| name | product-lifecycle-learning |
| description | Compare intended product outcomes against observed results to close the launch-to-learning loop: collect post-launch evidence, distinguish expected from observed from uncertain from inferred claims, update assumptions, assess feature health, and choose among continue/improve/harvest/pivot/pause/retire — including retirement lifecycles with deprecation, migration, customer treatment, and retained reusable learning. Do not use for incident postmortems or root-cause analysis (routes to incident-learning or site-reliability-engineering); do not use for analytics instrumentation or metric dashboard design (routes to product-analytics-and-measurement); do not use arbitrary thresholds as universal retirement rules — decisions require human judgment and context. |
| license | MIT |
| metadata | {"tags":"product-lifecycle-learning, post-launch-review, outcome-review, feature-health, assumption-update, retirement-decisions, deprecation, sunset-planning, retained-learning, evidence-ledger, epistemic-discipline, lifecycle-closure"} |
Product Lifecycle Learning
Close the loop from launch to learning. This skill compares what was intended against
what actually happened, maintains an evidence-backed assumption ledger, assesses
feature health, and makes disciplined continue/improve/harvest/pivot/pause/retire
decisions — including full retirement lifecycles. It produces a durable retained
learning record that feeds back into roadmap, analytics, adoption, experimentation,
and future specifications.
Loading Guide
Load only the reference or template relevant to the task. Do not load every file at once.
| File | Load when |
|---|
| references/discovery-brief.md | You need to understand how lifecycle-learning concepts map across skills and where this skill's boundaries are |
| references/epistemic-discipline.md | You need the full taxonomy for classifying claims as expected, observed, uncertain, or inferred |
| references/retirement-lifecycle.md | Planning a feature or product retirement, including deprecation, migration, customer treatment, and internal cleanup |
| references/feedback-destinations.md | Routing learning outputs to the right downstream skill — roadmap, analytics, adoption, experimentation, or specification |
| templates/outcome-review.md | Conducting a structured post-launch outcome review comparing expected vs. observed |
| templates/assumption-ledger-update.md | Updating the assumption ledger with new evidence and confidence shifts |
| templates/feature-health-record.md | Assessing feature health across multiple dimensions and surfacing signals |
| templates/retirement-decision.md | Making and recording a justified retirement or continuation decision |
| templates/sunset-plan.md | Planning deprecation communication, migration paths, customer treatment, and internal cleanup |
| templates/retained-learning-record.md | Capturing durable reusable learning that survives beyond the feature |
Core Methodology
The Launch-to-Learning Loop
LAUNCH → [OBSERVE] → [COMPARE] → [IDENTIFY GAPS] → [UPDATE ASSUMPTIONS] → [ASSESS HEALTH] → [DECIDE] → [CAPTURE LEARNING] → (feed back)
| | | | | | |
Collect Expected vs. Gap analysis Assumption Feature health Continue / Retained
outcome observed with confidence ledger update dimensions Improve / learning
data outcomes intervals Harvest / record
Pivot /
Pause /
Retire
The loop starts after launch (the feature or capability is live and generating data) and ends with a
durable learning artifact that feeds the next cycle of roadmap, analytics, adoption, experimentation,
and specification work.
Stage-by-Stage
| Stage | Input | Activity | Output |
|---|
| Observe | Analytics data, adoption metrics, user feedback, support tickets, operational metrics | Collect outcome evidence from observed behavior and system data. Distinguish signal from noise. Flag missing or low-confidence data. | Collected outcome data with confidence labels |
| Compare | Expected outcomes (from spec/roadmap), observed outcomes, confidence intervals | Compare the two; identify alignment, deviation, and surprise. Do not conflate expectation with observation. | Gap analysis: what matched, what diverged, what was ambiguous |
| Identify gaps | Gap analysis, assumption ledger | Identify which assumptions held and which broke. Distinguish between measurement gaps (could not observe) and outcome gaps (observed deviation). | Assumption gap register with confidence |
| Update assumptions | Assumption gap register, prior assumption ledger | Revise assumptions: strengthen confirmed ones, weaken contradicted ones, add new ones surfaced by the data. Record confidence shifts. | Updated assumption ledger. Use templates/assumption-ledger-update.md. |
| Assess health | Updated assumptions, adoption data, operational metrics, user feedback | Evaluate feature health across adoption, technical, operational, and strategic dimensions. Do not reduce to a single score. | Feature health assessment. Use templates/feature-health-record.md. |
| Decide | Feature health assessment, business context, portfolio priorities | Choose one of six lifecycle decisions. The decision requires human judgment; no automated threshold. | Decision record with accountable owner. Use templates/retirement-decision.md. |
| Capture learning | Decision record, gap analysis, updated assumptions, context | Produce a durable retained learning record: what was learned, why, and how it should inform future work. Not a transient meeting summary. | Retained learning record. Use . |
Epistemic Discipline
Every claim in lifecycle-learning output is classified into exactly one of four categories. These are not
conflated; a comparison is not an observation, and an inference is not a fact.
| Category | Definition | Example | Source |
|---|
| Expected | What was intended or predicted before launch | "We expected activation to reach 60% within 30 days" | Spec, roadmap, launch brief |
| Observed | What actually happened, measured from data | "Activation reached 43% at 30 days (95% CI: 39-47%)" | Analytics, adoption data, operational metrics |
| Uncertain | What is ambiguous, noisy, or contested | "Attribution is confounded by a simultaneous pricing change; cannot isolate feature effect" | Confidence intervals, conflicting signals, data-quality issues |
| Inferred | What is concluded from evidence, with reasoning | "The gap between expected 60% and observed 43% suggests the onboarding redesign did not reduce time-to-value as hypothesized; the pricing change confound means we cannot rule out an external cause" | Reasoned implication from evidence |
Full taxonomy and field guide in references/epistemic-discipline.md.
Lifecycle Decisions
Six outcomes are available after assessment. The choice requires human judgment informed by evidence;
no numeric threshold or automated rule replaces context and accountability.
| Decision | Meaning | Typical evidence profile | Follow-up |
|---|
| Continue | Keep as-is; feature is healthy | Outcomes match or exceed expectations; stable, low-risk | Schedule next review |
| Improve | Invest in enhancement | Adoption gap exists but fixable; underlying need confirmed | Feed roadmap and experimentation |
| Harvest | Reduce investment, maintain for existing users | Declining growth but stable base; not worth expanding | Monitor for retirement signals |
| Pivot | Change direction significantly | Need confirmed but current approach failed | Feed roadmap, discovery, experimentation |
| Pause | Temporarily halt investment | Ambiguous results, external confounds, or resource constraint | Schedule re-assessment with new evidence |
| Retire | Deprecate and remove | Sustained non-adoption, replacement exists, or strategic misalignment | Execute retirement lifecycle |
Retirement Lifecycle
When the decision is Retire, a structured retirement lifecycle covers the full path from deprecation
announcement through internal cleanup. Full detail in references/retirement-lifecycle.md.
| Phase | Activity | Template |
|---|
| Deprecation communication | Announce retirement: timeline, rationale, alternatives. Target affected users with segmentation. | templates/sunset-plan.md |
| Migration path | Provide migration tooling, documentation, and support for existing users. Define the recommended path. | templates/sunset-plan.md |
| Customer treatment | Support commitments during sunset: data export, grace periods, extended support windows, SLA preservation, refund/credit policies where applicable. Coordinate with customer-success. | templates/sunset-plan.md; route communication plans to conditional-customer-success |
| Internal cleanup | Remove feature flags, archive code, update documentation, retire monitoring and alerting, reclaim infrastructure. | templates/sunset-plan.md |
| Learning closure | Capture what the feature's lifecycle taught — not a postmortem, but a closure record that completes the learning loop. | templates/retained-learning-record.md |
Retained Learning Record
Every lifecycle-learning cycle produces a durable retained learning record — not a transient meeting
summary. The record captures:
- What the feature or capability was intended to achieve (expected outcomes)
- What actually happened (observed outcomes, with confidence)
- What was uncertain and why
- What assumptions were updated and how
- What decision was made (continue/improve/harvest/pivot/pause/retire) and who made it
- Why that decision was reached, with evidence
- What should inform future decisions — reusable patterns, anti-patterns, assumptions to test next time
- Where the learning was routed (roadmap, analytics, adoption, experimentation, specifications)
This record is the durable learning artifact. It is the evidence that the launch-to-learning loop
actually closed.
When Not to Use
This skill does not own:
- Incident postmortems, root-cause analysis, or operational incident review — these belong to
incident-learning (not yet landed) and ../site-reliability-engineering/SKILL.md. Lifecycle-learning consumes incident signals as input but does not produce postmortems.
- Analytics instrumentation, metric dashboard design, tracking-plan creation, or event taxonomy — these belong to ../product-analytics-and-measurement/SKILL.md. Lifecycle-learning consumes analytics data as input but does not own measurement infrastructure.
- Customer-success account management, renewal decisions, or health scoring — these belong to
conditional-customer-success (not yet landed). Lifecycle-learning routes retirement communication plans and customer-treatment strategies there.
- Roadmap prioritization or portfolio allocation — these belong to ../product-roadmapping-and-portfolio/SKILL.md. Lifecycle-learning feeds evidence into roadmap decisions but does not make them.
- Arbitrary or automated retirement thresholds — this skill never applies rules like "retire if DAU < 100" or "kill if NPS < 30" without context about the product, market, user base, and alternatives. Retirement decisions require human judgment and named accountability.
Routing and Feedback
Inputs (consumed by lifecycle-learning)
| Input | Source |
|---|
| Expected outcomes, acceptance criteria | ../spec-driven-development/SKILL.md, roadmap briefs |
| Observed outcomes, metric data, funnels, cohorts | ../product-analytics-and-measurement/SKILL.md |
| Adoption evidence, activation rates, retention signals | ../product-adoption/SKILL.md |
| Experiment results, readout learning entries | ../product-experimentation/SKILL.md |
| Incident signals, reliability data | ../site-reliability-engineering/SKILL.md, incident-learning |
| Customer feedback, support trends, health signals | conditional-customer-success |
Outputs (produced by lifecycle-learning, routed to)
At least five feedback destinations must be updated per cycle: roadmap, analytics, adoption,
experimentation, and specifications. Additional routing to customer-success and incident-learning
is conditional on the decision.
File Map
Related Skills