015 (4), gwn-orlando-2019 (3)
I8_footnote_in_name	team_raw holds penalty/bonus footnote text (penalty, bonus for, impeded, interference, second credit, ...)	2	1	ottawa-dbf-2014 (2)
I9_doubled_glyphs	team_name holds fake-bold doubled glyphs (three doubled letters in a row, '::', or a time with 4+ decimals)	0	0	-
I10_glued_text	team_name over 70 characters or holding instruction text (determines N, start first, minute gap, +N sec, Race N:, turn left/right, merge onto, Rang Équipe, ...)	1	1	halifax-manulife-2009 (1)
I11_club_wrap	club cell is a bare continuation word (Club, Centre, ...), or the team name ends with the first words of a club printed in the same edition (a two-line club read into the crew name)	5	2	tidbrf-2004 (3), dbc-nationals-2018 (2)
I12_split_places	race not marked combined in which a printed place is shared by finishers whose times differ by more than 0.5 s (divisions ranked separately inside one race)	320	65	can-am-ccc-2024 (50), dbc-nationals-2014 (32), gwn-mercer-2017 (29)
I13_short_key_span	short crew key (digits, or 1-2 letters and at most one digit) whose team_id spans more than one province (or, for events without a single province, more than one edition)	0	0	-
I14_bib_prefix_split	in a bib-printing file, team_name '<token> <rest>' whose <rest> is printed with the same bib in the same edition under another team_id (a division/progression code or seed read into the name)	13	2	can-am-ccc-2026 (8), gwn-challenge-2010 (5)
I15_place_time_order	timed, placed race (3+ entries, not combined) in which more than n // 2 pairs of crews have their place order reversed against their time order (places read from another column)	155	58	dbc-nationals-2014 (21), can-am-ccc-2024 (17), gwn-mercer-2017 (9)
I16_team_twice_in_race	entry whose team_id holds more than one entry in its race, any status (two crews merged into one team, a heading or same-name crew keyed alike, or a row the parser duplicated); C57, D80	41	15	lachine-2014 (6), ottawa-dbf-2024 (4), richmond-steveston-2023 (4)

Assumptions

Values the data relies on that no source prints, each with its basis, where it applies, how confident the project is and what would make it revisit the assumption. Process decisions are logged separately and are not listed here.

23 assumptions, from ASSUMPTIONS.md
ID	Assumption	Basis	Applies to	Confidence	Revisit if	Status	Source
A1	The Ottawa live Google Sheets with no printed year are 2026 results.	The Ottawa page already announces the 2027 dates; the raw snapshot is dated 2026-11-07.	`ottawa` 2026 published-sheet tabs (catalog year 2026).	high	The sheets are overwritten in 2027 (keep the 2026-11-07 snapshot).	active	owner 2026-11-07
A2	Regina crews printed only as entry codes (`AMR 1`, `S101`, `W3`) are kept as event-scoped identities: they race within their own edition, never merge with a team in another edition, and a code is never guessed to be a team name.	No code-to-name key exists on reginadragonboat.com or in Wayback copies. Safe; loses cross-event linkage for those crews.	`regina` 2013, 2014 (sprint and heats), 2017 (`lane_grid`, D15). Implemented in T6.	high	Team lists are obtained from the organizer (keep on the outreach list).	active	owner 2026-11-07
A3	The Lachine course is 200 m.	Winning times of 0:48–0:58 match 200 m standard-boat times. No file prints the distance.	All `lachine` races 2008–2024 with no printed distance: `default_distance_m: 200` in `sources.yaml`, `distance_inferred = TRUE`.	medium	Mission Dragon Boat or another source states a different course length.	active	owner 2026-11-07
A4	Halifax times printed as SS:FF with SS ≥ 30 are seconds and hundredths ("55:27" = 55.27 s), and plain seconds ≥ 60 are m:ss ("65.38" = 1:05.38). Rewritten in `time_raw` with a warning.	Read as minutes, SS:FF ≥ 30 would be implausibly slow for a 200–500 m race; plain seconds ≥ 60 can't be parsed as printed. Both styles appear in the same files (D12).	`halifax` `event_blocks` / `team_place_html` files.	high	A Halifax file states its time format differently.	active	D12
A5	The Ottawa 2013 alumni sheet's SS:FF times are seconds and hundredths ("26:24" = 26.24 s). Rewritten with a warning.	Same reasoning as A4: read as minutes the times would be implausible for a sprint.	`ottawa` 2013 alumni sheet (`titled_list`).	high	A source prints a different reading.	active	D19
A6	When no place is printed, or printed ranks are ambiguous (repeat within one race), place is the rank by finish time (`confidence = inferred`).	Finish order by time is what places mean when crews race the same distance together.	`lane_grid` (Regina, D15); `race_cards` cards with repeated ranks (D16); `titled_list` (D19); `race_rows` (D19); `csv_cards` ambiguous lane splits (D20).	high	A file shows places that disagree with time order (penalties, adjustments).	active	D15, D16, D19, D20
A7	When a list has no place column, the printed row order is the finish order (`confidence = inferred`).	Parsed as finish-order lists in D5, D14 and D19; the sources print no place, so the order is the only signal.	`ordered_list` (D14) rows without a printed place; `csv_blocks` tables with no rank column; `titled_list` lists with neither lane header nor rank; `standings` pseudo-races (D5).	medium	A source lists crews in lane or draw order instead.	active	D5, D14, D19
A8	Ottawa 2016 charity places are re-derived from the on-water time, not the printed "adjusted time including fundraising incentive".	The adjusted ranking is not a race result; the on-water time is.	`ottawa` 2016 charity finals (`titled_list`).	high	The project decides fundraising-adjusted standings should be kept.	active	D19
A9	`09:99.99` and `09:95.95` are timing placeholders meaning no time (NT), not real times.	No crew takes ~10 min over a sprint course; the values recur as fillers.	Ottawa 2025–26 Saturday sheets (`ordered_list`), 2026 charity challenge cup (`race_rows`), `csv_blocks`.	high	A sheet uses these values as real times.	active	D14, D19
A10	"DNC" (did not compete) is the same as DNS.	The Can-Am sheet's own note defines DNC as did not compete.	`csv_cards` (Can-Am 2026).	high	A source distinguishes DNC from DNS in a way that matters for ratings.	active	D20
A11	Printed Can-Am times already include penalties (footnote stars are stripped).	D20 reading of the sheet: the stars are penalty footnotes and the printed times are taken as final. Not confirmed by the organizer.	`csv_cards` (Can-Am 2026).	medium	The organizer publishes raw times plus penalties separately.	active	D20
A12	When neither the file name, link text nor URL path has a year, the year comes from the page the link was found on.	Ottawa posts year-less files (`/media/342461/sunday-race-results.pdf`) from per-year pages (`/teams/race-results/2019.aspx`).	`_year()` fallback in `ingest/catalog.py`.	medium	A source page links files from other years.	active	D7
A13	A race with no printed distance gets its event's `default_distance_m` (`distance_inferred = TRUE`).	Events that run a single course length.	`dz-500` (500 m), `dz-oddball-2k` (2000 m), `lachine` (200 m, A3). Applied in `build_db.race_distance`; a `race_overrides.yaml` entry beats both the printed value and the default (D82).	medium	A race in those events is printed at another distance (the printed value always wins).	active	D6, D82
A14	A club acronym (`PDBC`, `FCRCC`, `22D` …) names the club whose full name is printed beside it, and only that club; trailing age, gender, tier and boat-size words (`Snr B Mixed`, `Premier`, `SM`, `_10`, `PM2`) name a crew of the club, not a different club. Used for `teams.club_id` only, never to merge teams.	`ingest/normalize/club_aliases.yaml`: each acronym is expanded only when a source prints it in the team name with the full name in the club column (or in the same club column); 32 rows, cited by edition. Acronyms with conflicting evidence (CSBD, MDBC, SNS) stay unexpanded. Suffix words: Pan Am CCC names (`Pickering DBC Snr B Mixed`), DBC Nationals 2025 `SM` crews in `boat_size = small` races.	`ingest/identity.py` `club_match_keys` (C5); all teams with a club.	medium	A source shows one acronym used by two clubs, or a suffix word that is part of a club's name.	active	D40
A15	A trailing class suffix of a crew name (`Snr B Mixed`, `U21 Open`, `Masters C Women`, `60+ women`) is the crew's age and gender class: `Snr/Senior/Sr/Masters` + letter A-D = senior_a-d, `40+/50+/60+/70+` = senior_a-d, bare `Masters` = senior_a, `Junior/Jr` = u18, `U<nn>` = u<nn>, `Men` = open; a bare `Senior` gives no age. Used only where the race prints no class for that field.	pan-am-ccc-2019 pages print the class beside each crew and in the race's left margin (`U16`, `Snr C`, `U21` for race 1) while the title reads only "Mixed"; crews keep the same suffix across rounds. The letter and `+` mappings follow AGE_RULES for race titles.	`normalize.rounds.class_suffix` / `entry_class` in `build_db` (entries.age_group, gender_class, class_suffix); `teams.age_group`. 2,059 entries take their age class from a suffix.	medium	A source shows `Masters C` or a bare `Senior` meaning another age band, or a suffix that is part of a club's name.	active	D55
A16	A race with no printed or default distance is in the sprint bucket (≤ 500 m) when its winning time is under 4:00, else in the long bucket (≥ 1000 m); a race with neither distance nor time takes its edition's majority bucket.	DB winning times: 99th percentile 2:38 for 200 m and 3:09 for 500 m races, 1st percentile 4:12 for 1000 m. Null-distance winners cluster at 0:30–3:30 and 8:00–12:30, with 24 races between 3:30 and 8:00.	Rating pools in `ratings/data.py` (R1, D67): 7,972 rating races bucketed by winning time, 8 by edition majority (reports/ratings/events.md).	medium	A source prints a distance that contradicts the bucket, or 1000 m races under 4:00 appear (fast small-boat 1000 m).	superseded by A19 (D78)	D67
A17	A race with no printed boat size is a standard (20-seat) boat race.	Small-boat races are labelled (`Small`, `Petit`, `SB`, `10s`) wherever the DB marks them (487 races since C71, D102: the bare word in a race title counts, 316 before); standard is the festival default in every series the DB covers. A title naming both sizes is `mixed` (22 races) and not rated, so the assumption no longer covers it. Known gaps: a small crew entered in a standard division is rated there (hamilton-waterfest-2023, 8 races; C73), and 6 dbc-nationals-2023 races under a stray "Small" section line have no size word of their own.	Rating pools in `ratings/data.py` (R1, D67): races with `boat_size` NULL go to the `standard_*` pools (17,234 races).	medium	An event runs only small boats without printing it (a small-boat regatta or festival).	active	D67, D102
A18	An edition with no `start_date` is dated 1 July of its year for rating order (the year's midpoint, phase_2_SPEC §2).	No edition in the DB has a start_date (0 of 314).	Rating event timestamps in `ratings/data.py` (R1, D67): every rating race today. Editions sharing a date interleave by race order.	low	Edition dates are filled (MP task R1b): the within-year order of events then comes from the dates.	active	D67
A19	A race with no printed or event-default distance is rated at the distance its winning time points to: 35 s to under 62 s → 200 m; 1:40 to under 4:00 → 500 m; 7:00 or more → 2000 m. Any other winning time, or none, leaves the race unrated (`distance_unknown`). An event default (A13) is a distance and always comes first.	Validated (D79) on Canadian races that print their distance, defaults excluded (`data.band_validation`, generated in reports/ratings/events.md): 200 m band 96.2 % raw / 98.4 % clean (1,715 races; misses: 26 races at 250 m, 1 at 175 m); 500 m band 94.0 % raw / 100 % clean (2,609); 2000 m band 97.7 % raw / 99.7 % clean (296). 'Clean' left out 214 printed distances the winning time rules out (200 m in 1:40+, 500 m under 1:15 or 7:00+): TIDBRF 2019/2023-2026 and GWN 2018 '200 m' at about 2:15, Nationals 2023 '500 m' at 43-57 s, Pan Am 2015 '500 m' at 10 min. C59 (D82) traced them to the parser (section headings swallowed into the block above), to labels naming two distances ("500m + 2000m": such a label now gives no distance, so A19 decides) and to one misprint; after it, raw accuracy is 98.45 / 100 / 99.3 % and the one race left (gwn-challenge-2026 race 37, placeholder times) is kept as printed with its reason in `race_overrides.yaml`. The first band (under 1:15, D78) was 84.5 % / 93.0 %: 100 m super sprints (Ottawa, winners 20-34 s) and 250 m races (59-74 s) fell in it, so D79 narrowed it to 35-62 s, the widest range at ≥ 98 % clean. Owner chose winning-time inference over excluding null-distance races (D78).	`ratings/data.py` `rated_distance` (DATA_VERSION 3): 7,699 of 13,006 rating races take their distance from the winning time (D82); 393 null-distance races are `distance_unknown`.	medium	A clean accuracy drops below 98 % in the generated validation; a series with unlabelled 100 m, 250 m or 1000 m races is added; a `keep_printed` reason in `race_overrides.yaml` turns out wrong.	active	D78, D79, D82
A20	The leaderboard floor is the 10th percentile (rounded down) of the Nationals crews' rated regattas and rated races in the season at any rated distance (D79: all-distances activity), with an absolute minimum of 2 regattas and 6 races; seasons without Nationals inherit the last earlier floor. It applies to every view, judged on the crew's all-distances activity; a view also needs 3 rated races in it.	Owner decisions D78 §7 and D79 §1: a crew is ranked when it raced at least as much as the least-active crews at that season's Canadian National Championships. In the data every season's computed floor is at or below the minimum (most Nationals crews appear at one regatta a season), so 2 regattas / 6 races applies throughout (reports/ratings/eligibility.md, 2026-11-09).	`ratings/ranking.py` `season_floors`, `eligible`, `MIN_VIEW_RACES`; `meta.json` `eligibility`; ranked standard_all 264, standard_500 262, standard_200 235, standard_2000 4, small-boat views 0 (D79).	medium	It ranks too few crews in a view (today: every small-boat view), or coverage per season grows so the computed floor rises above the minimum (C58 may link Nationals crews to their other regattas).	active	D78, D79
A21	`dragonboat.net/media/675699/sunday-race-results.pdf` is the 2024 Ottawa Sunday results (the file prints no year).	PDF created 2024-06-28 (the festival ran June 21-23, 2024); a crew named "2024: A Space ODC"; linked as "SUNDAY RACE RESULTS" from `raceresults.aspx` in the Wayback captures of 2024-08-11 and 2024-11-03; its divisions and crews match the 2024 Saturday file.	`ottawa-dbf` 2024 Sunday races (`catalog_overrides.yaml`, 48 races).	high	The festival re-links or re-labels the file, or a different 2024 Sunday file appears at media/675696.	active	D100
A22	Every race on a Manitoba Paddling Association "Upstream / Down Stream" sheet is a pursuit-format race, including the 2023 finals sheet, which does not print the word.	The 2024 September sheet is headed "Sunday September 8th, 2024 - PURSUIT" over its rounds and finals; the 2023 totals sheet (Sept DB Results 2.pdf) names the two rounds "Pursuit 1" and "Pursuit 2"; the 2023 finals sheet has the same two-boat Upstream / Down Stream columns and "Nth fastest in ..." seeding from those rounds.	`winnipeg-september` 2023-2024: section "Pursuit ..." on 72 races (`two_lane`), so `relay_or_pursuit` leaves them unrated.	medium	MPA describes the September format (side-by-side lanes would make them ordinary two-boat races), or G10 parses the 2026 sheet and finds a different heading.	active	D104
A23	A club's home province is the one `club_provinces.yaml` gives it, or, for a club not listed there, the province where all of its crews race mostly, when there are enough of them: two for a club a sheet prints in a club column, three for a club that is only a shared beginning of crew names, and crews at two or more US regattas for a US club (D131). Any other club has no province.	No results sheet prints a club's province. Register entries (59) rest on a place in the club's own name (Pickering, Edmonton, Calgary, Fort Langley, Toronto, Vermont, NYC), on what the sheets print beside the name ("Dragnum - Dragon Boat Calgary", "Dasa/Lethbridge", "Catch22 Racing Ltd (New York)"), or on a clear majority of crews for a name that is one club (22Dragons 223 of 232 in QC). The all-crews-agree case holds for 618 clubs (845 before D131: two crews sharing a first word, or one visit to a US regatta, proved thin: "Dragonflies", "Toba").	The province shown for 4,561 crews (`home_province_basis` `club`), the Canadian-home rule (D78) and the province floor.	medium	A club moves, a listed name turns out to cover two clubs, or a source that prints club addresses is found (a governing body's club list).	active	D121, D131

This build

Built from commit 512abfd, data as of 2026-11-13: 100,595 crew results in 18,625 races across 337 editions; 15,104 races and 81,429 results feed the ratings of 11,697 crews.