The Islands Trust function study · computed 2026-09-03

A study of how the Islands Trust actually works, built from its own record.

This study does not ask whether the Islands Trust should exist. It asks a narrower question: how does the Trust actually run its meetings, make its decisions, and enforce its rules — measured against what it says it does, and measured against ordinary local government? Every number below comes from the Trust's own record: its meeting minutes, back to 2008, and the Trust's own recordings of its meetings, turned into word-for-word transcripts.

2,265 documents — Meeting minutes: every body of the Trust, 2008 to 2026 (Source. docs/trust-function-study/results/2026-09-03/minutes-meeting-counts.json (query meeting-counts@1 lib@2026-09-03.1, computed 2026-09-03))
1,028 meetings — Meeting recordings, turned into transcripts: 2021 to 2026, 3,050 hours of recordings, word for word (Source. docs/trust-function-study/results/2026-09-03/transcripts-meeting-length.json (query 2026-09-03.1, computed 2026-09-03))
7 other local governments — The same measures, run on other governments: the regional districts and the island municipality that would do this job instead — for comparison, what ordinary local government looks like on the same measures

The stance we started from

The study must be able to conclude "it functions about as well as any comparable body" if that is what the record shows.
docs/trust-function-study/README.md, §1 ("Stance"), written before any measure was computed

Before a single number was computed, the people building this study wrote down that it had to be allowed to come out the other way — that the Trust could turn out to work about as well as any similar body. That sentence was published first, on purpose, so nobody could later say the study was built to reach a conclusion it started with. On several of the measures below, that is exactly what the record shows.

The rules we work under

Every measure is written down before it is computed
Nobody decided what a number should say and then went looking for it. What counts as "closed session" or "a defeated motion" was written down first, the same way for every body.
Every number comes with what it is a share of
A number like "952 meetings were closed" means nothing on its own. This study always says 952 out of how many — here, 2,265.
No fair comparison, no claim
If there is nothing honest to measure the Trust against, the study says so and stops there. It does not call a number "good" or "bad" with nothing to set it beside.
Bad news for this project gets published, and published first
If a measure comes out in the Trust's favour, or against what this project usually argues, it goes in the study anyway — and it leads the findings, not buried at the bottom.
Officials are named; the public is not
Elected trustees and Trust staff are named when they act in their public jobs. A member of the public who spoke at a meeting is never named on this page.
Every measure is a script anyone can re-run
Each number comes from a small computer program, kept in this project's own code, that reads the Trust's public record and counts. Anyone can run the same program again and check it.

What we found

The first three findings below go against what this project usually argues. They are listed first on purpose, per the rule above: bad news for this project's own case gets published, and published first.

How often do meetings go behind closed doors — and is the Trust unusually secretive?

952 of 2,265 meetings (42%)

Out of. 2,265 minutes documents, every Trust body, 2008–2026

About 4 meetings out of every 10 close their doors to the public for part of the meeting. That sounds high. But the same count has now been run on 7 other local governments — the kind that would take over if the Trust were abolished. All 7 of them close their doors MORE often than the Trust does. Counting whole governments, the Trust is the most open of the 8 measured. Two things have to be said next to that, and both are the next two findings on this page. First, closing the door less often is not proof of better behaviour on its own — we tested that, directly, and it is not. Second, when we stop comparing whole governments and compare only the committees that actually decide land use, the Trust is no longer the most open one.

Where	Figure
Islands Trust (all bodies)	42% (952 of 2,265)
Capital Regional District (committee + board)	45% (108 of 240)
Regional District of Nanaimo (committee + board)	51.3% (20 of 39)
Cowichan Valley Regional District (committee + board)	56.1% (128 of 228)
Bowen Island Municipality (council)	59.4% (133 of 224)
qathet Regional District (planning committee + board)	60.2% (153 of 254)
Comox Valley Regional District (committee + board)	73.1% (288 of 394)
Sunshine Coast Regional District (committee + board)	88.9% (32 of 36)

Read this table with the qualification the study's own openness exhibit requires: “A lower closed-meeting rate is not on its own evidence of better procedural practice; measured directly against comparators (C1–C3, 2026-09-03, re-run 2026-09-04 with qathet included), the Trust's rate of discussing a named party's bylaw enforcement matter in open session is comparable to the two best-measured regional districts, not exceptional in either direction, and the Trust states its grounds for closing more consistently than most of them.”

The most important caveat. The Trust's own M1b caveat applies identically to every comparator here: an agenda template's 'CLOSED MEETING'/'IN-CAMERA' heading is not evidence by itself; every number in this row requires a carried motion or recorded closure evidence, not just the heading. For Bowen specifically, 128 of its 133 closed meetings explicitly cite s.90 or the Community Charter in the motion block; SCRD's closed meetings cite s.90/the Community Charter at an even higher rate (32 of 32 — every one); Cowichan's own closed meetings cite it at almost the same rate as Bowen's (115 of 128); qathet's own closed meetings cite it at 149 of 153 (97.4%) — strong corroboration across all four that the detection is finding genuine closed-session motions, not noise, and true for both of qathet's own bodies (the Planning Committee's 16 closed meetings and the Board's 137 are not different in this respect).

Compared with the one outside committee that does the same land-use job, is the Trust still the more open one?

No. That committee closes 16.5% of its meetings; the Trust's own local trust committees close 37.2%

Out of. 16 of 97 qathet Regional District planning committee meetings (2016-01-28 to 2026-08-26), against 591 of 1,587 local trust committee meetings in the Trust's own minutes

The finding above compares whole governments. That is not quite fair to either side: a regional board also runs water, sewer, fire, roads and staff, and a lot of what it closes the door for has nothing to do with land use. The fair comparison is committee to committee. Until now this study had no outside land-use committee with a long enough record to use. qathet Regional District's planning committee is the first: 97 meetings over ten years, doing the job a local trust committee does. It closes its doors far less often than a local trust committee does. For every 10 local trust committee meetings that close their doors, about 4 of that planning committee's meetings do. This is the most awkward result the comparison work has produced for this study's own headline, and it goes here, in the findings, not in a footnote at the bottom.

Where	Figure
qathet Regional District — planning committee	16.5% (16 of 97)
Islands Trust — local trust committees	37.2% (591 of 1,587)
Islands Trust — all bodies	42% (952 of 2,265)
qathet Regional District — regional board	87.3% (137 of 157)

Both facts in this table are true at the same time. qathet's regional board closes its doors far more often than the Trust does; qathet's planning committee closes them far less often than the Trust's local trust committees do. Narrowing to the closer comparison changes the answer, so the narrower answer is published too, beside the wider one.

The most important caveat. ★ ADVERSE FINDING, STATED PLAINLY, PER THE STUDY'S OWN METHOD RULES: qathet's Planning Committee — its own land-use body, the closest comparator to a Local Trust Committee of any body measured in this whole study, and the first with a corpus (97 meetings, 2016-2026) complete enough to trust as its OWN number rather than a single document or a recommends-only shadow of the Board — closes only 16.5% of its meetings (16 of 97). That is BELOW the Trust's own combined 42.0% AND below the Trust's own LTC-only 37.2% figure. Every prior comparator in this row that supported 'the Trust is more open' did so only at the combined (committee+board) level, because none had a genuinely trustworthy land-use-committee-only cut to check separately (SCRD's EASC is one document; CRD's EAC recommends bylaws rather than reading them and its own M1b was never split out from the Board in this table; CVRD's own EASC, at 86.7%, is the one prior exception that already ran HIGHER than its own Board). qathet's Planning Committee is the first clean, well-populated, apples-to-apples land-use-committee comparison this study has been able to run against a Local Trust Committee specifically, and it UNDERCUTS the 'Trust is most open' framing rather than extending it. This does not overturn the combined-basis finding above (qathet's own Regional Board, at 87.3%, still closes far more than the Trust, and the combined 60.2% still sits above the Trust's 42.0%) — both facts are true at once, and both are reported, because narrowing to the closer comparator changes the answer and the study's own rules require that to be stated rather than left to the more favourable combined number.

Is an open door actually a fairness problem — does the Trust discuss ordinary people's bylaw cases in public more than other governments do?

No, not more than the governments that would replace it — but the Trust's rate is the highest of the 8 measured: 2.21 times per 100 meetings, against 2.19 and 2.03 for the two nearest, and 4 governments showing none at all

Out of. 50 matches in 2,265 sets of Trust minutes, set against 7 other governments' own minutes (36 to 394 meetings each)

This project's own publisher raised a fair objection to the finding above: it is no credit to a government to hold its doors open if what it does with them open is talk about a named person's bylaw case in public. So we measured that, the same way, on every government in the study. The answer is that the Trust does it at about the same rate as the two comparable governments with enough record to measure — not more. 4 governments show none of it at all, and they are not all the same kind of "none": Sunshine Coast Regional District (EASC + Board, 2024-2026 only), with only 36 meetings on file, has too small a record for "none" to mean much. qathet Regional District (Planning Committee + Regional Board) does not have that excuse — its own record runs 254 meetings, in the same range as the comparators that DO show hits, and still shows none. This does not clear the Trust of anything: 50 occasions in its own minutes are still 50 occasions, and only 6 of them show any sign in the record that the person affected was told or was there. That fact stands on its own. What the comparison rules out is narrower, and it is the thing this project believed: that the Trust does this more than the governments that would replace it. It does not.

Where	Figure
Islands Trust (all LTCs + Trust Council + Executive Committee + Conservancy Board)	2.21 per 100 meetings (50 in 2,265)
Cowichan Valley Regional District (EASC + Board, 2017-2018 & 2023-2026)	2.19 per 100 meetings (5 in 228)
Comox Valley Regional District (EASC + Board)	2.03 per 100 meetings (8 in 394)
Capital Regional District (EAC + Board)	0.83 per 100 meetings (2 in 240)
Regional District of Nanaimo (EASC + Board, 2018-2019 only)	0 per 100 meetings (0 in 39)
Bowen Island Municipality (Council + COW)	0 per 100 meetings (0 in 224)
Sunshine Coast Regional District (EASC + Board, 2024-2026 only)	0 per 100 meetings (0 in 36)
qathet Regional District (Planning Committee + Regional Board)	0 per 100 meetings (0 in 254)

Two more checks point the same way. When the Trust does close its doors, it names the part of the law that lets it do so almost every time: only 2.3% of its closed meetings give no reason at all, and just 3 of the 8 governments measured do better than that. The worst at explaining itself is Comox Valley Regional District (EASC + Board), which gives no reason at all for 16.3% of the meetings it closes — and it closes 73.1% of its meetings, far more often than the Trust. And of the bylaw-enforcement business this study can see at all, the share the Trust handles in the open (17.2%) sits within a few points of 15.7% and 13.2%. The Trust's own transcripts add a separate figure that has no comparator and is never mixed in: 2.33 per 100 meetings across 1,028 recorded meetings. This test now runs against all 7 comparator governments the study measures, qathet Regional District included.

The most important caveat. The Trust has verbatim transcripts (1,028 meetings, 2021-2026) as well as minutes; every comparator has minutes only. This asymmetry cuts AGAINST the hypothesis, not for it: a matter discussed aloud in a district meeting but summarised out of its minutes is invisible to this measure, so a minutes-only corpus will systematically UNDER-count a comparator's true open-session enforcement rate relative to the Trust's transcript-inclusive figure. For that reason the headline comparison in this file is MINUTES-ONLY ON BOTH SIDES (the `trust` row above, 2,265 documents, against each comparator's own minutes corpus) -- the one fair comparison available. The Trust's transcript-inclusive rate is reported as a separate, clearly non-comparable reference row and must never be read against a comparator's minutes-only rate.

Sources. docs/trust-function-study/results/2026-09-03/fairness-c1-enforcement-rate.json (query 2026-09-04.c1.2, computed 2026-09-04) · docs/trust-function-study/results/2026-09-03/fairness-c2-stated-grounds.json (query 2026-09-04.c2.2, computed 2026-09-04) · docs/trust-function-study/results/2026-09-03/fairness-c3-open-vs-closed.json (query 2026-09-04.c3.2, computed 2026-09-04)

Do trustees ever vote against each other, or does the Trust just rubber-stamp things?

337 defeated (1.6%); recorded disagreement on 2.4% of votes

Out of. 21,620 recorded votes (“motions”) across 2,265 meetings

Out of every 1,000 votes the Trust takes, about 16 are defeated and about 24 show a trustee formally recorded as opposed. That is a small share. But set beside the regional district with the closest comparable minutes, it is bigger, not smaller: that comparator's minutes show a defeated vote in only 2 of every 1,000, and not one single recorded disagreement in ten years of meetings. The Trust's minutes show more visible disagreement among its own members than the comparator's do, not less.

Where	Figure
Islands Trust	defeated 1.6%, dissent 2.4%
Capital Regional District (committee + board)	defeated 1.9%, dissent 7.6%
Regional District of Nanaimo (committee + board)	defeated 2%, dissent 8.3%
Cowichan Valley Regional District (committee + board)	defeated 1.6%, dissent 3.5%
Bowen Island Municipality (council)	defeated 1.4%, dissent 11.6%
qathet Regional District (planning committee + board)	defeated 1.6%, dissent 10.2%
Comox Valley Regional District (committee + board)	defeated 0.2%, dissent 0%
Sunshine Coast Regional District (committee + board)	defeated 0.6%, dissent 1.7%

This row has been rewritten three times as comparators were added, which is worth knowing in itself. Across eight bodies the recorded-dissent share runs: Comox Valley 0%, Sunshine Coast 1.7%, Islands Trust 2.4%, Cowichan Valley 3.5%, Capital 7.6%, Nanaimo 8.3%, qathet 10.2%, Bowen Island 11.6%. The Trust sits third from the bottom: it records less visible disagreement than most comparable bodies, but not the least. Both figures below it rest on thin evidence — Comox Valley's 0% across ten years is far more likely a minute-taking convention than genuine unanimity, and Sunshine Coast's covers only 36 meetings. The sharpest single comparison is Bowen Island, the same island now governing itself, whose minutes record dissent nearly five times as often as the Trust's. An early version of this page said the Trust records MORE disagreement than its comparators; that rested on the Comox Valley figure alone and was wrong. CRD's, RDN's, Cowichan's, Bowen's, SCRD's AND qathet's defeated-shares (1.9%, 2.0%, 1.6%, 1.4%, 0.6%, 1.6%) all sit reasonably close to the Trust's own 1.6% (SCRD's is the furthest below, but still under 1 percentage point away; qathet's is an EXACT match) and are markedly higher than CVRD's 0.2%; on this specific figure, six of seven non-Trust bodies look MORE like the Trust than CVRD does, the opposite of what a simple 'Trust vs everyone else' framing would predict; this table draws no evaluative conclusion from that, per rule 3, but the number itself is worth reading carefully rather than assuming the Trust is always the outlier.

The most important caveat. LTC minutes since about 2019 record 'It was MOVED and SECONDED' with no names; dissent is recorded there only as 'Trustee X opposed', so the recorded-dissent share is a floor and named-vote rates are computed for Trust Council only.

Do applications get stuck, coming back to meeting after meeting without a decision?

15.7% of files show up at 3 or more separate meetings

Out of. 1,538 distinct application files named in the minutes

Most applications the Trust handles — about 7 in 10 — are named in the minutes on only one or two meeting dates before they are done. About 1 in 6 keeps coming back for three meetings or more; the single longest-running file in the whole record, a Galiano rezoning, was named at 35 different meetings over more than ten years. Against the regional district comparator with a usable count, the Trust's files move through FEWER meetings before they are done, not more.

Where	Figure
Islands Trust	15.7% (of 1,538 files)
Comox Valley Regional District	35.9% (of 117 files)

Capital and Nanaimo regional districts: not computable — 0 application-file identifiers found.

The most important caveat. This is a proxy: an appearance is any mention, not a deferral. A rezoning legitimately appears at receipt, community information meeting, public hearing, three readings and adoption (4–8 meetings); a development permit normally at one or two. Compare within a type, never across types.

Sources. docs/trust-function-study/results/2026-09-03/minutes-m2c-deferral-chains.json (query M2c@1 lib@2026-09-03.1, computed 2026-09-03) · docs/trust-function-study/results/2026-09-03/comparator-cvrd-m2c-deferral-chains.json (query M2c@1 lib@2026-09-03.1+comparator-adapter@1, computed 2026-09-03)

When the Trust talks about a named private person's bylaw case, does the record show that person getting a chance to answer?

48.5% of cases (32 of 66) show no sign in the record that the person answered

Out of. 66 cases with an address, file number or business name specific enough to search for later — 41 where the Trust discussed a named person's own bylaw-enforcement matter, 25 where someone else raised a different named person's matter at a delegation

"Right of reply" is a simple idea: if a government talks about a specific person by name, in public, that person should get a fair chance to answer back, on the same public record. We tested it. For every case where the Trust discussed a named person's own bylaw problem, or where someone else stood up and talked about a different named person's matter, we searched that meeting and the next two meetings of the same body for any sign that the person, or someone speaking for them, answered. In 48.5% of cases — 32 of 66 — we found no sign in the record that they did. That is not the same as saying the Trust kept them from answering. It only means the part of the record we can search does not show it happening. Most contact with someone facing a bylaw complaint happens off the public record entirely — a letter, a phone call, a conversation nobody wrote down — so finding nothing here is not proof that nothing happened. Broken down by island, the pattern for people's own enforcement cases is stark: on Denman, none of 3 cases show a reply in the record; on Hornby, none of 3; on Gabriola, 3 of 8 do. The full table, every island measured, is below.

Where	Figure
Denman Island	0 of 3 (0%)
Executive Committee	0 of 1 (0%)
Gabriola Island	3 of 8 (37.5%)
Gambier Island	1 of 3 (33.3%)
Hornby Island	0 of 3 (0%)
Mayne Island	0 of 1 (0%)
North Pender Island	0 of 2 (0%)
Salt Spring Island	5 of 10 (50%)
Saturna Island	1 of 3 (33.3%)
South Pender Island	1 of 1 (100%)
Thetis Island	1 of 4 (25%)
Trust Council	0 of 2 (0%)

This search can be wrong in both directions. search_terms are derived from the same identifiers used for anonymisation (civic address, file number, business name); a hit whose only strong tag was a lot/PID with no separately-searchable civic address or file number can produce a term too generic or too rare to find a genuine reply (or to avoid false positives) — both directions of error are possible and not separately measured here. On the other side: 'a representative speaks' is approximated by speaker_role NULL/'public' matching the term; it cannot distinguish the affected party or their actual representative from an unrelated member of the public who happens to mention the same address (e.g. a neighbour on an unrelated topic, or staff reading a letter from the party into the record, which is role NULL in some transcripts and would be miscounted as the party speaking). The 25 delegation cases — someone else raising a different named person's matter, not shown island by island above — show a reply in the record 88% of the time, far higher than the enforcement cases. The likeliest reason is that the same meeting keeps discussing the same address minutes later, not that the specific person named necessarily spoke themselves, so this figure is reported on its own rather than blended into the table above.

A few real cases from the record.

The most important caveat. No claim is made about why no reply was found — see F1's own note that most enforcement/owner contact happens off the public record (letters, phone calls); reply_found=false is not evidence of a due-process failure by itself.

Source. docs/trust-function-study/results/2026-09-03/fairness-f5-right-of-reply.json (query 2026-09-03.1, computed 2026-09-04)

When the Trust discusses a specific person's bylaw-enforcement problem in open session, does its own written record show that person knew it was happening?

6 of 50 (12%)

Out of. 50 occasions found in the Trust's own written minutes where a named person's or property's bylaw-enforcement matter was discussed in open session (24 more turn up in the word-for-word recordings, not counted in this line — see the plain-language note below)

Every one of these 74 cases is MODEL-CODED: a computer program (a language model) read each one and sorted it, not a person. That is stated plainly here because it matters — a computer's sorting has to be checked, not just trusted. So a sample of 26 of the 74 cases was checked again, and its accuracy is a measured number, not a guess: it agreed with the first sort 88.5% of the time on every measure at once, and 97.7% of the time measure by measure. That clears this project's own bar for trusting a result (80%), so the finding is published. The finding itself: looking only at the Trust's written minutes, in 6 of 50 cases (12%) does the record itself show any sign that the affected person was told, or was there. In the other 44, the record simply does not say — which is not the same as proof nobody was told. Most contact with someone facing a bylaw problem happens by letter or phone, off the public record. But it means that 88% of the time, a reader of the Trust's own official minutes has no way to know, from the minutes alone, whether the person being discussed knew about it.

The most important caveat. This is model coding, not human coding — a language model read each case's stored quote and sorted it, and its accuracy is measured against a re-examined sample, not assumed. The model only sees a short window of text around each mention; a longer quote might show more. And "the record does not say" is not the same as "nobody was told" — the Trust routinely handles notice to a property owner off the public record, by letter, which this method cannot see. Separately, in 12 of 74 cases, the content itself reads as squarely the kind of personal or legal information the Trust's own rules say belongs in a closed meeting, discussed in public anyway; in 7 more, the record shows it WAS in fact handled closed, with only a bare line about it left in the public minutes — the closed-door process working as intended. Only 7 of 74 cases show an actual decision or direction being given in the same meeting; most are a status update or a staff report with no vote in view.

Source. docs/trust-function-study/results/2026-09-04/fairness-f1-enforcement-fairness-coding.json (query 2026-09-04.1, computed 2026-09-04)

How long does it take a bylaw to go from first reading to becoming law?

273 days, typically

Out of. the middle value (median) of 212 Trust bylaws with both a first reading and an adoption date, out of 551 distinct bylaws seen

Half of the Trust's local bylaws that reach adoption take longer than about nine months from first reading to becoming law; half take less. Regional district comparators show much shorter medians — but this is the one measure in this study we will not let stand as a simple comparison, because it is not a fair one. See the caveat below before repeating this number against any comparator.

Where	Figure
Islands Trust (local trust committee bylaws)	median 273 days (n=212)
Comox Valley Regional District (Board)	median 28 days (n=424) — not a like-for-like population, see caveat
Capital Regional District (Board)	median 0 days (n=360) — not a like-for-like population, see caveat

The most important caveat. ★ The single most important caveat on this row, carried forward from the CVRD entry and now confirmed a sixth time: none of the comparator bylaw populations is matched by subject to the Trust's. The Trust's 551 LTC bylaws are overwhelmingly OCP/zoning amendments tied to individual land-use applications. Each comparator's own bylaw count is that body's FULL output — budget, tax-rate, service-establishment, borrowing, procedure and fee bylaws mixed in with land-use rezoning bylaws. This matters MOST for Bowen of all comparators: a regional-district Board at least splits business between a Board and an electoral-area committee, but Bowen's Council IS the whole municipality — water, sewer, fire, parks, finance, personnel AND land use bylaws all pass through the same 153-bylaw count, with no land-use-only subset available to isolate; SCRD's own 31 bylaws carry the same caveat, sharpened by the same EASC-barely-publishes-minutes finding noted throughout this table (SCRD's own land-use-only committee record is too thin to isolate a land-use-only bylaw subset from); qathet's own 114 Board bylaws carry the same full-service mix, though its 13 Planning Committee bylaw RECOMMENDATIONS are at least drawn from a genuinely land-use-focused body (see the qathet-specific caveat below). Reading any of the day-counts below as 'Bowen/SCRD/qathet approves land use N times faster than the Trust' is exactly the unfair comparison rule 3 warns against; nothing published from this table should make that claim without first isolating each comparator's own land-use (rezoning/OCP) bylaws specifically, which no pass has done yet.

Source. docs/trust-function-study/results/2026-09-03/minutes-m4a-bylaws.json (query M4a@1 lib@2026-09-03.1, computed 2026-09-03)

How long do Trust meetings actually run?

2.8 hours, typically; 26.5% run past 4 hours

Out of. 1,028 recorded meetings with a finished transcript, 2021–2026

Half of Trust meetings run under about 2 hours 48 minutes; half run longer. About 1 meeting in 4 runs past the 4-hour mark. The longest single meeting on record ran 9.41 hours. There is no published measure of "typical" meeting length for a comparable body anywhere in Canada, so this number is reported on its own, not against a benchmark.

The most important caveat. A recording is one stream file, not necessarily one meeting: Trust Council sessions span days and appear as several recordings; a meeting whose stream restarted appears as two recordings; a recording may start before call-to-order and run past adjournment.

Source. docs/trust-function-study/results/2026-09-03/transcripts-meeting-length.json (query 2026-09-03.1, computed 2026-09-03)

Who does most of the talking at a meeting?

only 30.9% of all speech is confidently tied to a named official

Out of. 2,578 hours of speech across 1,027 meetings

Of everything said at a Trust meeting, only about 3 words in 10 can be confidently pinned to a specific named trustee, chair or staff member by voice. Of that named speech, the great majority — about 9 in 10 — is trustees and the chair, not staff. The rest of all speech, about 7 words in 10, is not confidently attributed to anyone by name; most of that is the public, and officials the voice-matching system did not recognise.

The most important caveat. Public speakers are only labelled by hand (role 'public'), so 'unlabelled' contains most public speech and any official the voiceprint did not recognise.

Sources. docs/trust-function-study/results/2026-09-03/transcripts-attribution-coverage.json (query 2026-09-03.1, computed 2026-09-03) · docs/trust-function-study/results/2026-09-03/transcripts-m3a-floor-share.json (query 2026-09-03.1, computed 2026-09-03)

Do members of the public actually get to speak, and how much time do they get?

533 of 1,028 meetings (51.8%) have a findable public-comment period

Out of. 1,028 meetings with a finished transcript; 945 public-comment periods found within them

A findable public-comment period — a "town hall" or "delegations" section — turns up in about half of all meetings searched. Where one is found, it typically runs under 5 minutes and takes up about 4.2% of the whole meeting's speaking time. This is a search for the WORDS a chair uses to open public comment, not a read of the agenda, so it is a floor, not an exact count.

The most important caveat. The marker is a phrase search, not the agenda: a chair who opens town hall without one of the phrases, or with a phrase Whisper mis-heard, is missed (see marker_yield_pct); a chair mentioning 'town hall' in passing with a cue-like word can open a false span.

Source. docs/trust-function-study/results/2026-09-03/transcripts-m3c-public-participation.json (query 2026-09-03.1, computed 2026-09-03)

What does current zoning actually let someone build, island by island?

19,213 homes allowed by right; up to 26,749 counting secondary suites and similar add-ons

Out of. 18,446 of 21,521 zoned parcels (85.7%) across the twelve islands with a resolved zoning rule

This is not a measure of how the Trust behaves — it is a plain count of what its own zoning bylaws currently allow, parcel by parcel, on each of the twelve islands measured. Coverage is uneven: on Denman and Hornby, the rule is resolved for essentially every zoned parcel; on Thetis, only about 1 parcel in 17. See "what is missing and why" below for why Thetis is so low.

The most important caveat. A resolved rule is a floor, not a ceiling: several bylaws grant extra density through a map or schedule this study could not test parcel by parcel, so the true maximum on some islands is higher than the number shown here.

Source. docs/trust-function-study/results/2026-09-03/density-coverage.json

When the Trust enforces its bylaws, is it mostly about the environment, or about how people house themselves?

148 housing-related matters (19.4%) vs. 72 environmental matters (9.4%) — about 2.06× as many

Out of. 764 enforcement matters found in the minutes and transcripts, of which 502 (65.7%) could not be sorted into either category

The Trust's job, by law, is to "preserve and protect" the environment. Where this study could tell what a bylaw-enforcement matter was actually about, it was about twice as likely to be a dwelling or short-term-rental matter as an environmental one. But that "where it could tell" matters: nearly two-thirds of everything tagged as bylaw enforcement in the record could not be sorted into either bucket at all. And the Trust's own enforcement process only responds to complaints — it counts what neighbours choose to report, not what the Trust would prioritise if it went looking on its own.

The most important caveat. The Trust's own bylaw enforcement process is complaint-driven: 'with the new trust council policies and with some policies adopted by local trust committees... we're ensuring that there's a narrow focus to investigations so it's all set out in policy... and it's always been the case that we can't take a broad approach anyway, just by judicial decision... the new policies the trust council has adopted are making it clear that bylaw officers are to limit their investigations on the property to the complaint received unless... it involves... concerns about health and safety... [or] a development permit area... or set back from water.'

Source. docs/trust-function-study/results/2026-09-03/enforcement-e1-subject-mix.json (query e1-subject-mix@2026-09-03.1, computed 2026-09-03)

What we got wrong

Two claims this project made were withdrawn once the study caught up with its own record. Both are left here, in full, rather than quietly deleted.

Withdrawn: “The Islands Trust is markedly more open than the regional district that would replace it.”

The first version of this comparison used one regional district. On that single comparator, the Trust looked far more open than the government that would replace it. So the study kept adding comparators. There are now 7, and the corrected table below carries every one of them. Two further corrections have been made to this same claim since. The Trust does still close its doors least often of the 8 whole governments measured — but comparing whole governments is the loosest test available, and on the two tighter tests the study has since run, the claim does not survive as written. Compare only the committees that decide land use, and the Trust is no longer the most open. Ask what an open door is being used FOR, and the Trust turns out to be ordinary rather than exceptional in either direction. Both of those are findings on this page, not footnotes: “A lower closed-meeting rate is not on its own evidence of better procedural practice; measured directly against comparators (C1–C3, 2026-09-03, re-run 2026-09-04 with qathet included), the Trust's rate of discussing a named party's bylaw enforcement matter in open session is comparable to the two best-measured regional districts, not exceptional in either direction, and the Trust states its grounds for closing more consistently than most of them.”

Where	Figure
Islands Trust (all bodies)	42% (952 of 2,265)
Capital Regional District (committee + board)	45% (108 of 240)
Regional District of Nanaimo (committee + board)	51.3% (20 of 39)
Cowichan Valley Regional District (committee + board)	56.1% (128 of 228)
Bowen Island Municipality (council)	59.4% (133 of 224)
qathet Regional District (planning committee + board)	60.2% (153 of 254)
Comox Valley Regional District (committee + board)	73.1% (288 of 394)
Sunshine Coast Regional District (committee + board)	88.9% (32 of 36)

Caught by. Our own checks, in three rounds. This study's rule is to measure against more than one comparator before drawing a conclusion. The second and third comparator changed the answer the first time. Adding a comparator with a real land-use committee changed it again. Testing whether an open door is a good thing here — the publisher's own objection to his own finding — changed it a third time.

Sources. docs/trust-function-study/results/2026-09-03/benchmark-table.json · docs/trust-function-study/results/2026-09-03/comparator-qathet-m1b-in-camera.json (query M1b@1 lib@2026-09-03.1+comparator-adapter@2, computed 2026-09-04) · docs/trust-function-study/results/2026-09-03/fairness-c1-enforcement-rate.json (query 2026-09-04.c1.2, computed 2026-09-04)

Withdrawn: “47.1% of cases (32 of 68) show no sign in the record that the person answered”

A coordinator review of the C1 comparator measure (fairness-c1-enforcement-rate.json) found and locally filtered a 'stub report' false-positive pattern in F1's own published minutes rows -- a bare, content-free 'Bylaw Enforcement Report' heading ('None', or a trustee giving a brief update on Bylaw Enforcement with no further content) picking up a specificity tag that actually belonged to a different, nearby agenda item. C1's own self_check disclosed this on 2026-09-03, but the fix was applied only inside C1's own comparison arithmetic, not in F1 itself, so F1's published headline -- and F5, which re-derives its own cases from F1's classification logic rather than reading F1's file -- both still carried it. Fixed 2026-09-04: F1's own headline count and F5's own case count are now both recomputed with the fix applied at the source. Two of F5's 68 cases (South Pender LTC, 2012, both a bare 'Bylaw Enforcement Report None' heading) were not genuine cases and are removed. The corrected figures: F1's own count of named-party enforcement matters discussed in open session moves from the original "77 occasions (53 in minutes, 24 in transcripts) where a named party's enforcement matter was discussed in open session" to 74 occasions. F5's right-of-reply figure — the one published live on this page — moves from "47.1% of cases (32 of 68) show no sign in the record that the person answered" to 48.5% of cases (32 of 66) showing no sign in the record that the person answered — the number of cases with no sign of a reply did not change (32 either way); only the denominator did, because two of the removed cases were themselves the false positive. This correction makes the finding read very slightly WORSE for this project's own case — a smaller denominator with the same "no sign of reply" count raises the share from 47.1% to 48.5% — and is published anyway, per this study's own rule that adverse results are published (README §4, rule 4).

Caught by. The C1 comparator measure's own self-check (fairness-c1-enforcement-rate.json), which found the pattern while hand-verifying its comparator results on 2026-09-03 but, being a separate, narrower-scoped measure, filtered it only inside its own arithmetic rather than fixing F1 at the source. Caught as a gap between what one measure disclosed and what another measure had actually corrected, and fixed the following day.

Sources. docs/trust-function-study/results/2026-09-03/fairness-f1-enforcement-open-session.json (query 2026-09-04.3, computed 2026-09-04) · docs/trust-function-study/results/2026-09-03/fairness-f5-right-of-reply.json (query 2026-09-03.1, computed 2026-09-04) · docs/trust-function-study/results/2026-09-03/fairness-f5-right-of-reply.json

Withdrawn: “The Hornby Island official community plan is missing from the Trust's own site.”

The link this project had on file used an older, retired Trust web address. The plan itself is live, at a renamed address, on the Trust's own site.

Caught by. Our own automated link-check, which tests every document link this project holds against the Trust's live website on a schedule.

Source. docs/a11y-audit (accessibility audit, document id 453)

What is retrieved but not yet judged

These two lists are not findings. A computer search built them; nobody has read every row yet. A number drawn from either of them, before that reading happens, is not something this project would publish.

Trustees talking about how the Trust itself functions

9,889 passages — not a finding, a reading list.

Out of. found by a computer word-search across 708 meetings and 35 named speakers, out of 20,774 candidate passages the search looked at

A computer program searched every meeting for named trustees and the chair talking about the Trust's own capacity, its process, its mandate, or how it is run — phrases like "we never get to" or "staff capacity" or "this process." It found 9,889 matches. That is not a finding. It is a reading list. A computer search cannot tell the difference between a trustee complaining that the Trust is broken and a trustee explaining, calmly, how a normal process works. A person has to read every one of them, in context, and sort them by hand before any number drawn from this list can be called a finding.

Source. docs/trust-function-study/results/2026-09-03/transcripts-m5a-self-assessment.json (query 2026-09-03.1, computed 2026-09-03)

What is missing, and why

Minutes are linked to only about half of the meeting recordings
546 of 1,028 finished meeting transcripts have a matching minutes document linked to them so far. The rest exist as a transcript with no official minutes attached yet — the linking work is ongoing.
One regional district could not be measured — and we did not work around it
Sunshine Coast Regional District (its minutes host sits behind an active bot-blocking challenge). It publishes its minutes online and blocks automated reading of that public record. This study's rule is to respect that rather than find a way around it. Two districts that were once on this list have since been measured. Cowichan Valley was listed as blocked because its robots file names one of the well-known AI crawlers; a closer reading showed that line sits in a standard template, and that the rule governing a named archive fetcher like ours permits reading for reference. qathet's portal genuinely does disallow crawlers — the one verdict in this whole study that held up on a second look — so its 254 minutes documents (2016-2026) were collected by hand through a browser instead, the same way Bowen Island's were, and have now been measured. If the last district's records are wanted, the honest routes are a formal records request, an authorised export, or collecting them by hand — not further automated fetching.
Bowen Island — the one before-and-after this study still lacks
Bowen Island Municipality's council minutes WERE obtained — 414 documents, deduplicated to 224 distinct meetings, 2016 to 2026 — after the publisher downloaded them by hand through his own browser, because the portal disallows automated crawlers. This corpus starts seventeen years after incorporation, so it shows Bowen as a municipality and cannot show the transition itself. Bowen sat inside the Islands Trust Area until it incorporated as its own municipality in 1999 — the one island in this whole study with a real before-and-after under two different governments.
Thetis Island's zoning coverage is only 5.7%
On Thetis, many zoned parcels sit close to the boundary with a neighbouring island's separate bylaw, and a computer check could not safely tell which bylaw actually governs each one. Rather than guess, this study leaves those parcels unresolved.
Assessed property values are missing for 8 of the twelve islands
Salt Spring, North Pender, South Pender, Gambier, Mayne, Thetis, Saturna, Lasqueti have no assessed-value data in this project's records at all. Without it, this study cannot say anything about property tax, affordability, or cost-per-resident on those islands.

How to check us

Every number on this page links to the exact result file it came from, in this project's own public code repository. Each result file carries its own definition, its own denominator, the exact version of the script that produced it, and the date it was computed. Nothing here is hand-typed into a spreadsheet first.

scripts/study/minutes/run.sh <measure|all> [YYYY-MM-DD] recomputes the minutes measures against the live database; scripts/study/comparator/run.js does the same against the stored regional-district comparator files; scripts/study/fairness/run.sh and scripts/study/enforcement/e1-subject-mix.js recompute the fairness and enforcement measures. Anyone with access to the same public record can run the same scripts and compare the output.

Every result file, in the repository · Every measure script · The full study design

Disclosure

The publisher of this project has a personal interest in one Denman Island matter now before local officials. That matter is excluded from every evaluative example in this study — it is never used to illustrate a finding. Its meetings still count in every total on this page, the same as any other meeting, so the aggregate numbers above are not quietly cleaned up to leave it out.

The counterfactual study: Trust vs. non-Trust islands · The Trust's own 2022 governance review · The accessibility audit