Summary
Between May and July 2026, New York City's Commission on Government Efficiency, a 15-member charter revision commission, ran ten public hearings across the five boroughs and heard hundreds of witnesses before filing five ballot questions for the November 3 general election. OCC Research tracked the entire cycle with a governance memory system: a structured, citable record of who testified, what they asked for, and what the commission did with it, published as a bilingual public dashboard at occresearch.org/coge.
This case study reports what that record shows. Three findings stand out:
- 1The commission engaged most of what it heard, but the loudest ask did not reach the ballot. Of 21 recurring themes, four reached the November ballot; the final report hands the most-raised theme of the cycle, open primaries with 33 witnesses, to a future commission, with reasons it spells out.
- 2The official record ran weeks behind the process it documents. Video with automatic captions went up shortly after each hearing; the official transcripts are what lagged. As final-report drafting began, seven of the eleven meetings held had no official transcript posted; the final meeting's transcript link later pointed to the wrong file for roughly two weeks.
- 3Every ballot question can be measured with data the city already publishes, and each one exposes a data gap. The subject of ballot question 1, revocable consents, has no public dataset at all.
The Setting
Mayor Mamdani appointed the Commission on Government Efficiency (COGE) in 2026 with a mandate to propose efficiency-related charter amendments for the November 3 ballot. It had roughly eight weeks of public process: a June 4 kickoff, ten borough hearings from June 9 to July 13, a preliminary staff report released July 2 and adopted July 6, and a final meeting on July 23 at which the commission approved five ballot questions, its final report, and the filing resolution, each by a 15-0 roll call.
OCC Research followed the cycle as an independent observer. Our purpose was not advocacy for any proposal. It was to test a thesis: that public institutions lose the connective tissue between what they hear and what they decide, and that a structured memory of the process, built while the process runs, changes what the public can know.
What OCC Built
The system behind the dashboard is a scoped-down governance memory system (GMS v0) with three components:
Proposal Lifecycle Metadata (PLM): commissioners, resolutions, hearings, votes, and the charter chapters under review, each stored as a structured record with a resolvable identifier.
Recurring Themes and Frictions (RTF): patterns across testimony, built up hearing by hearing. Each theme carries its witness count, supporting quotes tied to specific speakers and meetings, and source chips that link claims to verifiable external documents.
Outcome Review Anchors (ORA): a scaffold keyed to the November 3 ballot, designed to populate once outcomes exist. The measurement framework below is, in effect, ORA's plan.
The workflow: we ingest hearing audio and official transcripts, turn each witness and claim into a structured record checked against a fixed format, correct names against adopted minutes, and check every claim on the public page against its source before publication. We enforced editorial standards line by line: plain English, hedged predictions, precise attribution (one witness means one witness), and quotes attached only to points they actually support.
Two numbers describe the resulting record: 21 recurring themes tracked across ten hearings, and, after the final report appeared, a report-check appended to every theme stating what the commission did with it.
Did the City Listen? What the Record Shows
The question that organized our analysis came from a Manhattan florist's plea at the first hearing: listen before legislating. After the final report, we checked every theme against it.
| Witnesses | Theme | Where it landed |
|---|---|---|
| 33 | Closed primaries shut out unaffiliated New Yorkers | Deferred with reasons |
| 28 | Contract-registration delay makes firms finance the city | Partly on the ballot |
| 21 | Build in-house digital teams instead of renting them | In the commitments |
| 20 | Write AI rules into the charter | Not in the final report |
| 20 | One accountable leader for the city's technology | Partly in |
| 20 | MWBE firms are certified but shut out of contracts | Not in the final report |
| 20 | Remove outdoor dining's revocable-consent requirement | On the ballot |
| 16 | Streamlining must not become a “fast pass for exploitation” | In the report, explicitly |
| 16 | Civil service hiring is the bottleneck under everything else | Deferred, plus commitments |
| 16 | Prescriptive procurement timelines; pay nonprofits on time | Partly, outside the charter |
| 15 | Protect oversight agencies' budgets and independence | Deferred with reasons |
| 14 | Untangle the rules that stall street redesigns | On the ballot |
| 12 | Consolidate fragmented permitting into a single authority | On the ballot |
| 8 | Counter-theme: do not streamline away the council's checks and balances | Partly heeded; changed outcomes |
| 6 | Stop treating nonprofit stewards like concessionaires | Deferred with a named path |
| 5 | Set targets, not rigid rules, for the rainy day fund | On the ballot |
| 5 | Proportional representation | Not in the final report |
| 4 | Give the city a comprehensive plan | Not in as asked |
| 4 | Make the city budget legible and comparable | Not in the final report |
| 1 | Listen before legislating | In the report's method |
Reading the scorecard, patterns emerge.
Volume did not decide outcomes; jurisdiction did. The four themes that reached the ballot ranked 7th, 12th, 13th, and 16th by witness count. The two most-raised themes, open primaries and contract registration, ran up against limits the commission said were outside an efficiency mandate or governed by state law. The final report is candid about this: on elections it concedes that “the bulk of testimony the Commission received on this issue was to promote open primaries,” recounts how the 2003, 2010, 2024, and 2025 commissions each declined or were rejected by voters, and concludes electoral reform did not fit an efficiency mandate. The preliminary report did not include the issue; the final report engages it and says no, with reasons.
Engagement, not adoption, is the honest measure of listening. Almost every theme with a substantial witness count got a written answer of some kind: a ballot question, a commitment by the administration (the 2-for-1 hiring restriction lifted, OMB review timelines, DCAS exploring new IT titles), or a deferral with stated reasons. By the standard of “did testimony leave a trace in the written record,” the commission's final report scores well. By the standard of “did the most-requested changes happen,” it does not. Both statements are true, and a memory system's job is to keep them distinguishable.
The counter-testimony mattered. The eight witnesses who warned against streamlining away the council's checks and balances, and the labor witnesses who asked that notification requirements survive any reform, can point to specific guardrails in the final package. Minority positions left fingerprints, which is worth knowing in a process that could otherwise be scored only by majorities.
The Official Record Ran Behind the Process
Every hearing had a stenographer, and the commission posted video of each hearing, with automatic captions, shortly after it ended, so the public could always watch the proceedings. The official transcripts moved slower, and the gap is documentable because OCC logged the commission's meetings page throughout, and the Internet Archive's Wayback Machine captured it independently.
- 1Jun 23Nineteen days in, a Wayback capture of the meetings page shows no official transcripts posted for any of the five meetings already held.
- 2Jun 23 – Jul 9The first four transcripts appear: the June 4 kickoff, June 9 Manhattan, June 10 Bronx, and June 11 Brooklyn. Two carry internal file dates (June 17, June 22) that predate the June 23 capture, so the files existed before the page linked them.
- 3Jul 9 · Jul 12Wayback captures show those same four transcripts, out of ten meetings held; no meeting after June 11 has a posted transcript.
- 4Jul 13The posting lag is raised on the record at the final hearing and answered on the record; that hearing's official transcript preserves the exchange verbatim. Seven of the eleven meetings held have no posted transcript as final-report drafting begins.
- 5Jul 14 – Jul 16The rush begins: the June 22 Queens transcript appears by July 14 (OCC's logged scrapes), and the July 16 Wayback capture shows eight posted, with June 23, June 30, and July 6 added within days of the exchange.
- 6Jul 22Nine posted (July 1 added). The July 8 and July 13 hearing transcripts are still absent the day before the commission's final vote.
- 7by Jul 26All ten public hearings' transcripts are posted (OCC's logged scrapes), most after the final report was drafted and the ballot set.
- 8Jul 26 – Aug 7The link labeled “July 23 Meeting Transcript” points to the July 13 hearing's file; the meeting that adopted the ballot questions has no accessible official transcript.
- 9Aug 7The genuine final-meeting transcript is posted, fifteen days after the meeting, and the link is corrected.
Two methodological notes. First, the city's own server metadata could not establish this timeline: a site redeploy on July 14 stamped every PDF with that date, so independent captures (Wayback, logged scrapes, filename dates) were the only way to reconstruct the sequence. Second, we report the sequence and let it speak; whether the posting pace was reasonable for a compressed eight-week process is a judgment we leave to readers.
The systemic point is not about any person or office. It is that a deliberative process and its official record ran on different clocks, and for the stretch that mattered most, the drafting of the final report, the public record trailed the deliberation it documents. An institution that wants its process to be auditable in real time needs its record to keep pace with its decisions, or needs someone outside the institution keeping the clock. This cycle, the latter is what happened.
The Five Ballot Questions
On November 3, New Yorkers will vote yes or no on five questions filed by the commission (exact texts, abstracts, full amendment text):
1 · Streamline the use of public space (ramps, benches, planters, cafes): revises revocable consent, removes the mandatory public hearing, modernizes notice, and merges the sidewalk and roadway cafe applications into one process the commission projects at about two months, against today's six to eight months for sidewalk cafes and three to five for roadway cafes.
2 · Reform the contracting process: procurement transparency, a simplified pre-contracting questionnaire, delegated contract approvals, and a simplified client-services plan, aimed at small businesses and nonprofits.
3 · Speed up street safety projects, office leasing, and activation of city land: simplifies review of street safety projects and fast-tracks the sale or lease of modest city properties, air rights over city-owned landmarks, and office leases.
4 · Simplify building permitting: a centralized construction permitting hub (one digital front door across the more-than-a-dozen agencies involved), broader certificate-of-occupancy authority within DOB, and waterfront permitting consolidated into DOB.
5 · Require contributions to the rainy day fund: sets a reserve target of 12% of city tax revenue, directs a contribution formula developed in consultation with the Comptroller, and aligns withdrawal criteria with state law.
How to Know If They Worked: A Measurement Framework
Each question makes an implicit empirical promise. Each promise can be checked against data the city already publishes, and in every case the checking also reveals a gap worth closing. Datasets below are on NYC Open Data unless noted; we verified all of them live in August 2026.
The framework's working rule is triangulation: no conclusion rests on a single dataset. Each question pairs an official aggregate against a computation anyone can rerun from raw records (the Comptroller's late-registration rate against Checkbook's per-contract dates), an activity count against a timing measure (cafe locations against the still-missing application clock), an inventory against the parcel-level flow beneath it (COLP's total against the BBLs that actually leave it), or a volume against a cohort's behavior (jobs approved against the share of each class that moves to permits). When the signals agree, the conclusion is sturdy; when they disagree, the disagreement is itself the finding, as the late-2024 COLP cliff shows. The tables below hold the baselines as we recorded them, before the vote.
The promiseCafe permitting drops from six-to-eight months to about two; public input survives the loss of the mandatory hearing.
Baselines todayThe commission's own processing-time figures (6–8 months sidewalk, 3–5 roadway); 2,056 locations in Dining Out NYC Locations, updated daily; permitted street events in NYC Permitted Event Information.
| Period | Events | Change |
|---|---|---|
| 2022 | 5,888 | post-COVID peak |
| 2023 | 4,686 | -20.4% |
| 2024 | 4,672 | -0.3% |
| 2025 | 4,811 | +3.0% |
| 2026 Jan-Jul | 2,177 | +0.6% vs 2025 Jan-Jul |
Success metricsMedian application-to-approval days for the merged cafe process; count and borough distribution of active cafe consents (equity check: do gains reach beyond Manhattan); volume of DOT's promised “alternative methods” of public input. A second, ready-made measure: permitted street-activity events (block parties, street events, festivals, plaza events, and similar) with their year-over-year change. The baseline is flat: 4,686 events in 2023, 4,672 in 2024, and 4,811 in 2025, a trend inside plus-or-minus 3%, with like-for-like January–July counts of 2,164 in 2025 and 2,177 in 2026. If streamlined consents open the public realm, post-passage years would need to break above that flat line; the events data is an adjacent activity signal, though, since Question 1 amends consents for fixed structures rather than one-day event permits.
The gapThe two-month promise cannot be verified with today's data, and it used to be verifiable: until August 2023 the city published sidewalk-cafe applications milestone by milestone, submission date, community board step, public hearing, council step, and issuance, in a dataset frozen when the program moved to DOT. Its replacement records only the date a license issues. Revocable consents, the instrument Question 1 amends, have no dataset at all, just individual hearing notices in the City Record. If the question passes, DOT resuming application-level publishing is the first accountability ask.
The promiseLess burden, more transparency, faster internal approvals.
Baselines todayMOCS itself testified that end-to-end procurement runs 312 days for bids and 387 for proposals. The Comptroller's office reports that 90.7% of human-services contracts arrived late for registration in FY24, up from 88.5% in FY23, with nearly 40% more than a year late. The Recent Contract Awards feed captured 3,203 award notices totaling roughly $25.5 billion in FY26.
| Measure | Baseline | Source |
|---|---|---|
| End-to-end procurement, bids | 312 days | MOCS testimony |
| End-to-end procurement, proposals | 387 days | MOCS testimony |
| Human-services contracts registered late | 88.5% FY23 to 90.7% FY24 | Comptroller |
| All expense contracts late, illustrative FY26 pull | 320 of 500 | Checkbook NYC API |
Success metricsShare of contracts registered after their start date (the retroactivity rate); days from solicitation to registration by agency and method; interest paid on late payments; MWBE award share. Sources: Checkbook NYC, the Comptroller's vendor contracting and payment dashboard, and the annual contracts summary reports. The lateness test itself is public and computable today: Checkbook NYC's API exposes each registered contract's start date and registration date, so a contract registered after its start date can be scored automatically; in an illustrative pull of 500 FY26 registered expense contracts for this report, 320 carried a registration date after their start date, several by more than four years.
The gapMOCS's Agency Procurement Indicators stopped appearing on the open data portal after FY15; the current indicators live in PDF reports. The single most useful transparency step this question could trigger is per-contract milestone data published in open, sortable form, which is what the human-services witnesses asked for at Staten Island.
The promiseDangerous intersections get fixed faster; surplus property stops sitting idle.
Baselines todayRoughly a thousand Vision Zero street improvement corridor projects recorded since the program began (SIP Corridor and companions); 85,546 crashes, 49,634 injuries, and 229 deaths citywide in 2025 (Motor Vehicle Collisions); 3,089 city-owned parcels categorized “property with no use” in City Owned and Leased Property, the surplus inventory the fast-track targets.
| Release | No-use parcels | BBLs that left since prior |
|---|---|---|
| Jun 2022 | 3,562 | — |
| Apr 2023 | 3,543 | 30 |
| Jun 2024 | 3,489 | 46 |
| Dec 2024 | 3,189 | 332 (reclassification-scale) |
| Jan 2026 | 3,132 | 50 |
Success metricsStreet improvement projects completed per year and, if the city will publish it, design-to-construction duration; injuries and deaths at treated locations before and after (the collisions data supports this directly); the count of no-use parcels moving to disposition or activation per year; time-to-lease for city office space. The no-use inventory now has a usable baseline: City Planning archives every COLP release, and the no-use count fell from 3,562 parcels in mid-2022 to 3,132 in the January 2026 release, a drift of roughly 2% a year, with one exception, a drop of about 300 parcels in late 2024 that shows how reclassification can move the total without any land changing hands. The defensible measure is parcel-level: in a normal release, 14 to 69 specific parcels leave the no-use category; if Question 3's fast-track works, that flow would need to rise and stay above that range.
The gapProject-level timeline data for street redesigns is thin; the Capital Projects Dashboard covers budgets and schedules at a coarser grain. A before/after safety evaluation is the strongest test this framework proposes, and the data for it already exists.
The promiseOne digital front door replaces a maze of more than a dozen agencies.
Baselines todayFor jobs approved in the first half of 2026 in DOB NOW job filings, 69,141 of them, the median filing-to-approval time inside DOB was 7 days, the median filing-to-first-permit time was 27 days, and the 90th percentile of first permits was 150 days. Those figures cover DOB's own pipeline only. A job, in DOB's vocabulary, is a construction project's application to the department, the filing that must be approved before the project's work permits can issue.
| H1 | Jobs approved | Permits issued |
|---|---|---|
| 2023 | 63,059 | 79,148 |
| 2024 | 70,337 | 83,646 |
| 2025 | 69,064 | 83,797 |
| 2026 | 69,142 | 86,350 |
| Class | Within 30 days | 90 days | 365 days |
|---|---|---|---|
| H1 2023 (n=63,059) | 30.8% | 41.9% | 51.5% |
| H1 2025 (n=69,064) | 35.4% | 45.2% | 53.9% |
Success metricsToday no dataset captures a project's full journey across DEP stormwater review, Parks tree protection, FDNY, DOT, and DOB; the fragmentation the question targets is mirrored in fragmented data. If the hub is built, its own logs become the measure: median end-to-end days from first submission to final permit across all agencies. Interim proxies: the DOB NOW medians above; approval and permit volume, which has its own baseline (job approvals have run flat near 69,000 to 70,000 per half-year since 2024, while work permits issued in the first half of the year rose from 79,148 in 2023 to 86,350 in 2026) and is a supporting signal only, since construction demand moves with the economy and rising volume counts as success only alongside falling timelines; and certificate-of-occupancy issuance (DOB NOW CO). A more granular test triangulates the three DOB datasets, following each project by its job number from filing through work permits to certificate of occupancy, and scoring each half-year's approvals as a class: of jobs approved in early 2023, 30.8% pulled a work permit within 30 days of approval; for the early-2025 class it was 35.4%. That share, a cohort conversion rate, moves with process speed rather than the construction cycle, and it was already drifting upward before the ballot; post-passage classes would need to beat that drift, not just continue it.
The gapThe hub should be born with a public performance dataset. Writing that expectation into implementation, not the charter text, is where advocacy attention could go next.
The promiseThe city saves by rule, not by mood.
Baselines todayThe Revenue Stabilization Fund holds about $1.97 billion, has had no target balance and no mandated deposits since its 2020 creation, and the FY27 preliminary plan proposed drawing $980 million, roughly half of it. The question's 12%-of-tax-revenue target sits inside the range the Comptroller's own proposal argued for (10% minimum, 16% goal). Against tax revenues on the order of $80 billion, the target implies a reserve near $9–10 billion, roughly five times today's balance.
| Measure | Value |
|---|---|
| Fund balance, Aug 2026 | about $1.97B |
| Charter target, 12% of tax revenue | about $9-10B |
| FY27 preliminary plan proposed draw | $980M |
| Comptroller proposed range | 10% floor, 16% goal |
Success metricsThe fund's balance as a share of tax revenue, year over year, against the 12% target; whether the contribution formula is adopted on schedule and deposits actually follow it in each adopted budget; withdrawal events checked against the stated criteria. Sources: the Comptroller's quarterly cash reports, the annual comprehensive financial report, and the State Comptroller's plan reviews.
The gapThe fund's balance is published in PDFs and press releases, not as an open, downloadable series on the data portal. A one-table dataset would make this the easiest question of the five to hold accountable.
Cross-cutting
Passage itself is measured first: certified results from the NYC Board of Elections (vote.nyc) after November 3. Beyond passage, two lenses apply to every question: equity (do faster cafes, permits, and dispositions distribute across boroughs and communities, all of the above datasets carry geography) and the counter-theme from the hearings (do the council's checks and balances and public-notice functions demonstrably survive, which is a records question as much as a policy one).
What This Says About Governance Memory
The through-line of this case study is that every one of its findings required connecting records that no single institution keeps connected: testimony to footnotes, footnotes to ballot outcomes, hearing dates to transcript posting dates, charter promises to open datasets. None of the individual facts were hidden. The connections were simply nobody's job.
That is the case for governance memory systems in one sentence: the connections should be somebody's job. This cycle, a one-person research practice with a structured pipeline could hold a fifteen-member commission's full public record, check its final report against every recurring theme, reconstruct its publication timeline when server metadata failed, and hand the next phase, outcome measurement, a ready framework. The system's outcome-review scaffold now has its anchor: November 3, five yes/no questions, and the baselines above recorded before the vote, which is the only time baselines can be recorded.
Method and Limitations
Sources: official transcripts and minutes (nyc.gov), meeting videos, the preliminary and adopted final reports and appendices, Wayback Machine captures of the commission's meetings page, OCC's logged scrapes, and the NYC Open Data and Comptroller sources linked above. We queried all dataset figures in August 2026, and they reflect data as of that date; the DOB timing figures cover filings approved January–June 2026 and carry the usual caveats of administrative data (professional certification paths, still-pending filings not yet counted). Witness counts reflect OCC's theme attribution of spoken testimony at the ten public hearings, not an official tally; the 625 written submissions the final report counts are not included, and written testimony appears here only where the final report cites it or the witness published it. The commission has since published the submissions, with personal information redacted, in combined batches on its written public comment page; the batch files carry August 3 server dates, after the final report was adopted. Our analysis reflects the concerns most raised in testimony, not external polling or program data, and whether any measured change is attributable to a ballot question rather than concurrent reforms is a question the before/after designs here can inform but not settle. Correction (August 11, 2026): an earlier version of this report said the July 9 and July 12 Wayback captures showed two posted transcripts; re-reading the captures shows four, two of them under filenames that do not contain the word transcript, and the timeline above reflects the corrected counts. A second correction, same date: the DOB timing baseline first published here came from a truncated data pull, 4,588 of 69,141 approvals, and skewed high; the full-population figures of 7 days median to approval and 27 to first permit replace it. A third correction (September 4, 2026): an earlier version said the commission publishes no archive of its written submissions and put their number at 603; the submissions are now posted on the commission's written public comment page, and the count here uses the final report's own figure of 625.
Governance that remembers. Institutional Memory as a Service.
Have thoughts or feedback on this research?
Othman@occresearch.org