The ideal path: how the same PM does the same job, inside Sift. Every step and every feature closes a specific As-Is barrier or serves a specific job, and the trace is shown. A feature with no parent pain or job is cut.
Focus persona: personas.md (Persona A). Main job: jtbd.md. As-Is source: cjm-as-is.html.
The same PM doing the same job, inside Sift. Each step traces to a concrete As-Is barrier or a job (the Closes line). A step that closes nothing is an orphan and is cut. Target emotions are designed goals contrasted against the As-Is emotion on that phase, not measured. Steps 3, 4, and 6 are the differentiators.
Scroll horizontally to move through the six steps.
PM connects CSV import and/or Intercom; feedback ingests and normalizes; PII scrubbed by default.
When the session looms, PM opens Sift and sees feedback already clustered into themes with volume, recency, and evidence count; no manual reading or tagging.
Each theme shows item count (n=X) and a low-signal badge when thin; strong themes are visibly distinct from thin ones.
PM clicks a theme, sees its supporting items (n=X), clicks any item to the raw original text; verifies a theme they already know is real, then trusts the novel ones; sets priority.
PM shares an evidence brief (top themes, evidence per theme, source quotes, recommended priority) as a link before the session, so the reasoning is legible without Sift access.
Challenged in the room ("my customer said X"), the PM pulls the theme live and shows n=X items and the verbatim quotes on the spot; the anecdote meets traceable evidence.
Orphan check: none. All six steps close a named As-Is barrier and serve a job. All five growth zones are covered: GZ1 to Phase 6, GZ2 to Phases 3 and 4, GZ3 to Phase 2, GZ4 to Phase 1, GZ5 to Phases 2 and 5.
The To-Be projected target overlaid on the As-Is evidence curve. The entirely-negative As-Is trough is designed to invert into a steady positive line, rising to its peak exactly where As-Is hit its floor. Phase 6 is the largest lift (-5 to +4), the deepest, best-evidenced As-Is pain becoming the moment the PM is most in control.
The dashed To-Be line is a designed goal, not measured data. Phase 6 (the amber band) is the biggest inversion: from -5 powerlessness, the As-Is floor and its strongest-evidenced pain, to +4 control. That single inversion is the product's core promise, aimed at a real, confirmed pain.
The existing MVP core, sharpened. This is the same list as the lean UX canvas Solutions, the JTBD matrix MVP core, and the CLAUDE.md MVP scope, reordered by priority with a parent and a success signal on each. One list, not a second parallel backlog. Every item has a parent barrier or job; an item with no parent is not added. The five MVP-spine rows are highlighted.
| # | Feature / capability | Parent (As-Is barrier / job) | Priority | Success signal (hypothesis) |
|---|---|---|---|---|
| 1 | Feedback ingestion - CSV + Intercom | As-Is P2 siloed (GZ4); J2 | MVP | 40% of signups reach a first synthesis on their own data within 7 days (aarrr.md Activation) |
| 2 | AI theme synthesis - transparent clusters with volume, recency, evidence count; basic ranking by signal | As-Is P1 no POV (GZ5) + P3 tagging (GZ3); Main job; J2 | MVP | First synthesis in under 20 minutes with at least one theme the PM recognizes as real (aarrr.md Activation) |
| 3 | Honest confidence display - item count n=X + low-signal badge | As-Is P4 cannot gauge strength (GZ2); J3 | MVPD2, mandatory | Activated users act on the synthesis; leading H1 condition. Objective 1: 60% cite Sift as primary roadmap input at 90 days |
| 4 | Evidence drill-down + inline citation - theme, evidence items (n=X), raw source text | As-Is P4 distrust (GZ2) + P6 unciteable evidence (GZ1); Main job; J3; E1 | MVP | Activated users open the drill-down at least once in the first session (the trust behavior named in live-research.md) |
| 5 | Shareable public evidence brief - no login to view | As-Is P5 does not land (GZ5) + P6 defend (GZ1); S1; J4 | MVP | 25% of signups attributed to referral / brief sharing (aarrr.md Referral) |
| 6 | Prioritization scoring - multi-factor (customer segment, revenue impact) | Main job (rank and interrogate) | Later | PM interrogates a ranked list rather than accepting a black-box order (basic volume/recency ranking already ships inside item 2) |
| 7 | "New signal since last session" digest | Weak parent: retention loop (aarrr.md) + H-J1 continuous sensing, which jtbd.md marks as an UNCONFIRMED hypothesis job. Not an As-Is barrier. | Laterheld with caveat | Week-4 retention 50% for activated users (aarrr.md Retention) |
| 8 | Roadmap link - native lightweight or Jira / Linear sync | J4 (get the team to the same why); the decision-theme-signal chain | Later | Roadmap items linked back to their themes; serves Segment B more than the core Segment A journey |
Adversarial check of the To-Be map and backlog: does every feature have a parent, does the backlog sharpen rather than fork the MVP core, and is "MVP" no broader than the riskiest assumption (H1).
Orphans cut: none new. The map confirms the earlier cuts, which correctly map to no To-Be step: J5 signal-loop / contribution tracker (D1, Segment C), real-time monitoring (anti-pattern per ux-patterns.md), team synthesis reproduction (unconfirmed job). The free/paid split is untouched: the drill-down (item 4) stays the paid differentiator, consistent with gating the trust mechanism as the paid value.