Skip to content

USER-RESEARCH.md

Latest commit

 

History

History
471 lines (428 loc) · 31.5 KB

USER-RESEARCH.md

File metadata and controls

471 lines (428 loc) · 31.5 KB
Raw

User research: synthetic personas and simulated interviews

Warning

These personas and interviews are synthetic. They were generated as a structured way to pressure-test the scorecard from every stakeholder angle at once. No real person said any of this. Each "quote" is a hypothesis to validate with a real user, not evidence of demand. This is consistent with how the project labels its own synthetic and provisional material (see the "re-verify" notes in expansion.md and the adversarial-verification method in expansion-research.md). Do not prioritize a roadmap off this document alone; use it to design the questions for real discovery.

Last assembled: 2026-06-30.

Why do this at all

The scorecard already serves two named users well (the inherited-feed transit manager and the program liaison). But it now ships a wide surface, NTD readiness, a conformance mark, a public API, an accessibility-coverage view, an equity overlay, a national realtime monitor, that touches stakeholders the two-user framing never interviews. Role-playing the full cast forces the question "who is each surface actually for, and where would they stall?" The synthesis lives in RESEARCH-ROADMAP.md, tagged so it complements the existing roadmap docs rather than restating them.

Method

  • Frame. Everyone who touches a GTFS feed's quality, or the grade the scorecard puts on it: the agency that owns the feed, the vendor that produces it, the standards bodies that define "good," the apps and analysts that consume it, the oversight and audit roles that hold agencies to it, and the people who operate the tool. Personas are composite archetypes of these real segments.
  • Protocol. Each card carries a goal, then a four-to-five line simulated interview: Values today (mapped only to features that actually exist in the repo), Gets stuck, Wants next, and Adopts / walks. Frictions feed the remediation backlog; wishes feed the expansion backlog in RESEARCH-ROADMAP.md.
  • Research basis. Persona needs and the high-stakes claims behind them were checked against primary sources (access date 2026-06-30). The full cited evidence base is in RESEARCH-ROADMAP.md; the load-bearing ones:
    • The NTD GTFS obligation that gives a small agency a non-optional reason to care: FTA requires fixed-route NTD reporters to publish and maintain a public GTFS feed from Report Year 2023 (FTA, RY2023 final rule), and the RY2026 policy requires a stable agency_id for every reporter represented in the feed, crosswalked to the five-digit NTD ID on P-50; the two values do not have to be equal (FTA, 2026 NTD Full Reporting Policy Manual).
    • The concrete feared consequence behind freshness: an expired feed stops a trip from appearing in Google Maps (Google Transit Partners, general errors, keep your feed from expiration).
    • That a validator-clean feed can still be rider-wrong, which is the gap the plain-language layer fills: MobilityData's grading scheme notes a feed "flagged as valid by an automated validator may contain undetected qualitative errors that are unsuitable for rider-facing purposes" (MobilityData GTFS Grading Scheme).
    • That errors are real but concentrated, so a per-code (not per-instance) rubric is defensible: 21% of 632 US feeds had at least one validator error, and ten error types accounted for 90% of occurrences (Devunuri & Lehe, Findings, 2024).
  • Honest tag. Where a persona's wish matches something already shipped or already planned, the roadmap marks it [corroborates …]; only genuinely new asks are [NET-NEW]. Independent triangulation onto an existing plan is a signal, not noise.

How to read a persona

Each card is one role compressed to its decision: what they get from the tool as it stands, the wall they hit, the next thing they would ask for, and the single condition that flips them to adopting or walking away.

Persona roster

# Persona Group Primary goal Top friction
A1 Dolores — manager, 18-bus city system, inherited the feed Fix My Feed Know if the vendor's feed is any good without learning GTFS The official monthly report is technical; she can't tell what to do first
A2 Ray — coordinator, rural + tribal dial-a-ride, fare-free Fix My Feed Not get dinged for running flexible, fare-free, seasonal service Worries a demand-response feed reads as broken or empty
A3 Priya — agency contracts/procurement officer Fix My Feed Write feed quality into the next vendor solicitation No language to require a quality bar a vendor must hit
B1 Marcus — engineer at a GTFS vendor serving many small agencies Produce & Vendor Ship feeds that pass before a client or app rejects them A public grade on his clients' feeds could read as blame
C1 Lena — MobilityData-style standards steward Steward the Standard Keep the ecosystem on the canonical validator and spec Needs assurance the tool tracks, not forks, the rules
C2 Hiro — state-program GTFS data steward (Cal-ITP-style) Steward the Standard See the grade map to the guideline agencies are held to Wants the rubric legible against the state's own checklist
D1 Sam — ingestion engineer at a trip-planner app Consume the Data Decide which feeds are safe to ingest, at scale Needs one machine-readable pull, not one request per agency
D2 Aisha — MPO / regional planner and modeler Consume the Data Use feed quality + equity to target where help goes Quality and need data live in different places
D3 Tomas — academic transit-data researcher Consume the Data Study how feed quality changes over time, citeably No stable, versioned, citable reference for the dataset
D4 Gloria — rider, no car, screen-reader user Consume the Data Trust that her bus shows up correctly in her app She never sees the feed; she only feels it when it breaks
E1 Frank — FTA / state-DOT NTD oversight staff Assure & Audit See which reporters meet the federal GTFS obligation No NTD-keyed national readiness view
E2 Naomi — transit / disability-access advocate Assure & Audit Push agencies to publish accessibility data riders need Hard to show coverage gaps without shaming small agencies
E3 Wei — accessibility specialist auditing the UI + VPAT Assure & Audit Confirm the AAA / 508 claim is real, not asserted The VPAT's functional-performance log is still unfilled
E4 Dana — journalist comparing agencies Assure & Audit Report a checkable claim about local transit data A comparison view risks becoming a small-agency leaderboard
F1 Ramona — program liaison / customer-success manager Operate Walk into an agency call knowing the three things to raise Cohort prep is manual; no saved notes or call export
F2 Chelsea — owner / maintainer Operate Keep it cheap, accessible, and worth a second visit Most code is built; the gating work is human and operational

Group A — Fix My Feed (the agency improving its own feed)

A1 — Dolores, manager of an 18-bus city system who inherited the feed

  • Goal. Find out whether the GTFS her predecessor's vendor set up is healthy, and what to fix first, without becoming a GTFS expert.
  • Values today. The plain-language letter grade with four category scores and the "top 3 things to fix" framed as fixes with effort hints; the NTD certification-readiness read (published, valid, current, agency_id present); the notice-to-fix knowledge base (/fix/<code>/) that turns a validator notice into one setting to change. "The state's monthly report lists foreign_key_violation and a 24-item checklist. This tells me my feed expires in 19 days and that's the thing that drops me from Google Maps."
  • Gets stuck. The fix pages describe the change generically; she does not know which button in her specific scheduling tool does it. She is not sure the alert will reach her before the feed lapses, and there is no obvious way to turn alerts on from the page. She half-expects a low grade to feel like a judgment on her agency.
  • Wants next. Per-vendor fix instructions that name the exact export setting in her tool; a self-serve "watch this feed, email me before it expires" button; a one-line "will riders still see me?" statement tying the grade to Google / Apple ingestion; a board-meeting one-pager she can paste into an agenda.
  • Adopts if it tells her something before it bites and gives her the fix in her own tool's words. Walks if it reads like the official report, one more technical artifact she cannot act on.

A2 — Ray, coordinator of a rural and tribal dial-a-ride, fare-free

  • Goal. Publish good data for demand-response, fare-free, seasonal service without the tool treating "different" as "broken."
  • Values today. The service_type: seasonal / demand_response handling so a between-seasons gap is scored fairly; fare_free: true crediting the fare component instead of docking it; GTFS-Flex awareness (ADR 0007) that checks whether a rider can actually book a trip; the neutral "Not yet published" for realtime rather than a zero. "Most tools assume a fixed-route city bus. This one knows a fare-free dial-a-ride isn't a failing feed."
  • Gets stuck. He builds GTFS by hand (National RTAP's free GTFS Builder) and is not sure the scorecard's framing reaches a tribal agency that shares a regional feed or has an FTA waiver. The flex booking check can read as a demand for fields his small operation does not produce.
  • Wants next. A path that recognizes a shared regional feed and a waived reporter without flagging them; the National RTAP support channel surfaced as a resource on the page; flex "how to book" rendered for the rider, not just checked for the producer.
  • Adopts if the tool stays neutral about service that is legitimately different. Walks if "incomplete" gets confused with "non-standard."

A3 — Priya, agency contracts and procurement officer

  • Goal. Make the next vendor contract require a feed quality bar, so the agency stops inheriting feeds nobody can vouch for.
  • Values today. The /procurement/ copy-paste RFP/contract clause; the conformance mark (valid, current, accessible) as a bright-line credential a contract can name; the CI Action / GitHub Marketplace gate that fails a build below a min-grade so a vendor proves quality before delivery.
  • Gets stuck. She needs the clause to reference a bar a vendor can be held to objectively, and a way to verify a delivered feed against it without running Java herself. She is unsure whether the conformance mark is recognized outside this tool.
  • Wants next. Clause language tied to the conformance mark and to the state guideline by name; an acceptance-test recipe (scorecard try --min-grade) she can hand a vendor as a deliverable gate; a dated certificate artifact for the contract file.
  • Adopts if it makes a vendor's quality contractually checkable. Walks if procurement does not recognize the output as a standard.

Group B — Produce & Vendor (those who generate the feeds)

B1 — Marcus, engineer at a GTFS vendor serving dozens of small agencies

  • Goal. Catch feed problems before a client, an app, or a state report does, across all the agencies he produces for.
  • Values today. The CI Action he can drop into his own publish pipeline to gate on grade and days-to-expiry; scorecard try for an instant grade on a zip before it ships; the rule links that point each finding at the canonical validator notice or GTFS Best Practice, so the fix is unambiguous; the badge a happy client can embed. "I can fail my own build on a bad feed before it reaches the agency's site."
  • Gets stuck. The vendor-accountability signal and the stale-feed-by-vendor view are framed for the program, not for him; a public grade on his clients' feeds could read as naming-and-shaming the vendor. He wants the same view turned constructively toward fixing, not toward exposure.
  • Wants next. A vendor-facing roll-up of all the feeds he produces, framed as a worklist; an auth-aware adapter so he can gate hosted/keyed feeds in CI; early warning when a software update of his quietly breaks one field across many client feeds at once.
  • Adopts if it helps him fix his clients' feeds before anyone else notices. Walks if the vendor view becomes a public blame board.

Group C — Steward the Standard (the ecosystem keepers)

C1 — Lena, MobilityData-style standards steward

  • Goal. Keep the community on one canonical validator and one spec, and not see a fork that diverges quietly.
  • Values today. The hard guardrail that the scorecard scores on top of the canonical validator and does not re-validate GTFS; the rule links that send every finding back to the validator rules page, gtfs.org best practices, or the spec reference; the Mobility Feed API reuse (ADR 0011) that skips re-validating identical bytes; the explicit crosswalk to the GTFS Grading Scheme's seven fields. "It cites our notices and our version. It's a layer, not a competing validator."
  • Gets stuck. Wants to know the pinned validator version (v8.0.1) and how fast the tool adopts a new release and new notices; worries a "grade" could be read as a competing authority to the validator's own output.
  • Wants next. A visible validator-version stamp and changelog on the methodology; a clear statement that the grade is interpretation, not certification; an upstream path to feed real-world fix patterns back to the community knowledge base.
  • Adopts if the tool stays a faithful, versioned layer over the canonical tools. Walks if it drifts into re-implementing or contradicting the validator.

C2 — Hiro, state-program GTFS data steward (Cal-ITP-style)

  • Goal. Give agencies a grade that maps cleanly to the state guideline they are actually measured against.
  • Values today. The crosswalk.md / "how this agency maps to the standards" section tying the four categories to the California Transit Data Guidelines v4.0, the Minimum GTFS Guidelines, the Grading Scheme, Google/Apple, and the NTD obligation; the rubric's anchoring to v4.0 compliance tiers; the honest note that the grade is "a data-quality lens, not the official compliance determination."
  • Gets stuck. The crosswalk is California-deep; agencies in states that run a GTFS program but no quality rubric (Colorado, Michigan, Minnesota, Oregon, Washington) get the program shown only as a resource, not as a bar the score maps to. The official monthly report and the scorecard tell overlapping but differently-shaped stories.
  • Wants next. Per-state guideline profiles so the rubric cites the right authority per state; an explicit alignment with the monthly report so a manager is not confused by two numbers; a partnership posture toward the state program rather than a parallel one.
  • Adopts if it amplifies the state's own bar in plainer language. Walks if it competes with the official report instead of translating it.

Group D — Consume the Data (downstream consumers, ending with the rider)

D1 — Sam, ingestion engineer at a trip-planner app

  • Goal. Decide which of thousands of feeds are safe to put into production, and re-check cheaply.
  • Values today. The flat /catalog.json and /catalog.csv (grade, score, feed URL, days-to-expiry, top fix) that answer in one request, not one per agency; the versioned static /api/v1/ (agencies, per-state aggregates, national stats); the map.geojson; the badge JSON. "A shared read on a feed's quality before I ingest it is exactly the 'shared understanding before production' the validator's authors say they built for."
  • Gets stuck. Roughly a third of real feeds currently fail an automated fetch (WAF / User-Agent 403s on government-hosted feeds), so a feed he could ingest may show as unreachable; he cannot subscribe to "tell me when this feed's grade or expiry changes," only re-poll.
  • Wants next. Resilient fetching so a blocked feed scores instead of reading as unreachable; a change-feed or webhook on grade/expiry change; GeoJSON and a stable schema he can pin.
  • Adopts if the catalog is reliable enough to gate ingestion. Walks if fetch gaps make the grades look unreliable.

D2 — Aisha, MPO / regional planner and modeler

  • Goal. Aim limited technical-assistance dollars at the agencies where poor data overlaps high need.
  • Values today. The /equity/ overlay (state-level ACS poverty, zero-vehicle, disability shares joined to grades); the program rollups with a worst-first attention queue; the dataset.parquet / CSV / JSON for her own analysis; the national map and all-routes views. "It flags high-need states carrying many low-grade feeds. That is a triage list."
  • Gets stuck. Equity is state-level; she works at tract and corridor scale. The rollups are configured by cohort, not drawn to her MPO boundary out of the box.
  • Wants next. The built tract-level equity refinement wired live; a custom cohort drawn to her region; ridership-weighted views so a big-ridership low-grade feed ranks above a tiny one.
  • Adopts if it sharpens where her program spends time. Walks if the geography is too coarse to act on.

D3 — Tomas, academic transit-data researcher

  • Goal. Study national feed-quality change over time and cite it in a paper.
  • Values today. The dated per-agency artifacts as a longitudinal record; the national quality trend (ADR 0020) and /trends/; the open dataset.{json,csv,parquet}; the documented, versioned read API.
  • Gets stuck. There is no stable citable reference (a versioned release, a DOI, a data dictionary, a methodology version stamp) he can point a reviewer at; the rubric weights can change, and without a pinned methodology version his numbers are not reproducible.
  • Wants next. A versioned, citable dataset release with a data dictionary; a methodology changelog with effective dates; a documented schema version on every artifact.
  • Adopts if it is a credible, citable research substrate. Walks if the methodology shifts under him with no version to pin.

D4 — Gloria, rider with no car who uses a screen reader

  • Goal. Trust that her bus appears, on time and correctly, in the app she uses.
  • Values today (indirectly). She never opens the scorecard. She benefits when an agency acts on the freshness warning before the feed expires and the app drops the route; when wheelchair_boarding is populated so her app knows a stop is accessible; when headsigns and readable stop names mean the app shows "Downtown" not "STOP 0041." The tool's whole value to her is upstream.
  • Gets stuck. Nothing the scorecard publishes today is written for her. If she did land on it (from a journalist's story, an advocate's post), the agency page is built for the manager, not the rider, and the accessibility-coverage view is a data surface, not a "is my bus okay?" answer.
  • Wants next. A rider-readable "is my agency's feed healthy?" lookup that leads with the expiry risk in human terms; the accessibility-coverage view made readable for a rider or advocate; Spanish, given who rides.
  • Adopts if there is ever a surface that answers her question in her language. Walks if the tool stays entirely producer-facing (which is fine, but then she is served only through the agency).

Group E — Assure & Audit (independent scrutiny and oversight)

E1 — Frank, FTA / state-DOT NTD oversight staff

  • Goal. See, nationally, which fixed-route reporters actually meet the federal GTFS obligation, and where the known feed-identity gaps are.
  • Values today. The /ntd/ national certification-readiness page reading ntd.json; the per-agency NTD readiness section (published, valid, current, agency_id present) and the optional agency_id-to-NTD-ID equality flag populated by scorecard ntd-crosswalk from the Transitland Atlas; the framing that the equality result carries no score. "This surfaces exactly the feed-identity and freshness gaps the rulemaking documented."
  • Gets stuck. RY2026 requires a stable agency_id and P-50 crosswalk, but not equality to the NTD ID, so the copy must keep those two facts separate; agencies with multiple datasets, multiple brandings, or a shared regional feed are the exact identity cases FTA flagged and the hardest to match cleanly.
  • Wants next. Readiness copy audited against final-rule (not proposed-rule) language; explicit handling of multi-dataset and shared-feed agencies; the national readiness counts exportable as evidence for the next rulemaking.
  • Adopts if the readiness view is accurate to the final rule. Walks if it overstates an obligation agencies do not actually carry.

E2 — Naomi, transit and disability-access advocate

  • Goal. Press agencies to publish the accessibility data riders depend on, without shaming the small ones into defensiveness.
  • Values today. Accessibility's prominent placement in the rubric (the two wheelchair components carry 40 of 100 rider-experience points) and the standalone accessibility sub-score (ADR 0006); the /access/ national accessibility-coverage view; the conformance mark's 90% accessibility floor; the firm "absence is shown neutrally, never a zero" principle. The research she cites is real: mobility-disabled riders reach far fewer accessible stops when the data and infrastructure are missing (J. Transport Geography, 2023).
  • Gets stuck. The coverage view measures what the feed states, not whether a stop is physically usable; she needs that caveat to stay loud so a "90% accessible" number is not misread as 90% of stops being usable. The view is a data surface, not an advocacy-ready story.
  • Wants next. A readable accessibility-coverage map for riders and advocates; the "states it, does not certify usability" caveat kept unmissable; the pathways/levels and station step-free signals surfaced together as one accessibility picture.
  • Adopts if it moves agencies to publish the fields without shaming them. Walks if the number gets read as a usability guarantee it cannot make.

E3 — Wei, accessibility specialist auditing the scorecard's own UI and VPAT

  • Goal. Confirm the WCAG 2.2 AAA / Section 508 claim is real, not asserted.
  • Values today. The published VPAT (508 edition) with per-criterion Supports / Partially Supports rulings and documented map exceptions; the merge-blocking axe / Lighthouse / pa11y gate; the contrast gate across every theme; the honest Partially Supports call on the national all-routes map. "This is a real AAA effort with documented exceptions, not a badge."
  • Gets stuck. The VPAT's functional-performance rows (302.1 Without Vision) say "verification in Phase 2," and the manual assistive-technology results log in accessibility-testing.md is scripted but awaiting a human AT pass. The strongest claim is the one without lived-experience evidence behind it yet.
  • Wants next. A dated NVDA+Firefox and VoiceOver+Safari walkthrough filling the results log; a screen-reader check of the live result-count and loading status messages; the map exceptions re-verified by an AT user.
  • Adopts if the functional-performance log gets filled by a real AT session. Walks if the AAA claim stays asserted where it should be demonstrated.

E4 — Dana, journalist comparing agencies

  • Goal. Publish a checkable claim about how local transit data quality compares.
  • Values today. The /leaderboard/ and per-state aggregates; the open, reproducible artifacts anyone can re-pull; the national "state of transit data" framing in the problems and trends pages; the /how-to-read/ explainer. "The numbers are public and reproducible, so I can cite them."
  • Gets stuck. The product principle is "no leaderboard that shames small agencies," but a journalist's comparison is exactly the use that can turn the leaderboard into a ranking-to-lose; she needs the methodology and the caveats in plain language so a story does not misread a low grade as a bad agency.
  • Wants next. A plain-language methodology and "what a grade does and does not mean" explainer written for press; a "claims you may and may not make" note; context (size band, ridership) shown next to any comparison so the story is fair.
  • Adopts if she can publish a verifiable, fairly-framed claim. Walks if the tool either hides comparison or invites an unfair one.

Group F — Operate (run the program and the tool)

F1 — Ramona, program liaison / customer-success manager

  • Goal. Open one screen before an agency check-in call that says how the data is doing and the three things to raise, across every agency she supports.
  • Values today. The program rollups ("needs attention" when expiring or regressed, worst-first); the per-agency printable call brief (/agency/<id>/brief/); the "what changed since last check" summary; the opt-in digest (notify.py) filtered to just her agencies; the liaison outreach copy for an expired feed. "The rollup is my Monday worklist; the brief is what I bring to the call."
  • Gets stuck. She cannot save private notes per agency, cannot draw a custom cohort to her own portfolio without editing YAML, and the shared-fix detection ("one export setting fixes these five agencies") is described but not yet a button. The public claim/verify endpoint that lets an agency turn on its own alerts is the missing piece of the retention loop.
  • Wants next. The supporter workspace (saved cohorts, per-agency notes, one-click call-prep export, shared-fix detection); month-over-month cohort movement; the self-serve claim/verify endpoint live so agencies subscribe themselves.
  • Adopts if it plans her week and prepares her calls. Walks if prep stays as manual as a spreadsheet.

F2 — Chelsea, owner and maintainer

  • Goal. Keep the tool cheap, accessible, and worth a second visit, without the scope creeping into a validator or a feed editor.
  • Values today. The static-artifact architecture (no API server in the render path) that keeps hosting trivial; the idempotent daily pipeline; the ADR trail (0001–0024) recording every non-obvious call; the CI gates (ruff, mypy, pytest, axe) that make it hard to break; the discipline that most of the roadmap is already built, with the remaining work being operational, not code.
  • Gets stuck. The gap between "built" and "live" is human and operational: the self-serve claim/verify endpoint, the human AT pass, resilient fetching so a third of feeds stop 403-ing, and tuning what counts as worth an alert. There is no view of whether agencies actually return.
  • Wants next. Close the built-but-not-deployed items in priority order; a light signal of repeat visits and which fix pages are the organic entry points; keep the cost guardrail and the no-shaming principle intact as coverage grows.
  • Adopts the discipline of shipping the operational tail over building new surfaces. Walks the project into trouble only by letting scope drift off the inherited-feed manager and the liaison.

Cross-cutting themes (what the cast agrees on)

  1. The tool is feature-rich and deployment-poor. The single most repeated friction across Dolores, Sam, Ramona, and Chelsea is not a missing feature; it is a built feature that is not yet live or not yet discoverable: the self-serve claim/verify endpoint, resilient fetching, the per-vendor fix wording, the human AT pass. The highest-leverage work is finishing the operational tail, not adding surfaces. This sharpens, rather than overturns, feature-roadmap.md and service-plan.md.
  2. Freshness is the load-bearing promise. The expiry-to-dropped-from-Maps chain is the one consequence every agency-side and consumer-side persona names as concrete and feared. Lead-time alerts, predictive freshness, and the claim/verify loop that delivers them are the retention engine, corroborated by Google's own guidance.
  3. Frame as fix, never failure, holds the whole product together, and is fragile at scale. The vendor (B1), the advocate (E2), and the journalist (E4) each describe a surface (vendor view, coverage view, leaderboard) that could tip into shaming. The no-leaderboard principle is a feature, and it needs explicit guardrails on exactly those three surfaces.
  4. "States it" is not "certifies it," and that caveat has to stay loud. The accessibility-coverage number, the conformance mark, and the NTD readiness flag all measure what a feed publishes, not real-world usability or official compliance. Frank, Naomi, Wei, and Hiro independently insist the caveat stay unmissable; it is already in the docs and must stay in the UI.
  5. Each audience wants its own surface over the same artifacts. The engine and the data are rich; the views are thin for some roles: a rider-readable lookup (Gloria), a vendor-facing worklist (Marcus), a press methodology page (Dana), a citable dataset release (Tomas), a supporter workspace (Ramona). Most are renders over data that already exists.
  6. Accuracy of claims about external obligations is a credibility lever. Frank's NTD final-rule nuance and Hiro's per-state-guideline gap are both about the tool stating the external bar correctly. Getting the federal and state framing exactly right is cheap and protects trust.

Honest limits of this exercise

This is simulated. It can surface plausible needs and obvious gaps, but it cannot tell you which are real, how many of each user exist, or what any of them would pay or adopt. It over-represents the maintainer's mental model of these roles and will miss what only a real liaison, a real vendor, or a real rider would surprise you with. Several personas (the vendor, the app developer, the rider) are roles the project has had little direct contact with, so their cards are the least trustworthy and the most in need of real discovery. Do not treat any "want" here as validated demand. The honest next step is real conversations with at least one person in each group, starting with the two the tool already serves (the inherited-feed manager and the liaison) because they are reachable and their feedback is load-bearing.

The triaged backlog, the sequenced plan, the traceability matrix, and the validation plan derived from these interviews are in RESEARCH-ROADMAP.md.