# Feedback Loop Field Guide

> 204 feedback loops used in product development, ordered by latency: the wait between doing something and finding out whether it worked. Spans five tracks (Build, Value, Reach, Team, Risk & Moat) with fast-proxy chains and balancing metrics. Maintained by Konrad Bloor.

- Author: Konrad Bloor (https://konradbloor.com · https://www.linkedin.com/in/konradb/)
- Version: 0.23.1
- License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
- Cite as: Konrad Bloor, "Feedback Loop Field Guide" v0.23.1, https://konradbloor.com/loops
- Canonical (interactive chart): https://konradbloor.com/loops
- The same loops organised by decision: https://konradbloor.com/decisions
- Machine-readable: https://konradbloor.com/loops.json

## Build · Can we make the thing?

### IDE inline feedback

- Decision it informs: Is this code correct?
- Latency: typically 300ms (range 50ms to 2s)
- Signal: high fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: errors as you type
- Tools: VS Code, IntelliJ, Copilot
- Notes: The innermost loop, feedback before you finish typing.

### Syntax / parser

- Decision it informs: Is this code correct?
- Latency: typically 500ms (range 50ms to 3s)
- Signal: high fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: pass / fail
- Tools: compiler, interpreter
- Notes: Binary, instant, free.

### Type system

- Decision it informs: Is this code correct?
- Latency: typically 2s (range 200ms to 30s)
- Signal: high fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: type errors
- Tools: TypeScript, mypy, Rust
- Notes: Shape errors inside your own walls.

### Hot reload / REPL

- Decision it informs: Does this behave as I expect?
- Latency: typically 3s (range 500ms to 15s)
- Signal: medium fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: looks right / doesn't
- Tools: Vite HMR, Fast Refresh, REPL, Storybook
- Notes: Visual/behavioral confirmation without a full rebuild.

### Runtime schema validation

- Decision it informs: Does outside data violate our assumptions?
- Latency: typically 1min (range 1s to 1hr)
- Signal: high fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: validation failures
- Tools: Zod, Pydantic, io-ts, protobuf
- Notes: The guard at the gate, catches what types can't: the outside world.

### Lint / static analysis

- Decision it informs: Does this follow our conventions?
- Latency: typically 30s (range 2s to 5min)
- Signal: medium fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: violations
- Tools: ESLint, Ruff, clippy
- Notes: Style and convention enforcement at build time.

### Coverage / mutation testing

- Decision it informs: Would our tests even notice this bug?
- Latency: typically 10min (range 2min to 1hr)
- Signal: medium fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: % covered, % mutants killed
- Tools: coverage gates, Stryker, mutmut
- Balanced by: Build & CI duration
- Notes: A loop about your loops: coverage says what the tests touch, mutation testing says whether they'd actually fail. Both are gameable; mutation less so.

### Accessibility checks

- Decision it informs: Can everyone actually use this?
- Latency: typically 10min (range 1min to 1d)
- Signal: medium fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: violations
- Tools: axe-core, eslint-plugin-jsx-a11y, manual audits
- Notes: Automated checks catch roughly a third of WCAG issues in seconds; the rest need a human with a screen reader. Cheap where it's automated, neglected where it isn't.

### Unit tests

- Decision it informs: Is the logic correct?
- Latency: typically 2min (range 5s to 15min)
- Signal: high fidelity · medium cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: pass / fail
- Tools: Jest, pytest, Vitest
- Fast proxy for: Error monitoring
- Notes: Fitness function for individual units of logic.

### Integration tests

- Decision it informs: Do the contracts between parts hold?
- Latency: typically 10min (range 1min to 1hr)
- Signal: high fidelity · medium cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: pass / fail
- Tools: Testcontainers, Playwright API
- Notes: Catches drift between components.

### Code review

- Decision it informs: Would another human build it this way?
- Latency: typically 12hr (range 30min to 3d)
- Signal: medium fidelity · medium cost · human-mediated · leading · push delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: comments, approval
- Tools: GitHub PRs, Graphite
- AI can shorten this loop
- Notes: Human loop between lint and CI, catches design, not just defects.

### Pair / mob programming

- Decision it informs: Would another human build it this way, while I'm still typing?
- Latency: typically 1min (range 5s to 1hr)
- Signal: medium fidelity · high cost · human-mediated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: in-flight corrections
- Tools: pair programming, mob programming
- Framework: XP
- Fast proxy for: Code review
- Notes: Code review at conversation speed. The objection arrives while your hands are still on the keyboard. The pair shares one context and one set of blind spots, so it complements async review rather than replacing it. Nearly impossible to measure, which is exactly why orgs undervalue it.

### AI code review

- Decision it informs: Does a tireless machine reviewer spot what I missed?
- Latency: typically 5min (range 1min to 30min)
- Signal: medium fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: findings
- Tools: Claude Code review, Copilot code review, Greptile
- Fast proxy for: Code review
- Notes: A new rung between lint and human review, catches real bugs in minutes and never gets tired. What it misses tells you what humans are still for.

### E2E tests

- Decision it informs: Do whole flows work?
- Latency: typically 30min (range 5min to 2hr)
- Signal: medium fidelity · high cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: pass / fail, flake %
- Tools: Playwright, Cypress
- Fast proxy for: Error monitoring
- AI can shorten this loop
- Notes: High coverage, flaky signal. Trades fidelity for breadth.

### Visual regression testing

- Decision it informs: Did this change break how it looks?
- Latency: typically 10min (range 1min to 1hr)
- Signal: medium fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: visual diffs flagged
- Tools: Percy, Chromatic, Playwright screenshots
- Fast proxy for: Bug reports
- AI can shorten this loop
- Notes: Screenshot diffs catch the CSS regression the unit tests can't see. The fidelity lever is diff-threshold tuning. Too tight and every antialiasing shift pages a human.

### Error monitoring

- Decision it informs: Is it correct in reality?
- Latency: typically 1hr (range 1min to 1d)
- Signal: high fidelity · low cost · automated · lagging · push delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: errors/hr, users affected
- Tools: Sentry, Rollbar, PostHog error tracking, Crashlytics
- Fast proxy for: Bug reports
- Actions: rollback or hotfix; add a regression test; mute and ticket if benign
- Notes: Reality's opinion of your test suite. On mobile, crash-free sessions % is this loop's headline number.

### Bug reports

- Decision it informs: What did every earlier loop miss?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · medium cost · human-mediated · lagging · push delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: reports/wk
- Tools: Jira, Linear, support inbox
- Balanced by: Throughput
- AI can shorten this loop
- Actions: fix, then ask which earlier loop should have caught it
- Notes: The most expensive way to find a defect.

### Alerts / on-call pages

- Decision it informs: Is production healthy right now?
- Latency: typically 15min (range 1min to 1hr)
- Signal: high fidelity · high cost · automated · lagging · push delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: pages
- Tools: PagerDuty, Opsgenie
- Actions: follow the runbook; rollback the recent deploy; declare an incident
- Notes: Production telling you something, loudly.

### APM / distributed tracing

- Decision it informs: Where is it slow and why?
- Latency: typically 1hr (range 1min to 1d)
- Signal: high fidelity · medium cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: ms p50–p99, error %
- Tools: OpenTelemetry, Datadog, Honeycomb, Grafana
- Notes: Spans and telemetry: latency, error rates, p99s. OpenTelemetry is the usual instrumentation.

### SLO / error budget burn

- Decision it informs: Are we reliable enough to keep shipping?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: high fidelity · medium cost · automated · lagging · push delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: % error budget burned
- Tools: SLO dashboards
- Actions: freeze features, spend the budget on reliability; renegotiate the SLO
- Notes: Feedback formalized as a contract.

### Incident post-mortems

- Decision it informs: What structural weakness did this expose?
- Latency: typically 1wk (range 3d to 1mo)
- Signal: high fidelity · high cost · human-mediated · lagging · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: action items
- Tools: Blameless reviews
- AI can shorten this loop
- Actions: fix the systemic cause; add the guardrail
- Notes: Delayed but very high fidelity.

### Canary / progressive rollout

- Decision it informs: Does this deploy hurt real traffic at 1% before it can hurt 100%?
- Latency: typically 30min (range 5min to 4hr)
- Signal: high fidelity · medium cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: error delta vs baseline
- Tools: feature flags (LaunchDarkly, Statsig), canary analysis, blue-green deploys
- Fast proxy for: Change failure rate
- Actions: halt and rollback; proceed to 100%
- Notes: The missing rung between CI and the pager, production feedback with the blast radius dialed down. Kill switches make rollback a config change, not a deploy.

### Synthetic monitoring / uptime checks

- Decision it informs: Is the front door open right now?
- Latency: typically 5min (range 1min to 1hr)
- Signal: medium fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: uptime %, failed checks
- Tools: Checkly, Pingdom, health probes
- Notes: Scripted users hitting critical paths every minute, catches the outage before a human does, but only on the paths you thought to script.

### Chaos engineering / game days

- Decision it informs: Does our resilience survive contact with induced failure?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: high fidelity · high cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: weaknesses found
- Tools: fault injection, game days, DR drills, backup restore drills
- Fast proxy for: Incident post-mortems
- Notes: Synthetic incidents on your schedule instead of real ones on the pager's, the complementary leading loop to the post-mortem. The backup-restore drill is the variant nobody runs until the disaster: an untested backup is a hope, not a plan.

### Benchmark tests

- Decision it informs: Did this change make it slower?
- Latency: typically 30min (range 5min to 4hr)
- Signal: medium fidelity · medium cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: growth, scale
- Measured in: ms, ops/s
- Tools: k6, criterion, CI perf gates
- Fast proxy for: APM / distributed tracing
- Notes: Performance regression caught before deploy.

### Load / stress testing

- Decision it informs: At what load does it actually break?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · medium cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: breaking point (RPS), p99 under load
- Tools: k6, Gatling, Locust
- Fast proxy for: Capacity / scaling signals
- Notes: The performance sibling loop of chaos engineering: synthetic 10x on your schedule instead of real 10x on launch day. Staging never quite matches production. Treat the number as a bound, not a forecast.

### Capacity / scaling signals

- Decision it informs: Will this architecture survive growth?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: medium fidelity · medium cost · automated · leading · push delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: % utilization
- Tools: autoscaler metrics, load forecasts
- Notes: Demand signal disguised as infrastructure pain.

### Bundle size / asset weight

- Decision it informs: Did this change bloat the download?
- Latency: typically 3min (range 30s to 30min)
- Signal: high fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: kB
- Tools: size-limit, bundlewatch, webpack-bundle-analyzer
- Fast proxy for: Lighthouse / synthetic performance
- Notes: Fastest front-end loop: caught on the PR, before a page ever renders.

### Lighthouse / synthetic performance

- Decision it informs: Is the page fast in the lab?
- Latency: typically 10min (range 1min to 1hr)
- Signal: high fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: score 0–100
- Tools: Lighthouse CI, WebPageTest
- Fast proxy for: Core Web Vitals (field / RUM)
- Notes: Lab data, deterministic and leading, but a synthetic proxy for what real users feel.

### Memory / resource utilization

- Decision it informs: Are we leaking memory or heading for an OOM?
- Latency: typically 1hr (range 1min to 3d)
- Signal: medium fidelity · medium cost · automated · leading · push delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: MB, OOM kills
- Tools: heap profiler, cgroup / RSS metrics, OOM-kill alerts
- Notes: Slow creep in staging, sudden OOM in prod. Same loop, two speeds.

### Core Web Vitals (field / RUM)

- Decision it informs: Is the page fast for real users?
- Latency: typically 1mo (range 1d to 3mo)
- Signal: high fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: ms at p75, CLS
- Tools: CrUX, web-vitals RUM, Search Console
- Notes: LCP, INP, CLS at p75 over a 28-day rolling window. The field truth Lighthouse only estimates, and a ranking signal.

### Secrets detection

- Decision it informs: Did we just leak a credential?
- Latency: typically 1min (range 2s to 10min)
- Signal: high fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: leaks found
- Tools: gitleaks, GitHub secret scanning
- Notes: One of the few security loops that's instant.

### Threat modeling

- Decision it informs: Where would an attacker go, before we build it?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: medium fidelity · medium cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: threats identified
- Tools: STRIDE, attack trees, design review
- Fast proxy for: Pen test / red team
- AI can shorten this loop
- Actions: redesign the risky surface; backlog the mitigations
- Notes: The flaw caught at the whiteboard costs a conversation; the same flaw caught by the pen test costs a rebuild.

### SAST / code security scan

- Decision it informs: Did we just write a vulnerability?
- Latency: typically 30min (range 2min to 1d)
- Signal: medium fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: vulns by severity
- Tools: Semgrep, CodeQL
- Fast proxy for: Pen test / red team
- AI can shorten this loop
- Notes: Scans the code you wrote, at commit time. Noisy, fidelity depends on triage discipline.

### SCA / dependency vulnerabilities

- Decision it informs: Did we inherit a vulnerability, and can it reach us?
- Latency: typically 1d (range 10min to 1wk)
- Signal: medium fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: open CVEs by severity
- Tools: Dependabot, Snyk, OSV / CVE feeds
- Fast proxy for: Pen test / red team
- AI can shorten this loop
- Actions: patch or pin; triage reachability; accept the risk with an expiry date
- Notes: Fires on someone else's event, a CVE lands against code you haven't touched in a year. Noisy: most hits sit in code paths you never call, so reachability triage is the fidelity lever.

### Cloud misconfiguration / CSPM

- Decision it informs: Is our infrastructure drifting into exposure?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: high fidelity · medium cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: open misconfigurations
- Tools: AWS Config / Security Hub, Wiz / Prisma, tfsec / IaC scans
- Fast proxy for: Pen test / red team
- Notes: The public-S3-bucket loop. Code scanners read the repo; this reads the live cloud, where drift happens without a commit.

### Runtime security detection

- Decision it informs: Is someone attacking us right now?
- Latency: typically 1hr (range 1min to 1d)
- Signal: medium fidelity · medium cost · automated · lagging · push delivery · continuous cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: alerts
- Tools: WAF, SIEM / audit logs, anomaly alerts
- Notes: The lagging half of security that scans can't cover, detection instead of prevention. Noisy enough that triage discipline is the real fidelity.

### DAST / automated web scanning

- Decision it informs: Does the running app expose known attack patterns?
- Latency: typically 1wk (range 1hr to 1mo)
- Signal: medium fidelity · low cost · automated · leading · push delivery · periodic cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: findings per scan
- Tools: OWASP ZAP, Burp scans in CI
- Fast proxy for: Pen test / red team
- Notes: The automated sibling loop of the pen test, probes the deployed app on a schedule instead of an engagement. Finds the known patterns; the human finds the novel ones.

### Open vulnerability age / patch SLA

- Decision it informs: Are we closing holes faster than we open them?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: high fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: median days open by severity
- Tools: vuln management dashboards, SLA by severity
- AI can shorten this loop
- Actions: set SLAs by severity; burn down the backlog
- Notes: Every scanner feeds a backlog, and this is the loop that watches whether the backlog is shrinking. It's also the number auditors actually check.

### Bug bounty / VDP

- Decision it informs: What can an unbounded adversary pool find that we can't?
- Latency: typically 2wk (range 1d to 3mo)
- Signal: high fidelity · medium cost · human-mediated · leading · push delivery · continuous cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: valid reports
- Tools: HackerOne, Bugcrowd, security.txt / VDP
- AI can shorten this loop
- Notes: Continuous where the pen test is commissioned: reports arrive unbidden from an unbounded researcher pool, and you pay per find instead of per engagement. Triage load is the hidden cost.

### Pen test / red team

- Decision it informs: Can an adversary actually get in?
- Latency: typically 3mo (range 1wk to 1yr)
- Signal: high fidelity · high cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: leadership · Stages: growth, scale
- Measured in: findings by severity
- Tools: external pen test, red team exercise
- Actions: remediate criticals first; re-test to confirm
- Notes: High fidelity, very slow cycle. The tabletop drill is the same loop run against your response process instead of your perimeter.

### Build & CI duration

- Decision it informs: Is our own loop degrading?
- Latency: typically 1d (range 5min to 1mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: minutes
- Tools: CI dashboards
- Balanced by: Coverage / mutation testing
- Notes: A feedback loop about your feedback loops.

### Code complexity & coupling metrics

- Decision it informs: Is this change making the code structurally worse?
- Latency: typically 10min (range 2min to 1d)
- Signal: medium fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: complexity per function, duplication %, module coupling
- Tools: SonarQube, radon, ESLint complexity rules
- Fast proxy for: Hotspot analysis (churn × complexity)
- Actions: refactor only if the file is hot; tune thresholds to advisory
- Notes: Cyclomatic/cognitive complexity, duplication, module coupling, on the PR like lint, and easy to game: thresholds invite threshold-dodging. The OO metrics (LCOM, CBO, Law of Demeter) don't transfer to FP, where the analogues are module coupling and purity boundaries.

### Unused code / dead exports

- Decision it informs: Are we accumulating unused code?
- Latency: typically 5min (range 30s to 1d)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · event-kind
- Owner: ic · Stages: growth, scale
- Measured in: unused files / exports / deps, dead-code %
- Tools: knip, ts-prune, depcheck
- Fast proxy for: Tech debt accumulation
- Actions: rule out dynamic and entry-point use before deleting; ratchet net-new, don't gate the backlog
- Notes: Whole-project analysis for files, exports, and dependencies nothing imports. TypeScript and lint catch unused locals inside a file, but only project-wide analysis catches an export no other file uses. Medium fidelity: dynamic imports, framework entry points, and genuine public API read as false positives, so treat hits as candidates to review, not delete-on-sight. Best run as a per-PR ratchet on net-new code rather than a gate on the whole backlog.

### Hotspot analysis (churn × complexity)

- Decision it informs: Where is complexity actually costing us?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: high fidelity · low cost · automated · leading · pull delivery · periodic cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: churn × complexity rank, files flagged
- Tools: CodeScene, code-maat, git log frequency analysis
- Fast proxy for: Tech debt accumulation
- Actions: schedule a refactoring sprint; add tests before touching
- Notes: Churn × complexity is what makes the static numbers mean something: complexity in a file nobody touches is trivia; in a file with 40 commits this quarter it's a fire. Lab snapshot vs field truth, the same split as Lighthouse and Core Web Vitals, for code structure.

### Rework rate / code churn

- Decision it informs: How much recent code are we already rewriting?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: % of code churned within N weeks
- Tools: CodeScene, GitClear, git-of-theseus
- Fast proxy for: Hotspot analysis (churn × complexity)
- Notes: The share of new code rewritten again within weeks. High rework signals unclear requirements or a thrashing design, not productivity. A repo-wide leading proxy for where hotspots will form.

### Dependency freshness (libyear)

- Decision it informs: How far behind the ecosystem are we drifting?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: high fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: libyears behind
- Tools: libyear, Renovate, Dependabot version updates
- Fast proxy for: Open vulnerability age / patch SLA
- Actions: turn on auto-update PRs; schedule an upgrade spike
- Notes: Dependency drift is the precondition for CVEs hurting you. Current means the patch is a version bump; three majors behind means the patch is a migration project. Supply-chain tech debt, and it sets your future time-to-patch.

### PR cycle time

- Decision it informs: Is friction accumulating?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: hours open
- Tools: LinearB, Swarmia
- Fast proxy for: Lead time for changes
- Notes: Leading indicator of tech debt and process drag.

### Deployment frequency

- Decision it informs: How often do we ship to production?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: deploys/wk
- Tools: CI/CD analytics, Sleuth, DX
- Framework: DORA
- Balanced by: Change failure rate, Mean time to restore
- Notes: How many OODA cycles you complete per unit time, the cadence of every other Build loop.

### App store review gate

- Decision it informs: Will the gatekeeper let this release through?
- Latency: typically 2d (range 1hr to 2wk)
- Signal: medium fidelity · low cost · human-mediated · lagging · push delivery · on-demand cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: review turnaround, rejection rate
- Tools: App Store review, Play Console staged rollout
- Actions: keep an expedite playbook; server-drive features to skip the gate; stage rollouts behind flags
- Notes: The only loop in Build whose latency belongs to someone else. It caps your deploy frequency, breaks your hotfix MTTR, and its rejections arrive with the clarity of a fortune cookie.

### Lead time for changes

- Decision it informs: How long from commit to production?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: hours commit → prod
- Tools: CI/CD analytics, LinearB
- Framework: DORA
- Balanced by: Change failure rate
- Notes: The physical length of your ship loop. Everything downstream inherits this latency.

### Change failure rate

- Decision it informs: What fraction of deploys break something?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: medium fidelity · medium cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: % of deploys
- Tools: incident tagging, deploy annotations
- Framework: DORA
- Balanced by: Deployment frequency, Lead time for changes
- Actions: shrink batch size; invest in pre-deploy gates
- Notes: Speed without this is just shipping bugs faster.

### Mean time to restore

- Decision it informs: How fast do we recover when it breaks?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: medium fidelity · medium cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: hours to restore
- Tools: PagerDuty analytics, incident tooling
- Framework: DORA
- Balanced by: Deployment frequency
- Actions: invest in rollback tooling; run recovery drills
- Notes: Measured monthly as a trend, even though each incident resolves in hours.

### Task cycle time

- Decision it informs: How long does work-in-progress stay in progress?
- Latency: typically 2wk (range 1wk to 1mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: days
- Tools: Jira/Linear analytics, cumulative flow
- Notes: Idea-to-done for a unit of work. Pre-PMF it's your iteration speed. Slow cycle time throttles learning. Later it also catches queueing and handoff waste that PR cycle time misses.

### WIP

- Decision it informs: How much work is in flight right now?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: high fidelity · low cost · automated · leading · pull delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: items in flight
- Tools: board WIP limits, cumulative flow diagram
- Framework: Kanban
- Fast proxy for: Task cycle time
- Balanced by: Throughput
- Actions: stop starting, start finishing
- Notes: The control knob for the whole flow system. Little's Law says cycle time = WIP / throughput, so too much in flight slows everything. Doubles as an early warning of overcommitment and burnout.

### Work item age

- Decision it informs: Which in-progress items are going stale unnoticed?
- Latency: typically 3d (range 1d to 1wk)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: days in progress
- Tools: aging WIP charts
- Framework: Kanban
- Fast proxy for: Task cycle time
- Actions: swarm the stale item; split it or kill it
- Notes: The complementary leading loop to cycle time. Cycle time tells you after an item finishes, age warns you while it can still be saved.

### Daily standup

- Decision it informs: Is anyone blocked, and is today's plan still right?
- Latency: typically 1d (range 4hr to 3d)
- Signal: medium fidelity · low cost · human-mediated · leading · push delivery · periodic cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: blockers surfaced
- Tools: daily standup, async check-in bots
- Framework: Scrum
- Fast proxy for: Work item age
- Actions: unblock now; replan the day
- Notes: The human sensor for aging work. A good standup notices a stuck ticket days before the metric does. Failure mode: status theatre, reporting to the manager instead of unblocking each other.

### Throughput

- Decision it informs: How many items actually finish per week?
- Latency: typically 2wk (range 1wk to 1mo)
- Signal: medium fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: items/wk
- Tools: items completed per period, Monte Carlo forecasting
- Framework: Kanban
- Balanced by: WIP, Bug reports, Sprint predictability (say:do ratio)
- Notes: The blunt counterpart to velocity, counts finished things, not estimated points.

### Sprint predictability (say:do ratio)

- Decision it informs: Can we trust our own plans?
- Latency: typically 1mo (range 2wk to 3mo)
- Signal: low fidelity · low cost · automated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: % of committed work delivered
- Tools: velocity vs commitment, Jira sprint reports
- Framework: Scrum
- Fast proxy for: Lead time for changes
- Balanced by: Throughput
- Actions: shrink the batch; stop estimating, start counting
- Notes: Sandbagging and story-point inflation both flatter this number while draining ambition, the Goodhart poster child of agile metrics. Throughput balances it by counting what actually finished.

### Flow efficiency

- Decision it informs: How much of cycle time is work vs waiting?
- Latency: typically 1mo (range 2wk to 3mo)
- Signal: medium fidelity · medium cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: % touch time
- Tools: touch time / total time, blocked-time tracking
- Framework: Kanban
- Notes: Typically 15% or less. Most cycle time is queueing, which is why adding people rarely helps.

### Tech debt accumulation

- Decision it informs: Is the codebase still an asset?
- Latency: typically 1yr (range 3mo to 3yr)
- Signal: low fidelity · high cost · human-mediated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: survey score, velocity trend
- Tools: dev surveys, velocity trends
- AI can shorten this loop
- Actions: dedicate a capacity %; decide: rewrite or strangle
- Notes: Noisy, contested, and usually noticed too late.

### Feasibility spike / throwaway prototype

- Decision it informs: Can we even build this, for real?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: high fidelity · medium cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: go / no-go
- Tools: spike branch, proof-of-concept
- Framework: Cagan product risks
- Notes: This is Cagan's feasibility risk, and the prototype is built to be thrown away.

### Technology / model evaluation

- Decision it informs: Is this tech good enough to bet on?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · medium cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: eval score vs bar
- Tools: eval harness, vendor POC
- Framework: Cagan product risks
- Notes: Feasibility for the AI era, does the model clear the bar before you commit?

### Design / RFC review

- Decision it informs: Will this design survive contact with other engineers?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · medium cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: blocking comments resolved
- Tools: RFCs / design docs, ADRs, architecture review
- Fast proxy for: Capacity / scaling signals
- AI can shorten this loop
- Notes: The cheapest place to kill a bad architecture. The same flaw found by the capacity loop two years later costs a rewrite. Failure mode: rubber-stamp reviews that optimise for approval speed.

### Data quality tests

- Decision it informs: Is the data still shaped the way downstream consumers assume?
- Latency: typically 15min (range 1min to 6hr)
- Signal: high fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: growth, scale
- Measured in: assertion failures
- Tools: dbt tests, Great Expectations, Elementary
- Notes: Unit tests for the warehouse, assertions that fail the pipeline before a bad number reaches a dashboard or a model.

### Pipeline freshness / data SLAs

- Decision it informs: Did the data arrive on time, or are the dashboards stale and nobody's caught it?
- Latency: typically 1hr (range 5min to 1d)
- Signal: high fidelity · low cost · automated · lagging · push delivery · continuous cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: hours stale
- Tools: Airflow / Dagster alerts, dbt source freshness, Monte Carlo
- Notes: Stale data fails silently. Decisions keep flowing off last week's numbers unless something watches arrival times.

### Data anomaly detection

- Decision it informs: Did the data break in a way nobody thought to assert?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: medium fidelity · medium cost · automated · lagging · push delivery · continuous cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: anomalies flagged
- Tools: Monte Carlo, Elementary, row-count / distribution anomalies
- Notes: dbt tests catch what you asserted; this catches what you didn't, volume drops, distribution drift, silent nulls. The error monitoring of the data stack.

### Eval suite / golden-set regression

- Decision it informs: Did this prompt or model change make the product dumber?
- Latency: typically 30min (range 2min to 1d)
- Signal: medium fidelity · medium cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: % of golden set passed
- Tools: promptfoo, Braintrust, LangSmith
- Fast proxy for: LLM-as-judge scoring
- Balanced by: Token / cost per request, Model latency / time-to-first-token
- Actions: block the change; expand the golden set
- Notes: The unit tests of AI behavior, run on every prompt tweak and model swap. Fidelity is capped by how well the golden set represents the wild.

### LLM-as-judge scoring

- Decision it informs: Is production output quality drifting, at a scale no human can read?
- Latency: typically 1hr (range 1min to 1d)
- Signal: medium fidelity · medium cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: judge score 1–5
- Tools: LLM-as-judge pipelines, Langfuse, Braintrust
- Fast proxy for: Thumbs up/down on AI output
- Notes: A model grading a model, cheap enough to score everything, biased in ways you must periodically audit against human labels.

### AI red-teaming / adversarial evals

- Decision it informs: Can the model be made to do what it shouldn't?
- Latency: typically 2wk (range 1d to 3mo)
- Signal: high fidelity · medium cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: successful attacks per suite
- Tools: prompt injection suites, jailbreak evals, manual red-teaming
- AI can shorten this loop
- Notes: The pen test of the AI stack: evals measure whether the model is good, this measures whether it holds up. Injection via retrieved content is the finding that surprises everyone.

### Token / cost per request

- Decision it informs: Did this change blow up the cost or context per task?
- Latency: typically 1hr (range 1min to 1d)
- Signal: high fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: $ per task, tokens
- Tools: Helicone / Langfuse, provider usage dashboards, cost per trace
- Fast proxy for: AI / inference spend
- Balanced by: Eval suite / golden-set regression
- Actions: cache and truncate context; route to a cheaper model
- Notes: The new unit economics. One prompt tweak can 10x COGS overnight. Watch cost per task, not the monthly bill.

### Model latency / time-to-first-token

- Decision it informs: How long until the user sees the first token?
- Latency: typically 1hr (range 1min to 1d)
- Signal: high fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: ms to first token
- Tools: TTFT dashboards, APM spans, provider status
- Balanced by: Eval suite / golden-set regression
- Notes: Streaming hides total latency but not time-to-first-token. Part of it sits with the provider, not you.

### Model & output drift

- Decision it informs: Has the upstream model or the user mix shifted under us?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · medium cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: score drift
- Tools: canary prompts, output distribution monitors, Langfuse dashboards
- Actions: pin the model version; refresh prompts and evals
- Notes: Providers swap weights behind stable API names; users drift too. Scheduled canary prompts catch what evals only catch on deploy.

## Value · Is it worth making?

### Dogfooding

- Decision it informs: Can we stand to use it ourselves?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: medium fidelity · low cost · human-mediated · leading · push delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: internal complaints
- Tools: internal releases
- Fast proxy for: Usability tests
- Notes: Fast but biased. You are not your user.

### Beta / early-access program

- Decision it informs: How does it fare with friendly outsiders before the real crowd?
- Latency: typically 1wk (range 3d to 1mo)
- Signal: high fidelity · medium cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: issues, beta adoption %
- Tools: early-access cohorts, TestFlight, design partners
- Fast proxy for: Feature adoption / activation
- Notes: The rung between dogfooding and GA, users who forgive rough edges but aren't blinded by having built it.

### Sprint review / stakeholder demo

- Decision it informs: Is the increment what stakeholders actually wanted?
- Latency: typically 2wk (range 1wk to 1mo)
- Signal: medium fidelity · medium cost · human-mediated · leading · push delivery · periodic cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: stakeholder acceptance
- Tools: sprint reviews, stakeholder demos
- Framework: Scrum
- Fast proxy for: Feature adoption / activation
- Notes: The retro inspects the process; the review inspects the product. Stakeholders nod politely in demos. Acceptance here is not adoption.

### Usability tests

- Decision it informs: Can people actually use it?
- Latency: typically 1wk (range 1d to 2wk)
- Signal: high fidelity · medium cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: issues per 5 users
- Tools: moderated sessions, Maze
- AI can shorten this loop
- Notes: Five users find most of the problems.

### Task success rate

- Decision it informs: Can they finish the task at all?
- Latency: typically 1wk (range 1d to 2wk)
- Signal: high fidelity · medium cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: % completed
- Tools: moderated sessions, Maze, Lyssna
- Balanced by: Time on task
- Notes: Effectiveness, the headline usability number. Binary per user, brutal in aggregate.

### First-click / findability

- Decision it informs: Do they even know where to start?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: high fidelity · low cost · automated · leading · pull delivery · on-demand cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: % correct first click
- Tools: tree testing (Treejack), first-click tests, Maze / Lyssna
- Fast proxy for: Task success rate
- Notes: A correct first click predicts ~87% task success vs. ~46% for a wrong one, the 'identify the task' signal.

### Clicks / interaction cost

- Decision it informs: How far do they wander off the ideal path?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: clicks vs ideal path
- Tools: path metrics (Maze / Lyssna), session replay, analytics funnels
- Notes: Lostness: clicks taken vs. clicks needed. The raw 'number of clicks' made comparable across users.

### Time on task

- Decision it informs: How long does the task take?
- Latency: typically 1wk (range 1d to 2wk)
- Signal: medium fidelity · medium cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: minutes per task
- Tools: moderated sessions, session timers, Maze / Lyssna
- Balanced by: Task success rate
- Notes: Efficiency, noisy with small n; pair it with success rate or it misleads.

### Task ease (SEQ / SUS)

- Decision it informs: How hard did that feel?
- Latency: typically 1d (range 1min to 1wk)
- Signal: medium fidelity · low cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: SEQ 1–7, SUS 0–100
- Tools: Single Ease Question (SEQ), System Usability Scale (SUS), post-task survey
- Notes: Satisfaction. SEQ is one question the moment the task ends; SUS scores the whole system 0–100.

### Session recordings / rage clicks

- Decision it informs: Where do real users struggle?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: medium fidelity · low cost · automated · lagging · pull delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: rage clicks, replays
- Tools: FullStory, Hotjar, PostHog replays
- AI can shorten this loop
- Notes: Qualitative signal at quantitative scale.

### Funnel / drop-off analysis

- Decision it informs: Where exactly do users abandon the flow?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: high fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: stage conversion %
- Tools: Amplitude funnels, GA4, PostHog funnels
- Fast proxy for: Feature adoption / activation
- Actions: redesign the cliff step; remove a step entirely
- Notes: A session recording shows one user struggling; the funnel shows the cliff a thousand walked off. The complementary aggregate loop to the recording.

### Thumbs up/down on AI output

- Decision it informs: Do users think the AI's answers are any good?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: low fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: % thumbs-up, regenerate %
- Tools: thumbs up/down, regenerate rate, edits to accepted output
- Fast proxy for: Retention curves
- Notes: Sparse and polarized, but the only label stream that comes from the people you actually serve. Regenerates and abandoned sessions are the behavioural check on the thumbs.

### AI synthesis of qualitative streams

- Decision it informs: What are thousands of tickets, calls and reviews saying this week?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: medium fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: themes/wk
- Tools: LLM tagging / clustering, Gong AI, support copilots
- Notes: Compresses the slowest qualitative loops from months to days, the same tickets and calls, read by something that never gets bored.

### Public reviews (G2 / app stores / Reddit)

- Decision it informs: What do people say about us where we can't moderate?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · low cost · human-mediated · lagging · pull delivery · continuous cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: stars, mentions
- Tools: G2, app store reviews, Reddit / HN mentions
- AI can shorten this loop
- Notes: Unsolicited, public, and permanent. Feedback that doubles as marketing or anti-marketing. Watch the complaints that repeat, not the star average.

### Support tickets

- Decision it informs: What is confusing or broken in practice?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · medium cost · human-mediated · lagging · push delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: tickets/wk
- Tools: Zendesk, Intercom
- AI can shorten this loop
- Notes: Reveals doc gaps and silent feature failures too.

### Time to first response

- Decision it informs: How quickly do customers hear a human?
- Latency: typically 4hr (range 1hr to 1d)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: hours
- Tools: Zendesk SLAs, Intercom
- Balanced by: First contact resolution
- AI can shorten this loop
- Notes: The fastest service loop, and the one customers judge you on before anything is even fixed.

### Resolution time

- Decision it informs: How long until the problem is actually solved?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: medium fidelity · medium cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: hours, days
- Tools: ticket analytics
- Balanced by: Ticket CSAT
- AI can shorten this loop
- Notes: MTTR for customer pain rather than infrastructure.

### First contact resolution

- Decision it informs: Do we fix it in one touch or bounce people around?
- Latency: typically 2wk (range 1wk to 1mo)
- Signal: medium fidelity · medium cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: % resolved first touch
- Tools: ticket reopens, transfer tracking
- Balanced by: Time to first response
- Notes: Low FCR is a process smell: knowledge gaps, bad tooling, or unclear ownership.

### Ticket CSAT

- Decision it informs: Was that interaction any good?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: low fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: % satisfied
- Tools: post-ticket surveys
- Balanced by: Resolution time
- Notes: Response bias runs hot. Mostly the delighted and the furious answer.

### Contact rate / tickets per customer

- Decision it informs: Is the product generating support burden?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: tickets per 100 customers
- Tools: ticket volume normalized by MAU
- Fast proxy for: Revenue churn / GRR
- Actions: fix the top ticket driver; add self-serve deflection
- Notes: Rising contact rate leads churn. The product is asking for help on customers' behalf.

### Fake door / prototype tests

- Decision it informs: Does anyone want this before we build it?
- Latency: typically 1wk (range 1d to 2wk)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: pre-pmf, growth
- Measured in: % clicked, signed up
- Tools: landing page tests, clickable prototypes
- Fast proxy for: Feature adoption / activation
- Notes: Cheapest possible value signal, pre-build.

### Sean Ellis PMF survey

- Decision it informs: How disappointed would users be if this disappeared?
- Latency: typically 2wk (range 3d to 1mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · on-demand cadence · event-kind
- Owner: leadership · Stages: pre-pmf, growth
- Measured in: % very disappointed
- Tools: 40% very-disappointed threshold, Superhuman PMF engine
- Framework: Sean Ellis
- Fast proxy for: Retention curves
- Actions: double down on the very-disappointed segment
- Notes: The canonical fast proxy for product-market fit, retention's answer, months earlier. The segment that says 'very disappointed' tells you who to build for.

### Willingness-to-pay research

- Decision it informs: What would they actually pay, before we bet the pricing page on it?
- Latency: typically 2wk (range 1wk to 1mo)
- Signal: medium fidelity · medium cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: acceptable price range
- Tools: Van Westendorp, conjoint analysis, pricing interviews
- Framework: Van Westendorp
- Fast proxy for: Pricing power
- Notes: The research loop underneath Pricing power. Stated willingness runs high, but the shape of the curve is real. Months faster than finding out at renewal.

### Feature request board / upvotes

- Decision it informs: What do engaged users think they want?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · low cost · human-mediated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: votes
- Tools: Canny, Productboard, GitHub issues
- AI can shorten this loop
- Notes: A demand signal from the vocal minority. Treat votes as evidence of pain, not as a spec. The comments are worth more than the counts.

### Weekly discovery interviews

- Decision it informs: What problem are people actually hiring us for?
- Latency: typically 1wk (range 3d to 2wk)
- Signal: high fidelity · medium cost · human-mediated · leading · pull delivery · periodic cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: interviews/wk, insights
- Tools: Teresa Torres' continuous discovery, opportunity solution trees
- AI can shorten this loop
- Notes: Torres' core move is compressing the value loop from quarterly research to weekly touchpoints.

### Kano survey

- Decision it informs: Is this feature a must-have, a delighter, or indifferent?
- Latency: typically 2wk (range 1wk to 1mo)
- Signal: medium fidelity · medium cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: category per feature
- Tools: functional/dysfunctional question pairs
- Framework: Kano
- Notes: A survey instrument, not just a 2x2. Classifies features by how their presence and absence feel.

### Opportunity scoring (importance vs satisfaction)

- Decision it informs: Which customer outcomes are important but badly served?
- Latency: typically 2wk (range 1wk to 1mo)
- Signal: medium fidelity · medium cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: importance − satisfaction
- Tools: ODI surveys, opportunity landscape
- Framework: ODI
- Notes: In Ulwick's outcome-driven framing, underserved outcomes are where value hides.

### Feature adoption / activation

- Decision it informs: Do people reach the value moment?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: % reaching value moment
- Tools: Amplitude, PostHog, GA
- Fast proxy for: Retention curves
- Actions: improve discoverability; move it into onboarding; kill the feature
- Notes: Leading proxy for retention.

### A/B experiments

- Decision it informs: Did this change move the outcome metric?
- Latency: typically 2wk (range 1wk to 2mo)
- Signal: high fidelity · medium cost · automated · leading · pull delivery · continuous cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: % lift (significant)
- Tools: Statsig, GrowthBook, Optimizely
- Balanced by: Retention curves
- Actions: ship the winner; kill the change; iterate the hypothesis
- Notes: High fidelity if powered properly; loop length set by traffic.

### NPS / CSAT

- Decision it informs: How do people feel about us?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: low fidelity · low cost · automated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: NPS −100…100, CSAT %
- Tools: Delighted, in-app surveys
- Fast proxy for: Retention curves
- Balanced by: Gross margin, Stickiness (DAU/MAU)
- Notes: Popular, cheap, and noisy.

### Retention curves

- Decision it informs: Does it keep solving the problem?
- Latency: typically 3mo (range 1mo to 6mo)
- Signal: high fidelity · high cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: leadership · Stages: pre-pmf, growth, scale
- Measured in: % retained by cohort
- Tools: cohort analysis
- Fast proxy for: LTV
- Balanced by: A/B experiments, Trial / freemium conversion
- Actions: fix the leakiest cohort; revisit the ICP
- Notes: You cannot read it faster than time passes. Use activation as the fast proxy.

### Trial / freemium conversion

- Decision it informs: Do free users become paying customers?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: high fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: trial→paid %
- Tools: trial→paid dashboards, PLG funnels
- Fast proxy for: MRR / ARR growth
- Balanced by: Retention curves
- Actions: tune paywall timing; shorten path to the value moment; requalify who gets a trial
- Notes: The commercial moment of truth for product-led growth, value delivered but not yet paid for. Gaming it with pressure tactics shows up two quarters later in retention.

### Stickiness (DAU/MAU)

- Decision it informs: Is the product becoming a habit?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: DAU/MAU %
- Tools: Amplitude, Mixpanel, DAU/MAU dashboards
- Fast proxy for: Retention curves
- Balanced by: NPS / CSAT
- Actions: build the habit hook; re-check the product's natural cadence
- Notes: The fast proxy for retention, but only meaningful against the product's natural cadence. A payroll product at 20% DAU/MAU is a triumph; a social product at 20% is dying.

### Churn interviews / exit surveys

- Decision it informs: Why did they actually leave?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: high fidelity · medium cost · human-mediated · lagging · pull delivery · continuous cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: reasons coded
- Tools: cancellation-flow surveys, churned-customer calls
- AI can shorten this loop
- Actions: fix the top churn driver; tighten targeting upstream
- Notes: Win/loss analysis for the customers you already had. The cancel-flow survey lies less than you'd think; the call afterwards lies least.

### Expansion / referral revenue

- Decision it informs: Is it valuable enough to buy more of?
- Latency: typically 6mo (range 3mo to 1yr)
- Signal: high fidelity · high cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: NRR %
- Tools: NRR, referral tracking
- Balanced by: Revenue churn / GRR
- Notes: The strongest value signal there is, and the slowest.

### Win/loss analysis

- Decision it informs: Why do we win and lose deals?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: high fidelity · medium cost · human-mediated · lagging · pull delivery · on-demand cadence · event-kind
- Owner: leadership · Stages: growth, scale
- Measured in: win rate %, reasons
- Tools: sales call reviews, Gong
- Balanced by: Pricing realization / discount rate
- AI can shorten this loop
- Actions: arm sales against the top objection; fix the losable reasons
- Notes: Competitive truth from people who just made a real choice.

## Reach · Can we get it to people?

### Paid ads CTR / CPA

- Decision it informs: Does this message pull this audience?
- Latency: typically 1d (range 1hr to 3d)
- Signal: medium fidelity · medium cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: CTR %, $ CPA
- Tools: Google Ads, Meta Ads
- Fast proxy for: ROAS / channel CAC
- Balanced by: Landing conversion rate (CVR)
- Notes: The fastest reach loop, Weinberg's rational early choice for exactly this reason.

### ROAS / channel CAC

- Decision it informs: Is this channel economically viable?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: medium fidelity · high cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: return multiple
- Tools: attribution tooling
- Fast proxy for: LTV:CAC ratio
- Balanced by: Ad-spend liquidation
- Actions: shift budget between channels; kill the channel
- Notes: Fidelity degrades as attribution gets murkier.

### Ad-spend liquidation

- Decision it informs: How fast does ad spend come back as cash?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: high fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: % of spend recovered, days to recoup
- Tools: blended spend vs revenue, Shopify
- Fast proxy for: LTV:CAC ratio
- Balanced by: ROAS / channel CAC
- Notes: Cash-recovery speed, not profitability. 100%+ day-0 liquidation lets you recycle spend almost without limit; a slow liquidator funds the gap from its own pocket. Blended, it needs no attribution.

### CPM / auction pressure

- Decision it informs: What is the market charging us for attention?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: $ per 1k impressions
- Tools: Meta CPM, TikTok CPM, auction insights
- Notes: Market feedback, not just cost. Rising CPM means competition or a creative the auction dislikes.

### Cost per click (CPC)

- Decision it informs: What is the auction charging us per click?
- Latency: typically 1d (range 1hr to 3d)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: $ per click
- Tools: Google Ads, Meta Ads
- Fast proxy for: ROAS / channel CAC
- Balanced by: Landing conversion rate (CVR)
- Notes: CPM divided by CTR. A rising CPC means attention got pricier or the creative stopped earning clicks. Cheap clicks are only cheap if they convert, so read it against CVR.

### Landing conversion rate (CVR)

- Decision it informs: Does the click become a customer action?
- Latency: typically 3d (range 1d to 1mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: % of clicks converting
- Tools: pixel events, GA4 funnels
- Fast proxy for: ROAS / channel CAC
- Balanced by: Paid ads CTR / CPA, Cost per click (CPC)
- Notes: Separates ad problems from page problems; CTR without CVR is expensive applause.

### Quality / relevance score

- Decision it informs: Does the platform think this ad deserves cheap reach?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: score 1–10
- Tools: Google Ads quality score, Meta relevance diagnostics
- Notes: The platform's own model grading you, a proxy loop for user response, machine-mediated.

### Frequency / creative fatigue

- Decision it informs: Has this creative worn out its audience?
- Latency: typically 2wk (range 3d to 1mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: frequency, CTR decay
- Tools: frequency caps, CTR decay by cohort
- Actions: rotate creative; expand the audience
- Notes: Leading indicator of CPA rot; when it climbs, feed the machine new creative.

### Incrementality / lift tests

- Decision it informs: Did the spend cause the revenue, or just claim it?
- Latency: typically 1mo (range 2wk to 3mo)
- Signal: high fidelity · high cost · automated · lagging · pull delivery · on-demand cadence · event-kind
- Owner: leadership · Stages: scale
- Measured in: incremental lift %
- Tools: geo holdouts, conversion lift studies
- Notes: The audit of every faster paid loop. Attribution flatters, holdouts tell the truth.

### Influencer / sponsorship performance

- Decision it informs: Did borrowed audiences convert to ours?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: low fidelity · high cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: code redemptions, CPA
- Tools: promo codes, dedicated landing pages, sponsor-read attribution
- Notes: Paid reach with a human face. Attribution is murkier than ads, trust transfer is stronger. Codes and dedicated URLs are the only meter you can trust.

### Last-click attribution

- Decision it informs: Which channel touched them last?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: low fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: % credit by channel
- Tools: GA4 default reports, platform dashboards
- Fast proxy for: Incrementality / lift tests
- Notes: Fast, free, and systematically wrong. Over-credits bottom-funnel, erases everything that built the intent.

### Multi-touch / data-driven attribution

- Decision it informs: How should credit split across the journey?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · medium cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: % credit by touch
- Tools: GA4 DDA, MTA platforms
- Fast proxy for: Incrementality / lift tests
- Notes: Better than last-click, but privacy changes (iOS, cookie loss) have blinded much of what it once saw.

### Self-reported attribution

- Decision it informs: Where do customers say they found us?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · low cost · human-mediated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: % by source
- Tools: 'How did you hear about us?' field
- Notes: Catches dark social and word of mouth that pixels can't see; biased by memory, candid about influence.

### Marketing mix modeling

- Decision it informs: How should budget split across all channels?
- Latency: typically 6mo (range 3mo to 1yr)
- Signal: low fidelity · high cost · automated · lagging · pull delivery · periodic cadence · event-kind
- Owner: leadership · Stages: scale
- Measured in: $ effect per channel
- Tools: MMM (Robyn, Meridian)
- Notes: The slowest reach loop, statistical, assumption-heavy, but the only one that sees the whole board.

### Delivered / inbox placement

- Decision it informs: Did it even reach the inbox?
- Latency: typically 15min (range 1min to 4hr)
- Signal: high fidelity · low cost · automated · leading · pull delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: % inboxed
- Tools: ESP delivery logs, seed / inbox-placement tests
- Notes: Delivered isn't the same as inboxed. The spam folder counts as delivered.

### Bounce rate

- Decision it informs: Are we mailing dead addresses?
- Latency: typically 10min (range 30s to 2d)
- Signal: high fidelity · low cost · automated · leading · pull delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: % bounced
- Tools: ESP bounce logs, list verification
- Notes: Hard bounces return in seconds; soft bounces retry over days.

### Sender reputation

- Decision it informs: Is our sender reputation healthy?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: reputation, complaint %
- Tools: Postmaster Tools, SNDS, complaint feedback loops
- Notes: Reputation erodes or heals over days to weeks. Complaints are the real killer.

### Spam / reject rate

- Decision it informs: Are mailboxes blocking this send?
- Latency: typically 5min (range 30s to 1hr)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: % rejected
- Tools: Postmaster Tools, SNDS, complaint feedback loops
- Fast proxy for: Sender reputation
- Notes: Rejects and spam-folder placement show up within minutes, a hard signal a send is being filtered.

### Open rate

- Decision it informs: Did the subject line earn a look?
- Latency: typically 4hr (range 5min to 3d)
- Signal: low fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: % opened
- Tools: ESP open tracking
- Balanced by: Unsubscribe / list churn
- Notes: Increasingly unreliable. Apple Mail Privacy Protection auto-fires opens and inflates the number.

### Click-through rate

- Decision it informs: Did the content earn a click, not just a look?
- Latency: typically 4hr (range 5min to 3d)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: % clicked
- Tools: ESP click tracking, UTM links
- Notes: The email signal Apple didn't break. A click is intent in a way an auto-fired open never was.

### Unsubscribe / list churn

- Decision it informs: Are we burning the list to hit this month's number?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: high fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: % per send
- Tools: ESP unsubscribe reports, net list growth
- Fast proxy for: Sender reputation
- Balanced by: Open rate
- Actions: cut send frequency; resegment the list
- Notes: Every send spends trust; unsubscribes are the receipt. Rising unsubs lead spam complaints, which lead reputation damage.

### Cold outreach reply rates

- Decision it informs: Does the pitch resonate with this segment?
- Latency: typically 3d (range 1d to 2wk)
- Signal: medium fidelity · low cost · human-mediated · leading · push delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: % replied
- Tools: sequences, reply tracking
- Fast proxy for: Meetings booked / pipeline
- Actions: rewrite the pitch; change the segment
- Notes: Cheap message-market fit signal.

### Meetings booked / pipeline

- Decision it informs: Is outbound converting to real demand?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: medium fidelity · medium cost · human-mediated · leading · pull delivery · continuous cadence · trend-kind
- Owner: leadership · Stages: pre-pmf, growth, scale
- Measured in: meetings, $ pipeline
- Tools: CRM
- Fast proxy for: MRR / ARR growth
- Balanced by: Pipeline stage conversion
- Notes: Leading indicator for revenue, lagging for messaging.

### Pipeline stage conversion

- Decision it informs: Where does the sales machine actually jam?
- Latency: typically 3mo (range 1mo to 6mo)
- Signal: medium fidelity · low cost · automated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: % conversion by stage
- Tools: CRM stage reports, demo→proposal→close rates
- Balanced by: Meetings booked / pipeline
- Actions: fix the jam stage; qualify harder upstream
- Notes: Pipeline says how much is in the machine; win/loss says why deals die; this says where. The jam is usually one stage, and it's usually not the one sales says it is.

### Sales cycle length

- Decision it informs: How long from first meeting to signature?
- Latency: typically 3mo (range 1mo to 6mo)
- Signal: medium fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: days to close
- Tools: CRM stage timestamps
- Notes: The lead time of the revenue machine. Lengthening cycles are an early sign of weakening urgency, wrong ICP, or a new blocker in procurement.

### Content engagement

- Decision it informs: Does this topic and angle resonate?
- Latency: typically 3d (range 1hr to 2wk)
- Signal: low fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: shares, time on page
- Tools: shares, comments, time on page
- Notes: Fast but weakly correlated with buying intent.

### Keyword research (difficulty vs. potential)

- Decision it informs: Is this keyword worth targeting?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: medium fidelity · low cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: volume vs difficulty
- Tools: Ahrefs / Semrush Keywords Explorer, Google Keyword Planner, Search Console, AlsoAsked
- Fast proxy for: Keyword rankings / SERP position
- AI can shorten this loop
- Notes: A cheap forecast up front, but Google takes months to confirm the bet. Content's version of a feasibility spike.

### Technical SEO / indexation

- Decision it informs: Can Google even crawl and index us?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: high fidelity · low cost · automated · leading · pull delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: pages indexed
- Tools: Search Console coverage, Screaming Frog, Sitebulb, Ahrefs / Semrush site audit
- Notes: The precondition for everything else. An unindexed page can't rank at all.

### Keyword rankings / SERP position

- Decision it informs: Where do we actually rank for our targets?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: medium fidelity · medium cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: SERP position
- Tools: rank tracker, Search Console, Ahrefs / Semrush / SE Ranking
- Fast proxy for: Organic traffic & CTR
- Actions: refresh the content; build links; target easier keywords
- Notes: Daily numbers, monthly truth. The loop length belongs to Google, not you.

### Organic traffic & CTR

- Decision it informs: Is search actually sending clicks?
- Latency: typically 1mo (range 3d to 3mo)
- Signal: high fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: clicks, CTR %
- Tools: Search Console, GA4, Plausible
- Notes: You can rank #3 and still get no clicks. The title and meta have to earn them.

### Domain authority / backlink profile

- Decision it informs: Is our link authority growing?
- Latency: typically 3mo (range 1wk to 1yr)
- Signal: medium fidelity · medium cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: DR/DA 0–100
- Tools: Ahrefs DR / Moz DA, Majestic, referring-domain monitors
- Fast proxy for: Keyword rankings / SERP position
- Notes: It feeds rankings rather than reflecting them. Compounds slowly, decays if you stop earning links.

### Content decay / refresh signal

- Decision it informs: Which pages are slipping in traffic?
- Latency: typically 6mo (range 1mo to 1yr)
- Signal: medium fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: % traffic lost
- Tools: Search Console, GA4, Ahrefs / Semrush traffic trends
- Actions: refresh or consolidate; redirect and prune
- Notes: Rankings rot silently. The fix is usually refreshing old winners, not writing new.

### AI search visibility (AEO / GEO)

- Decision it informs: Do the answer engines mention us when buyers ask?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: low fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: % of answers citing you
- Tools: Profound, share-of-model tracking, LLM brand monitoring
- Notes: Search traffic is migrating into answers. Measurement is young and noisy, but being absent from the answer is a silent zero.

### App store ratings & ASO

- Decision it informs: Does the store surface us, and do the stars invite the tap?
- Latency: typically 1mo (range 1d to 3mo)
- Signal: medium fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: avg rating, keyword rank
- Tools: App Store Connect, Play Console, ASO keyword tools
- Actions: prompt ratings after value moments; answer reviews; fix the top-complained bug
- Notes: SEO's mobile sibling loop, with ranking driven by ratings, retention and keyword relevance. The rating is a lagging echo of product quality that then leads installs.

### Views / impressions

- Decision it informs: Did the algorithm pick this up?
- Latency: typically 6hr (range 15min to 2d)
- Signal: low fidelity · low cost · automated · leading · pull delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: views
- Tools: TikTok analytics, YouTube Studio, IG insights
- Balanced by: Follows-per-view
- Notes: The fastest reach signal that exists, and the noisiest. It shows distribution luck more than durable demand.

### Hook rate / swipe-aways

- Decision it informs: Do the first three seconds earn the next thirty?
- Latency: typically 1d (range 1hr to 3d)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: % past 3 seconds
- Tools: 3-sec retention, swipe-away rate, thumbstop
- Notes: The most actionable creative signal. It isolates the open from the content.

### Audience retention / watch time

- Decision it informs: Where exactly do people leave?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: high fidelity · low cost · automated · leading · pull delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: % watched, drop-off point
- Tools: avg view duration, retention curves, watch time
- Notes: Second-by-second feedback on content quality; the closest thing organic has to a session recording.

### Completion / loop rate

- Decision it informs: Do they watch to the end, or rewatch?
- Latency: typically 12hr (range 1hr to 3d)
- Signal: high fidelity · low cost · automated · leading · pull delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: % completed, rewatches
- Tools: platform analytics, average view duration, loops / rewatches
- Notes: Completions and rewatches are the signal short-form algorithms reward hardest.

### Shares & saves

- Decision it informs: Is it worth passing on or keeping?
- Latency: typically 1d (range 1hr to 3d)
- Signal: high fidelity · low cost · automated · leading · pull delivery · continuous cadence · event-kind
- Owner: ic · Stages: pre-pmf, growth, scale
- Measured in: shares, saves
- Tools: platform analytics, shares, saves / bookmarks
- Notes: The real currency of short-form distribution. Saves and shares outrank likes by a mile.

### Comments / audience conversation

- Decision it informs: What is the audience actually saying back?
- Latency: typically 6hr (range 15min to 3d)
- Signal: medium fidelity · low cost · human-mediated · leading · pull delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: comments per video
- Tools: comment sections, pinned-comment replies, comment sentiment
- AI can shorten this loop
- Notes: The qualitative half of engagement. Comments tell you why the retention curve did what it did, and hand you the next video's hook. Replying early also feeds the algorithm.

### Follows-per-view

- Decision it informs: Does a view convert into a follower?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: medium fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: follows per 1k views
- Tools: platform analytics, follows attributed to post
- Balanced by: Views / impressions
- Notes: Reach that doesn't convert to follows is rented attention, not a compounding audience.

### Engagement rate

- Decision it informs: Does this provoke a response?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: % engaged
- Tools: likes, comments, shares, saves
- Notes: Comments and saves outweigh likes. Platforms rank by effortful signals.

### Comments & DMs

- Decision it informs: What is the audience saying back, in words?
- Latency: typically 6hr (range 15min to 3d)
- Signal: medium fidelity · low cost · human-mediated · leading · pull delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: comments, DMs
- Tools: comment threads, DM inbox, social listening
- AI can shorten this loop
- Notes: Engagement rate counts the conversation; reading it is a different loop. DMs are where the buying questions arrive.

### Follower / subscriber growth

- Decision it informs: Are we compounding an audience or renting attention?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: medium fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: net follows/mo
- Tools: subs per video, follower conversion
- Notes: The compounding metric. Views are the loop, followers are what the loop accumulates.

### Referral / K-factor

- Decision it informs: Does usage generate more usage?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: medium fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: K-factor
- Tools: invite tracking
- Notes: Compounding loops need cycles before the signal is real.

### PR / press coverage

- Decision it informs: Does the story land with media?
- Latency: typically 2wk (range 1d to 3mo)
- Signal: low fidelity · medium cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: leadership · Stages: pre-pmf, growth, scale
- Measured in: placements, reach
- Tools: coverage tracking
- Notes: Unpredictable timing, spiky signal.

### Partnership pipeline

- Decision it informs: Can others distribute us?
- Latency: typically 3mo (range 1mo to 1yr)
- Signal: medium fidelity · high cost · human-mediated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: partner-sourced $
- Tools: co-sell tracking, revenue attribution
- Notes: Slow to start, slow to stop.

### Events / webinar performance

- Decision it informs: Did showing up in person (or live) create pipeline?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · high cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: attendance, pipeline $
- Tools: registration / attendance, pipeline sourced, booth conversations
- Notes: Spiky, expensive, and hard to attribute, but the only channel where the market talks back in full sentences.

### Community activity & sentiment

- Decision it informs: Is the community compounding or decaying?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: medium fidelity · medium cost · human-mediated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: active members, posts/wk
- Tools: Discord / Slack activity, forum posts, unprompted advocacy
- Notes: Slow to build, fast to read once it exists, the channel where users answer each other and tell you the truth in public.

### Brand awareness / recall

- Decision it informs: Do people think of us unprompted?
- Latency: typically 1yr (range 3mo to 3yr)
- Signal: low fidelity · high cost · human-mediated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: scale
- Measured in: % unaided recall
- Tools: surveys, share of voice
- Notes: Builds slowly, decays slowly. Parallelize brand bets to keep some signal flowing.

## Team · Can we sustain the people doing the thing?

### Time-to-hire

- Decision it informs: How long from opening a role to a signed offer?
- Latency: typically 1mo (range 2wk to 3mo)
- Signal: medium fidelity · medium cost · human-mediated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: days role → offer
- Tools: ATS funnel metrics, stage conversion
- Balanced by: Quality of hire / ramp time
- Notes: The recruitment loop's own cycle time. Slow hiring loops mean losing candidates to faster ones.

### Sourcing response rate

- Decision it informs: Does our pitch open doors with talent we want?
- Latency: typically 3d (range 1d to 2wk)
- Signal: medium fidelity · low cost · human-mediated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: % replied
- Tools: outreach reply rates, ATS sourcing metrics
- Fast proxy for: Offer acceptance rate
- Notes: Cold outreach reply rates for hiring, message-market fit where the market is candidates. Fast feedback on employer brand.

### Interview signal

- Decision it informs: Will this person succeed here?
- Latency: typically 1wk (range 1hr to 1mo)
- Signal: low fidelity · high cost · human-mediated · leading · pull delivery · continuous cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: hire / no-hire confidence
- Tools: structured loops, work samples
- Fast proxy for: Quality of hire / ramp time
- Notes: Everyone's most confident low-fidelity loop.

### Offer acceptance rate

- Decision it informs: Is our pitch to talent working?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: medium fidelity · low cost · human-mediated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: % accepted
- Tools: ATS metrics
- Notes: Market feedback on your employer brand and comp.

### Compensation benchmarking

- Decision it informs: Are we paying market before the market tells us?
- Latency: typically 6mo (range 1mo to 1yr)
- Signal: medium fidelity · medium cost · human-mediated · leading · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: percentile vs market
- Tools: Radford / Pave, offer decline reasons
- Fast proxy for: Offer acceptance rate, Attrition / regretted departures
- Actions: adjust bands ahead of cycle; target retention grants at flight risks
- Notes: Offer acceptance and attrition are the complementary lagging loops. By the time they move, you've been below market for a year.

### Quality of hire / ramp time

- Decision it informs: Was the interview signal right?
- Latency: typically 6mo (range 3mo to 1yr)
- Signal: medium fidelity · high cost · human-mediated · lagging · pull delivery · periodic cadence · event-kind
- Owner: leadership · Stages: growth, scale
- Measured in: ramp months, manager rating
- Tools: ramp milestones, manager review
- Balanced by: Time-to-hire
- Notes: 6–12 months to close the loop the interview opened.

### Onboarding ramp / time-to-first-commit

- Decision it informs: Is the new hire reaching productive ground on schedule?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · event-kind
- Owner: team · Stages: growth, scale
- Measured in: days to first PR
- Tools: first PR merged, 30-60-90 milestones
- Fast proxy for: Quality of hire / ramp time
- Notes: The fast proxy for quality-of-hire, and a test of your own onboarding, docs and dev environment as much as of the hire.

### 1:1 sentiment

- Decision it informs: How is this person really doing?
- Latency: typically 1wk (range 3d to 2wk)
- Signal: medium fidelity · medium cost · human-mediated · leading · pull delivery · periodic cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: sentiment
- Tools: weekly 1:1s
- Fast proxy for: Attrition / regretted departures
- Actions: adjust the load; escalate support early
- Notes: The manager's fastest loop, and easy to over-read.

### Team retrospectives

- Decision it informs: Is our process helping or hurting?
- Latency: typically 2wk (range 1wk to 1mo)
- Signal: medium fidelity · medium cost · human-mediated · leading · pull delivery · periodic cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: improvements shipped
- Tools: sprint retros
- Actions: pick one improvement and assign an owner
- Notes: An integration test for the team, if run honestly.

### Developer experience survey

- Decision it informs: Where does the day-to-day work actually hurt?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: medium fidelity · low cost · human-mediated · leading · pull delivery · periodic cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: DevEx score
- Tools: DX, SPACE framework, internal DevEx surveys
- Fast proxy for: Tech debt accumulation
- Notes: Tech debt measured from the inside. The people slowed down by it are the instrument. Quarterly signal for a problem you can fix weekly.

### eNPS / engagement surveys

- Decision it informs: Is morale trending?
- Latency: typically 3mo (range 1mo to 6mo)
- Signal: low fidelity · low cost · automated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: eNPS −100…100
- Tools: CultureAmp, Officevibe
- Fast proxy for: Attrition / regretted departures
- AI can shorten this loop
- Notes: Quarterly cadence, survey-fatigue noise.

### Pulse surveys

- Decision it informs: How is the team this week, not last quarter?
- Latency: typically 1wk (range 1d to 2wk)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · periodic cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: pulse score
- Tools: weekly pulse tools, Officevibe, Slack polls
- Fast proxy for: eNPS / engagement surveys
- AI can shorten this loop
- Actions: act visibly on the top pain; close the loop in public
- Notes: eNPS at a cadence that can still change the outcome. Two questions weekly beat forty questions quarterly. Survey fatigue is the failure mode.

### Attrition / regretted departures

- Decision it informs: Did we lose people we needed?
- Latency: typically 1yr (range 3mo to 2yr)
- Signal: high fidelity · high cost · human-mediated · lagging · push delivery · continuous cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: %, regretted count
- Tools: exit interviews, Glassdoor
- Actions: fix the systemic cause; start succession planning
- Notes: Unambiguous, and far too late.

### Meeting load / focus time

- Decision it informs: Do makers still have time to make?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: focus hrs/wk
- Tools: calendar analytics, focus-time metrics
- Fast proxy for: Attrition / regretted departures
- Notes: Burnout's leading indicator hides in the calendar. Fully automated and almost never looked at.

### On-call load / pages per person

- Decision it informs: Is the pager burning anyone out?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: high fidelity · low cost · automated · leading · pull delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: pages per person/wk
- Tools: PagerDuty analytics, pages per rotation, night pages
- Fast proxy for: Attrition / regretted departures
- Actions: fix the top alert sources; grow the rotation
- Notes: Reliability debt lands on people before it lands on dashboards. Night pages predict resignations.

### Bus factor / knowledge concentration

- Decision it informs: What breaks if this one person leaves?
- Latency: typically 3mo (range 1mo to 1yr)
- Signal: medium fidelity · low cost · automated · leading · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: % single-owner files
- Tools: git ownership analysis, CodeScene knowledge maps
- Actions: pair on single-owner areas; write down the tribal knowledge
- Notes: Git-derived like hotspots, but measuring fragility instead of cost: the files only one person has ever touched. Nobody runs this loop until the person resigns. The point is to run it before.

### Growth / promotion readiness

- Decision it informs: Are people growing here?
- Latency: typically 6mo (range 3mo to 1yr)
- Signal: medium fidelity · medium cost · human-mediated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: promotions, readiness
- Tools: career frameworks, calibration
- Notes: Slow loop; skip-levels and skill-gap reviews are the faster proxies.

### Performance reviews / 360 feedback

- Decision it informs: Is each person growing, stuck, or misplaced?
- Latency: typically 3mo (range 1mo to 6mo)
- Signal: medium fidelity · high cost · human-mediated · lagging · pull delivery · periodic cadence · event-kind
- Owner: leadership · Stages: growth, scale
- Measured in: ratings distribution
- Tools: review cycles, 360 feedback, calibration
- Notes: High ceremony, twice a year, and lossy. Nothing in a good review should be news if the 1:1 loop is working.

### OKR check-ins / goal scoring

- Decision it informs: Are we progressing on what we said mattered, and were they the right goals?
- Latency: typically 3mo (range 1wk to 6mo)
- Signal: medium fidelity · medium cost · human-mediated · leading · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: OKR score 0–1, confidence
- Tools: quarterly OKRs, weekly confidence check-ins
- Framework: OKR
- Actions: kill the stalled key result; reallocate focus
- Notes: Every other loop here is a sensor; this is where the org decides which sensors it steers by. Failure mode: set-and-forget, where grading day is the first time anyone looks.

## Risk & Moat · Do the economics allow success to continue?

### Cloud spend / burn alerts

- Decision it informs: Are costs running away?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: high fidelity · low cost · automated · leading · push delivery · continuous cadence · event-kind
- Owner: leadership · Stages: pre-pmf, growth, scale
- Measured in: $/day vs budget
- Tools: billing alerts, FinOps dashboards
- Fast proxy for: Runway / burn multiple
- Actions: kill zombie infra; rightsize instances
- Notes: One of the few fast loops in the risk track.

### AI / inference spend

- Decision it informs: Is inference eating into gross margin off the radar?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: high fidelity · low cost · automated · lagging · push delivery · continuous cadence · trend-kind
- Owner: leadership · Stages: pre-pmf, growth, scale
- Measured in: $/mo, % of COGS
- Tools: provider billing alerts, FinOps dashboards
- Fast proxy for: Gross margin
- Notes: The line item that turns software margins into hardware margins. Its fast proxy is cost-per-request in Build.

### Runway / burn multiple

- Decision it informs: How long can we keep playing?
- Latency: typically 3mo (range 1mo to 6mo)
- Signal: high fidelity · low cost · automated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: pre-pmf, growth, scale
- Measured in: months of cash
- Tools: finance reviews
- Balanced by: MRR / ARR growth
- Actions: cut burn; raise; grow into the gap
- Notes: A readout of every other loop, and you still have to check it.

### Investor / fundraising feedback

- Decision it informs: Does the capital market believe our story?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: low fidelity · medium cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: leadership · Stages: pre-pmf, growth
- Measured in: term sheets, passes
- Tools: partner meetings, term sheets, pass notes
- Notes: Noisy and polite. Passes rarely say why. But a fundraise is the one time the market prices your whole strategy at once.

### Gross margin

- Decision it informs: Does each unit sold actually make money?
- Latency: typically 3mo (range 1mo to 6mo)
- Signal: high fidelity · low cost · automated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: %
- Tools: monthly close, FinOps
- Balanced by: NPS / CSAT
- Notes: The ceiling on every other financial metric; software margins hide infra and support costs.

### Pricing realization / discount rate

- Decision it informs: How much of list price do we actually collect?
- Latency: typically 3mo (range 1mo to 1yr)
- Signal: high fidelity · low cost · automated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: % of list realized, avg discount %
- Tools: CPQ reports, discount approval logs
- Fast proxy for: Pricing power
- Balanced by: Win/loss analysis
- Actions: tighten discount approvals; reprice the list
- Notes: The sales team's revealed opinion of your pricing power. Heavy discounting shows up here quarters before it shows up in gross margin.

### MRR / ARR growth

- Decision it informs: Is revenue compounding?
- Latency: typically 3mo (range 1mo to 1yr)
- Signal: high fidelity · low cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: leadership · Stages: pre-pmf, growth, scale
- Measured in: $ ARR, % growth
- Tools: revenue dashboards
- Balanced by: Revenue churn / GRR, Runway / burn multiple
- Notes: The scoreboard everyone watches; explains nothing on its own.

### Revenue churn / GRR

- Decision it informs: How much revenue leaks before expansion papers over it?
- Latency: typically 3mo (range 1mo to 1yr)
- Signal: high fidelity · medium cost · automated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: % revenue retained
- Tools: gross revenue retention
- Balanced by: MRR / ARR growth, Expansion / referral revenue, Pricing power
- Actions: fix the churn drivers; invest in customer success
- Notes: GRR is the stricter number; NRR lets expansion hide a leaky bucket.

### Cash collection / DSO

- Decision it informs: Is billed revenue becoming actual cash?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: high fidelity · low cost · automated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: days outstanding
- Tools: days sales outstanding, aging receivables
- Fast proxy for: Runway / burn multiple
- Notes: Revenue is an opinion, cash is a fact. Rising DSO shortens runway while the ARR chart still smiles.

### Budget vs actuals (monthly close)

- Decision it informs: Are we spending what we planned, where we planned?
- Latency: typically 1mo (range 1wk to 3mo)
- Signal: high fidelity · medium cost · human-mediated · lagging · pull delivery · periodic cadence · event-kind
- Owner: leadership · Stages: growth, scale
- Measured in: % variance vs plan
- Tools: monthly close, headcount vs plan, variance reports
- Fast proxy for: Runway / burn multiple
- Actions: reforecast; freeze discretionary spend
- Notes: Runway says how long the money lasts; the close says where the plan is leaking. Headcount is the line item that matters. Most of the rest is rounding.

### Revenue forecast accuracy

- Decision it informs: Do our revenue forecasts mean anything?
- Latency: typically 3mo (range 1mo to 6mo)
- Signal: medium fidelity · low cost · human-mediated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: forecast error %
- Tools: forecast vs actual by quarter, pipeline-weighted forecasts
- Actions: recalibrate stage weights; sandbag-proof the pipeline review
- Notes: The say:do ratio of the revenue side. A forecast that is always wrong in the same direction is a negotiating position, not a forecast.

### 13-week cash flow forecast

- Decision it informs: Will the cash be there when the bills land?
- Latency: typically 1wk (range 1d to 1mo)
- Signal: high fidelity · medium cost · human-mediated · leading · pull delivery · periodic cadence · event-kind
- Owner: leadership · Stages: growth, scale
- Measured in: weeks of cover, forecast vs actual
- Tools: rolling 13-week model
- Fast proxy for: Runway / burn multiple
- Notes: DSO tells you receivables are slipping; this tells you whether payroll clears in week nine. Standard practice at scale, survival practice in a crunch.

### Revenue concentration

- Decision it informs: How much of the company walks out if one logo does?
- Latency: typically 3mo (range 1mo to 6mo)
- Signal: high fidelity · low cost · automated · leading · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: % ARR in top 5
- Tools: top-N customer share of ARR
- Notes: A single number that reprices every other metric. 40% in one account means their roadmap is your roadmap.

### Blended CAC

- Decision it informs: What does a customer actually cost, all-in?
- Latency: typically 3mo (range 1mo to 6mo)
- Signal: medium fidelity · medium cost · automated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: $ per customer
- Tools: S&M spend / new customers
- Fast proxy for: LTV:CAC ratio
- Notes: Channel CAC (in Reach) tells you where to spend; blended CAC tells you if the machine works.

### CAC payback period

- Decision it informs: How long until a customer repays their acquisition cost?
- Latency: typically 3mo (range 1mo to 1yr)
- Signal: medium fidelity · medium cost · automated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: growth, scale
- Measured in: months
- Tools: cohort payback curves
- Fast proxy for: LTV:CAC ratio
- Notes: A faster, less model-dependent proxy for LTV:CAC, often the better operating metric.

### LTV

- Decision it informs: What is a customer worth over their lifetime?
- Latency: typically 1yr (range 3mo to 2yr)
- Signal: low fidelity · high cost · automated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: scale
- Measured in: $ lifetime value
- Tools: cohort LTV models
- Notes: A forecast dressed as a measurement, fidelity depends entirely on churn assumptions.

### LTV:CAC ratio

- Decision it informs: Is the whole growth machine economically sound?
- Latency: typically 1yr (range 3mo to 2yr)
- Signal: low fidelity · high cost · automated · lagging · pull delivery · periodic cadence · trend-kind
- Owner: leadership · Stages: scale
- Measured in: ratio
- Tools: unit economics reviews
- Notes: The classic 3x rule of thumb; inherits the noise of both its inputs.

### Platform / API dependency changes

- Decision it informs: Can a platform kill us by email?
- Latency: typically 1mo (range 1d to 1yr)
- Signal: low fidelity · medium cost · human-mediated · leading · pull delivery · continuous cadence · event-kind
- Owner: leadership · Stages: pre-pmf, growth, scale
- Measured in: breaking changes flagged
- Tools: changelog monitoring, app store relations
- Notes: Most startups have no loop here until the late-night email arrives.

### Vendor / provider concentration

- Decision it informs: Which single supplier could take us down or reprice us?
- Latency: typically 3mo (range 1mo to 1yr)
- Signal: low fidelity · low cost · human-mediated · leading · pull delivery · periodic cadence · event-kind
- Owner: leadership · Stages: growth, scale
- Measured in: % spend per vendor
- Tools: dependency inventory, provider reviews, exit-cost estimates
- Notes: Platform risk's quieter sibling. You chose these dependencies. The AI era added a new one: a single model provider under everything.

### Regulatory & legal signals

- Decision it informs: Are we allowed to keep doing this?
- Latency: typically 6mo (range 1mo to 2yr)
- Signal: medium fidelity · high cost · human-mediated · leading · pull delivery · periodic cadence · event-kind
- Owner: leadership · Stages: pre-pmf, growth, scale
- Measured in: open issues
- Tools: counsel reviews, policy tracking, audits
- AI can shorten this loop
- Notes: Ignored until crisis; loop length set by governments.

### Compliance audits (SOC2 / ISO)

- Decision it informs: Can we prove we do what we claim?
- Latency: typically 3mo (range 1mo to 1yr)
- Signal: high fidelity · high cost · human-mediated · leading · pull delivery · periodic cadence · event-kind
- Owner: leadership · Stages: growth, scale
- Measured in: findings, attestation
- Tools: SOC2 Type II, ISO 27001, continuous compliance monitors
- AI can shorten this loop
- Notes: An annual exam with continuous homework. Compliance platforms turned a yearly panic into a standing loop, mostly.

### Fraud / abuse / chargeback rate

- Decision it informs: Is someone exploiting the product faster than we notice?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: medium fidelity · medium cost · automated · lagging · push delivery · continuous cadence · trend-kind
- Owner: team · Stages: growth, scale
- Measured in: chargeback %, abuse reports
- Tools: chargeback monitoring, abuse reports, anomaly detection
- Notes: One of the few genuinely fast Risk loops, and one where slow response compounds: card networks, app stores and users all keep score.

### Competitor launches

- Decision it informs: Is our position eroding?
- Latency: typically 1mo (range 1wk to 6mo)
- Signal: low fidelity · low cost · human-mediated · leading · pull delivery · continuous cadence · event-kind
- Owner: leadership · Stages: pre-pmf, growth, scale
- Measured in: launches tracked
- Tools: competitive intel, teardowns
- AI can shorten this loop
- Notes: Indirect feedback about your own strategy.

### Churn by tenure / lock-in strength

- Decision it informs: Do we keep winning once we've won?
- Latency: typically 1yr (range 3mo to 3yr)
- Signal: medium fidelity · high cost · automated · lagging · pull delivery · continuous cadence · trend-kind
- Owner: leadership · Stages: scale
- Measured in: % churn by tenure
- Tools: cohort churn, integration depth
- Notes: So slow it's closer to strategic radar than an operational loop.

### Pricing power

- Decision it informs: Can we raise prices without losing customers?
- Latency: typically 1yr (range 6mo to 3yr)
- Signal: medium fidelity · high cost · human-mediated · lagging · pull delivery · on-demand cadence · event-kind
- Owner: leadership · Stages: scale
- Measured in: % increase held
- Tools: price tests, renewal negotiations
- Balanced by: Revenue churn / GRR
- Actions: raise prices; add value before raising
- Notes: The ultimate lagging indicator of brand and moat combined.

### Pre-mortems

- Decision it informs: How might this fail before it does?
- Latency: typically 1d (range 1hr to 1wk)
- Signal: low fidelity · low cost · human-mediated · leading · pull delivery · on-demand cadence · event-kind
- Owner: team · Stages: pre-pmf, growth, scale
- Measured in: risks surfaced
- Tools: structured pre-mortem sessions
- Fast proxy for: Incident post-mortems
- AI can shorten this loop
- Notes: Synthetic feedback from a hypothetical future, the weird one, and cheap.

---

CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Cite as: Konrad Bloor, "Feedback Loop Field Guide" v0.23.1, https://konradbloor.com/loops