Warning
These personas and interviews are synthetic. They were generated as a
structured brainstorming device — not conducted with real people. No real
neighbor, property manager, judge, or acoustician said any of this. The panel
exists to pressure-test the product from every side of a noise dispute at once;
it is not evidence of demand, and it does not substitute for talking to
real users in a real dispute. Treat every "quote" as a hypothesis to validate,
not a finding. This is consistent with how the project labels its synthetic test
sessions (see audits/methodology-and-limitations.md).
Where a persona "values" something, it is mapped to a feature that actually exists in the repo today — no invented capabilities. Legal and acoustics claims below are summarized from cited public sources and are jurisdiction-dependent; nothing here is legal advice.
Last assembled: 2026-06-30.
A noise dispute is unusual among software problems: the people on every side of it are adversaries by default. The dog owner, the complaining neighbor, the property manager who has to choose between two tenants, the board, the skeptic who knows dBFS isn't SPL — each wants something different from the same six numbers per event. Role-playing the full cast forces the question this tool lives or dies on: does a levels-only, no-audio record actually help, given what ordinances and adjudicators say they will accept? The synthesis is tagged so it stays honest and doesn't become a wishlist.
Each card compresses a simulated interview to five lines: Goal · Values today
(real features) · Gets stuck · Wants next · Adopts / walks. "Values today" points
only at shipped behavior — the no-audio gate, the methodology/limitations section,
the quiet-hours violations export, calibration honesty, the day×hour heatmap, the
PWA, the accessible report. Frictions feed Remediations (R) and wishes feed
Expansions (E) in RESEARCH-ROADMAP.md.
- Sampling frame. Everyone a real barking/noise dispute pulls in: the person building a record (the dog owner defending against vague complaints; a renter accused of noise more broadly); the people who adjudicate (the complaining neighbor, the property manager/landlord, the HOA board / mediator); the people who decide whether to trust the number (an acoustics skeptic, an accessibility user who has to read the report at all); the people who vet the posture (a privacy advocate, a tenant-rights attorney); and the people who operate and build it (a Raspberry-Pi self-hoster, the owner/maintainer).
- Protocol. For each persona: a goal, a walkthrough of the surfaces they'd touch (the report, the violations export, the live meter, the PWA, the Pi service), what worked against the current implementation, where they'd stall, what they'd want next, and the one thing that makes them adopt or walk.
- Synthesis. Frictions → Remediations; wishes → Expansions. Each carries the personas who raised it, a rough priority, and an effort estimate, with a traceability matrix in the roadmap.
- Effort scale. S ≈ an afternoon · M ≈ a day or two · L ≈ a week+.
The cast's incentives, frustrations, and "is this even admissible" worries are modeled on public sources, not imagination. High-stakes legal/acoustics claims are cross-checked against ≥2 sources. Full evidence-to-item mapping is in the roadmap.
- What barking-nuisance ordinances actually ask for. Many jurisdictions require the complainant to keep a dated, timed log of barking and its duration, and several require a second household to corroborate before enforcement; common thresholds are framed as "incessant" (≈30 continuous minutes) or "intermittent" (≈60 minutes on-and-off in 24 h). See Contra Costa County Noisy Animal Ordinance, OC Animal Care barking-dog complaint process, LA Animal Services — Nuisance Barking, and Justia — Barking Dogs and Neighbors' Legal Rights. Critically for this tool: these sources weight a credible timed log, and some contemplate audio/video, but evidence rules are jurisdiction-specific.
- How tenants/PMs are told to document noise. Guidance converges on a clinical, timestamped log (date, exact start/end, duration, source description, impact) and a written paper trail; a decibel reading "adds objective weight." See Flex — apartment noise-complaint template, Local Noise Laws — documenting a noise complaint, and LawInfo — handling tenant noise complaints.
- How HOAs adjudicate. Boards enforce CC&R nuisance/quiet-hours provisions through a notice-and-hearing process (in CA, Davis-Stirling internal dispute resolution), and want precise, timestamped incidents. See MBK Chapman — California HOAs and noise complaints and Manning & Meyers — managing noise complaints in condominiums.
- Quiet enjoyment. The implied covenant gives tenants peaceful possession; a substantial interference (not a minor annoyance) can breach it and obligates the landlord to act. See Cornell LII — covenant of quiet enjoyment and Texas State Law Library — noise (landlord/tenant).
- Why no-audio is the whole design (recording law). Roughly a dozen states require all-party consent to record a private conversation, and an illegally captured recording is often inadmissible and can carry criminal penalties; a device left running in a home that captured audio could implicate household members, guests, or sound carrying from a neighbor. See RCFP Reporter's Recording Guide, Recording Law — two-party-consent states (2026), and Justia — recording phone calls & conversations, 50-state survey. (State counts/lists differ between sources — e.g., whether Delaware/Oregon are counted — which is exactly why the tool claims no jurisdiction-specific compliance.)
- Why dBFS ≠ SPL, and why uncalibrated readings are relative only. dBFS is referenced to digital full scale (always ≤ 0) and is not sound pressure level (dB SPL, referenced to 20 µPa); dB(A) is A-weighted SPL and is what noise regulations use. Uncalibrated phone/browser/Pi mics measure relative level and can be off by 5–15 dB or more. See Wikipedia — dBFS and AudioMasterclass — 0 dB vs 0 dBFS.
- What a "real" meter is (IEC 61672). Sound level meters are graded Class 1 (precision) vs Class 2 (general purpose) with defined tolerances and frequency ranges; environmental-noise regulations often specify Class 1, occupational rules accept Class 2. See SoftdB — Class 1 vs Class 2, IEC 61672-1:2013, and NoiseMeters — type/class definitions.
- Phone/app accuracy in practice. NIOSH found that of 192 evaluated apps, only a handful met a ±2 dB(A) criterion vs a Type-1 reference; its own app is "accurate within ±2 dB(A)" tested in a reverberant chamber and only meets IEC 61672 Type 2 with a calibrated external microphone, and NIOSH does not position it for compliance. See NIOSH/CDC — Sound Level Meter App and Kardous & Shaw 2014 — evaluation of smartphone sound apps (PMC).
| # | Persona | Group | Primary goal | Top friction |
|---|---|---|---|---|
| P1 | Chelsea — dog owner, primary user | Log & Defend | Show the real pattern when a neighbor complains about Olive | Worries the report reads as "she built a case" not "here's the truth" |
| P2 | Marcus — renter accused of general noise | Log & Defend | Push back on a vague "you're too loud" with data | His noise isn't a dog; wants the same record for footfall/music |
| P3 | Diane — downstairs complaining neighbor | Adjudicate | Be believed that it's genuinely disruptive at night | A device run by the accused looks self-serving to her |
| P4 | Raymond — property manager / landlord | Adjudicate | Decide fairly between two tenants without picking sides | Can't tell if a tenant's log is honest or cherry-picked |
| P5 | Yolanda — HOA board member & mediator | Adjudicate | Apply the CC&R quiet-hours rule defensibly | Needs timestamped incidents she can take to a hearing |
| P6 | Dr. Petra Voss — acoustical consultant / skeptic | Trust the Measurement | Refuse to let dBFS masquerade as SPL | "This isn't a Class-1 meter; the number is relative" |
| P7 | Greg — accessibility user of the report (screen reader) | Trust the Measurement | Actually read the report the PM forwarded him | Charts are useless to him if there's no text equivalent |
| P8 | Nadia — privacy & digital-rights advocate | Assure & Audit | Confirm nothing is recording her or the neighbors | Most "noise apps" quietly record audio |
| P9 | Tom — tenant-rights / legal-aid attorney | Assure & Audit | Know whether this helps or hurts a client's case | Worries clients over-read a chart as proof of source |
| P10 | Sam — Raspberry-Pi maker / self-hoster | Operate / Build | Run it unattended and trust it's local-only | Wants proof it never phones home |
| P11 | Chelsea (maintainer hat) — owner/maintainer | Operate / Build | Keep the no-audio gate un-breakable across changes | One careless commit could add an audio path |
- Goal: when the downstairs neighbor complains about Olive, replace "she's always barking" with an honest, time-stamped picture of when sound actually crossed a threshold and for how long.
- Values today: the no-audio guarantee (she can run it in her own apartment without recording herself, guests, or anyone next door); the quiet-hours violations export that lists every event with a within/outside flag so it can't be a cherry-picked subset; the day×hour calendar heatmap that shows the real pattern; the methodology + limitations section that makes it read as honest, not adversarial.
- Gets stuck: the limitations are present but buried — she worries a stressed neighbor or PM skims the charts and misses "this can't prove Olive made the sound." She's not sure what threshold/quiet-hours to set for her building's rule.
- Wants next: a plain-language "what this can and cannot prove" cover page on the export; a way to map quiet-hours to her actual lease/ordinance; one-tap "send to property manager" from her phone.
- Adopts if: the report visibly protects her and the neighbor from being over-read. Walks if: it ever looks like a tool for building a case against someone — that's the opposite of why she made it.
- Goal: his upstairs neighbor reported "constant stomping and bass." He wants the same honest record — was it actually loud, when, for how long — to discuss calmly with management.
- Values today: the tool measures levels and event metadata only, so it works for footfall/music exactly as it does for barking; the PWA means he doesn't need a Raspberry Pi; local-only / no egress means his data isn't going anywhere.
- Gets stuck: defaults are tuned for barking-ish events (threshold −35 dBFS, min-duration 0.4 s, 1.0 s debounce); he's unsure how to tune for sustained music vs short thumps. No guidance for "my noise is different from a dog."
- Wants next: tuning presets for a few noise types; a clearer live meter read on the PWA while he adjusts; a neutral framing so management doesn't read it as him building a counter-case.
- Adopts if: he can produce something that makes the conversation factual. Walks if: tuning is guesswork and the numbers feel arbitrary.
- Goal: be taken seriously that the barking genuinely wakes her — she's not exaggerating, and she resents being framed as the unreasonable one.
- Values today: that the export is honest by construction — it shows all events, not a flattering subset; that it explicitly disclaims source attribution, which (counterintuitively) makes her trust it more, because it isn't pretending to be something it's not.
- Gets stuck: a record produced by the accused party's own device reads as self-serving to her; she has no way to corroborate with her own experience, and the ordinance she looked up seems to want her log too.
- Wants next: a way to align her own dated log against the device's timeline; an explanation, in language she'd accept, of why there's no audio (right now it looks like the data is conveniently incomplete).
- Adopts if: the framing acknowledges her side and lets her corroborate. Walks if: it's presented as "proof the neighbor is fine," which would harden the dispute.
- Goal: adjudicate between two tenants without becoming the bad guy to either, and without exposing himself to a quiet-enjoyment complaint from the side he doesn't side with.
- Values today: a timestamped, reproducible record beats "he-said-she-said"; the violations CSV/HTML is exactly the format his documentation guidance asks for (date, start/end, duration); the measurement-conditions / lineage section tells him where and how it was measured.
- Gets stuck: he can't tell whether a tenant edited the SQLite file or moved the device to a louder spot; "within quiet hours" sounds like a verdict but is only "a level crossed a threshold here." He needs to defend his decision if challenged.
- Wants next: a tamper-evident/signed export he can trust; a one-line "what this does and doesn't establish" he can paste into a notice; the ability to weigh it alongside the complaining household's own log.
- Adopts if: it makes his decision more defensible, not less. Walks if: a tenant can quietly fake it, or it invites a tenant to over-claim.
- Goal: apply the community's quiet-hours covenant consistently and survive a hearing or internal dispute-resolution step.
- Values today: configurable quiet-hours she can set to the CC&R window (default 22:00–08:00, wraps midnight); the honest export that won't get torn apart for cherry-picking; the methodology section that pre-empts "your evidence is junk."
- Gets stuck: she needs incident counts and durations mapped to how the covenant is written (e.g., "X minutes during quiet hours over N nights"), and a neutral tone so a board action doesn't look like harassment of one homeowner.
- Wants next: an ordinance/CC&R-style rollup (duration-in-window per day) that stops short of declaring a violation; a mediator-friendly "for both parties" framing; guidance that the device's number is informational, not a regulatory measurement.
- Adopts if: it helps her be fair and defensible. Walks if: it pushes a verdict the board can't actually stand behind.
- Goal: stop a relative number from being paraded as an absolute one. If it's not a calibrated Class-1 (or at least Class-2) measurement, it must not read like dB(A).
- Values today: that the project already says this out loud — uncalibrated
dBFS is relative, not SPL; calibration offset defaults to 0.0 and the report
states so; the limitations list "relative, not absolute," "no source
attribution," and "placement-dependent" verbatim; the calibration helper
(
olive-calibrate) exists and the session lineage records placement/offset. - Gets stuck: a non-expert reader still sees a chart of "dB-ish" numbers and infers loudness in the regulatory sense; the calibration path needs a reference meter most users don't own, and even a NIOSH-grade phone app is only ±2 dB(A) and only Type 2 with an external calibrated mic — she wants that uncertainty shown, not implied.
- Wants next: an unmissable "uncalibrated → relative only" banner; calibration that records the reference instrument's make/class in the session; an honest uncertainty band rather than point values; the coarse bark/ambient "hint" kept clearly hedged or hidden in adjudication exports.
- Adopts if: the tool never lets a relative number impersonate SPL. Walks if: a single screen implies courtroom-grade measurement.
- Goal: read the report the property manager forwarded — he's a tenant in the building too, and the charts are meaningless to him without text.
- Values today: the report is built to WCAG 2.2 AA: every chart
<figure>has a data-table equivalent with a caption, every<svg>carriesrole="img"+ anaria-label, the heatmap prints the count as text in each cell (meaning never depends on color), there's a skip link, one<h1>, logical headings, visible focus, and a structural a11y gate plus pa11y/axe in CI. - Gets stuck: the strong a11y story is documented for the HTML report; if the PM exports/prints a PDF and forwards that, he's not sure the table equivalents and tags survive. The violations export specifically hasn't been walked with a screen reader in the audit.
- Wants next: confirmed a11y parity for the PDF/print path and the violations export; a screen-reader walkthrough artifact for the export, not just the main report.
- Adopts if: he can extract every number without sighted help. Walks if: the forwarded artifact is an untagged image of a chart.
- Goal: verify that a device marketed as "listens in your home" genuinely cannot record anyone — the bar most "noise monitor" products fail.
- Values today: the no-audio guarantee proven by construction — frames are
reduced to a dBFS number in memory and discarded, there is deliberately no API that
writes a frame, the SQLite schema has no column that can hold audio, and
merge-blocking tests (static
wave/soundfile/.tobytesscan,"wb"AST scan, schema introspection) fail the build if that changes; plus no-egress proof (import scan + asocketbooby-trap) and asystemdunit withPrivateNetwork. - Gets stuck: all of this is real but lives in audit docs; the reader of a report (a neighbor, a PM, a judge) has no idea the no-audio choice is a deliberate privacy/recording-law posture rather than a missing feature.
- Wants next: a short, reader-facing "why there is no audio" note embedded in the report itself; the no-audio/no-egress guarantees surfaced where a layperson sees them.
- Adopts if: the privacy posture is legible to non-engineers, not just provable in tests. Walks if: marketing ever softens "never records" into "doesn't usually."
- Goal: advise a client whether running this helps or hurts — and make sure nobody over-reads a chart in a way that backfires.
- Values today: the relentless "informational, not adversarial" framing; the recording-law rationale (level-only sidesteps two-party-consent / eavesdropping exposure because there is no captured conversation to consent to, subpoena, or leak); the explicit "no source attribution" and "not legal compliance in any jurisdiction" disclaimers.
- Gets stuck: ordinances and adjudicators vary wildly — some want a corroborating second household, some contemplate audio/video, all are local; a client could wave a heatmap and say "this proves it," which is exactly what the tool says it does not do.
- Wants next: the jurisdiction-dependence stated plainly in the export; a reader-facing limits/recording-law note; an option to suppress the bark/ambient "hint" in anything shown to an adjudicator; clarity that "within quiet hours" ≠ "violation."
- Adopts if: it lowers the chance a client overclaims. Walks if: the output invites someone to treat a relative level as legal proof of who did what.
- Goal: run it headless on a Pi in his own unit, unattended, and know it's local-only and resilient.
- Values today:
scripts/setup-pi.sh+ the hardenedsystemdunit (PrivateNetwork,ProtectSystem, auto-restart);resilient_sourcereconnect with capped backoff; the atomic heartbeat JSON for a watchdog; WAL durability,user_versionschema migrations, and config-driven retention pruning; a zero-runtime-dependency pure-Python core (only the optionalliveextra needs PortAudio). - Gets stuck: calibrating without a reference meter; knowing the device hasn't silently drifted or been bumped to a louder spot; reading the report on a headless box.
- Wants next: a guided calibration flow (ideally recording the reference instrument used); a placement/coverage health check surfaced from the existing frame-coverage accounting; an easy way to ship the HTML report off the Pi for viewing without adding network surface.
- Adopts if: it's boring and trustworthy and provably offline. Walks if: it needs cloud anything, or coverage problems hide silently.
- Goal: keep the central design gate — no audio, ever — unbreakable as the code evolves, and keep the honesty (limitations, calibration) non-optional.
- Values today: the gate is enforced by construction and by CI: no-audio static
- schema tests, no-egress tests, a report-content test that asserts the methodology + limitations text is present, deterministic SVG reports with snapshot tests, a structural a11y gate, and coverage thresholds — all merge-blocking. The PWA is a parallel implementation with its own Node tests sharing the same guarantees.
- Gets stuck: keeping Python and the PWA semantically in sync; making sure new surfaces (exports, the heatmap, future features) inherit the limitations text and the a11y/data-table rules rather than re-deriving them; the limitations are canonical in one doc but rendered in several places.
- Wants next: a single canonical source for the limitations/recording-law text that every surface (report, violations export, PWA, a future cover page) pulls from; parity tests across Python/PWA; a contributor note that any new export must carry the limitations and pass the a11y gate.
- Adopts if: it stays impossible to ship an audio path or a limitations-free artifact. Walks if: the guarantees become promises in a README instead of tests.
- The limitations are correct but buried. The single most-repeated friction — from the owner (P1), both adjudicators (P4, P5), the skeptic (P6), the advocate (P8), and the attorney (P9) — is that the honest caveats live in a methodology section a stressed layperson skims past. The fix is to lift "what this can and cannot prove" to a cover page on every export. Cheap; highest-leverage.
- "Relative, not SPL" must be unmissable. P6 is the loudest, but P4 and P9 also risk reading a dBFS chart as regulatory loudness. The repo already says it; the gap is prominence and an honest uncertainty story, plus calibration that records the reference instrument. (dBFS vs SPL; NIOSH app ±2 dB / Type 2 only with external mic.)
- No-audio is a feature to advertise to readers, not just a test. P8 and P9 note the privacy/recording-law rationale is invisible to the neighbor/PM/judge who reads the report. Surfacing a plain-language "why there's no audio" turns the constraint into a trust signal. (RCFP; Recording Law.)
- Adjudicators need it to be tamper-evident and corroborable. P4 and P5 can't act on a record the accused could have edited; P3 and the ordinances want a second household's log to count. A signed/hashed export and a "bring your own log to align" affordance address both. (ordinance two-household practice.)
- Map to the rule, but never declare the verdict. Everyone in Group B wants the numbers expressed against their rule (a duration-in-quiet-hours rollup), and everyone in Groups C/D insists "within quiet hours" must stay "a level crossed a threshold here," not "a violation occurred." The roadmap must do both at once.
- Accessibility can't stop at the HTML. P7 reminds us the strong a11y story is audited for the main HTML report; the forwarded artifact (PDF/print, violations export) is what an adjudicator actually shares, and it needs the same parity.
This is simulated. It can surface plausible needs and obvious gaps from many sides of
a dispute, but it cannot tell you which are real, how many people would actually
use this, or whether any ordinance/PM/board near you would accept the output. It
over-represents the author's mental model and the public sources cited, and it will
miss what only real disputants surprise you with. Several findings hinge on
jurisdiction-specific law (recording consent, what evidence is admissible, how a
board must give notice) — the cited sources illustrate the shape of these rules, not
the rule where any given user lives. Do not prioritize a roadmap off this panel
alone. Use it to design questions for, and lower the cost of, real conversations
with a tenant in a dispute, a property manager, an animal-control officer, and an
acoustician. Triaged, evidence-tagged backlog: RESEARCH-ROADMAP.md.