Field Reports

Case studies from teams dogfooding the parts.

Each report explains what failed, what evidence caught it, what changed, and which human decision remained outside the joke.

Field report

The TTL Treat That Expired on Purpose

Incident flavor line: The bowl went stale while the meeting decided whether the meeting was needed.

Problem: A Time Bowl sat open during refinement until the original intent was no longer reliable.
What failed: The team expected cart state to be permanent even though the decision context had expired.
What evidence caught it: TTL countdown, expiration event, and expired-cart recovery message.
What changed: The receipt now names expiry and asks the user to re-add only relevant snacks.
Human decision required: A human product owner confirms whether the stale bowl represents real demand.
What this proves: TTL behavior is observable and recoverable.
What this does not prove: It does not prove the feature should still ship.
Before/after artifact: Before: stale cart. After: expired-bowl recovery receipt.
Related product: TTL Timeout Treats
Related lesson: Fail-Forward Dogfood Loops
Related glossary terms: TTL, fail-forward state, dogfood record
Related blueprint: Cart TTL test script

Team exercise: Turn the failure into one observable test before widening rollout.

Review prompt: What evidence proves the fix, what does not prove it, and who owns the next human decision?

Last updated UTC: 2026-06-11T22:21:35Z

Field report

Incident flavor line: Semantic drift learned to wear a second collar.

Problem: Autonomy-washing and bounded creative arm appeared twice with slightly different meanings.
What failed: Seed data allowed duplicate concepts to create inconsistent teaching paths.
What evidence caught it: Canonical glossary registry, duplicate detection, and alias metadata.
What changed: Duplicates merge into canonical entries with aliases and backreferences.
Human decision required: A human reviewer chooses the canonical wording.
What this proves: The glossary can preserve meaning without multiplying terms.
What this does not prove: It does not prove the glossary is complete.
Before/after artifact: Before: repeated term cards. After: one canonical card with also-known-as chips.
Related product: Glossary Drift Biscuits
Related lesson: Evidence Dashboard Literacy
Related glossary terms: Semantic drift, alias, evidence trace
Related blueprint: Glossary drift review

Team exercise: Turn the failure into one observable test before widening rollout.

Review prompt: What evidence proves the fix, what does not prove it, and who owns the next human decision?

Last updated UTC: 2026-06-11T22:21:35Z

Field report

Incident flavor line: The claim boundary bit through the packaging.

Problem: A draft phrase sounded like BarkBowl certified ecosystem safety.
What failed: Copy implied more authority than static evidence could support.
What evidence caught it: Autonomy-washing red-team scan and claim-boundary checklist.
What changed: The phrase was downgraded to a bounded, inspectable claim.
Human decision required: Human reviewer confirms final public wording.
What this proves: Unsafe capability language can be caught before release.
What this does not prove: It does not certify the wider ecosystem.
Before/after artifact: Before: certification-like phrase. After: public-safe evidence statement.
Related product: Claim Boundary Dental Stick
Related lesson: Autonomy-Washing: Claims That Need a Muzzle
Related glossary terms: Claim boundary, autonomy-washing, bounded gloss
Related blueprint: Claim boundary rewrite

Team exercise: Turn the failure into one observable test before widening rollout.

Review prompt: What evidence proves the fix, what does not prove it, and who owns the next human decision?

Last updated UTC: 2026-06-11T22:21:35Z

Field report

Incident flavor line: The release bag said fresh; the kennel label said old.

Problem: The ZIP filename, WordPress header, public diagnostics, and docs did not match.
What failed: Version parity checks were not treated as product behavior.
What evidence caught it: Version parity audit, SHA-256 checksums, public settings endpoint, and Evaluation Lab.
What changed: Active release surfaces now move together per minor release.
Human decision required: Human deployment check confirms WordPress Admin state.
What this proves: Source package labels are internally consistent.
What this does not prove: It does not prove production was updated.
Before/after artifact: Before: split labels. After: one release version across headers, docs, exports, and evidence.
Related product: Version Parity Kibble
Related lesson: Dogfooding Needs Rings
Related glossary terms: Version parity, release package parity, static ledger
Related blueprint: Version parity release note

Team exercise: Turn the failure into one observable test before widening rollout.

Review prompt: What evidence proves the fix, what does not prove it, and who owns the next human decision?

Last updated UTC: 2026-06-11T22:21:35Z

Field report

Incident flavor line: The Checkup barked, then learned to leave a note.

Problem: A team got a useful diagnosis but had no artifact to discuss.
What failed: The learning loop ended before it became a reviewable handoff.
What evidence caught it: Guided quiz result, copied decision record, recommendations, and local-only boundary text.
What changed: Every diagnosis now includes why it happened, estimates, and copyable next steps.
Human decision required: Human team decides which recommendation becomes work.
What this proves: The quiz can route learning into a reviewable artifact.
What this does not prove: It does not approve the action.
Before/after artifact: Before: diagnosis only. After: draft decision record with human-review boundary.
Related product: Prompt Loop Chow
Related lesson: Guided Quiz Diagnosis
Related glossary terms: Quiz diagnosis, decision matrix, evidence trace
Related blueprint: 2-minute diagnosis record

Team exercise: Turn the failure into one observable test before widening rollout.

Review prompt: What evidence proves the fix, what does not prove it, and who owns the next human decision?

Last updated UTC: 2026-06-11T22:21:35Z

Field report

Incident flavor line: The mature dog declined the chew.

Problem: A proposed change had low evidence, high blast radius, and no owner.
What failed: The team assumed every finding needed mutation.
What evidence caught it: No-op decision record and resource-closure checklist.
What changed: No-op became a first-class structural action with a receipt.
Human decision required: Human owner decides when evidence justifies reopening the issue.
What this proves: Change restraint can be documented.
What this does not prove: It does not mean never change.
Before/after artifact: Before: risky action. After: explicit no-op receipt.
Related product: No-Op Senior Blend
Related lesson: No-Op Is a Feature
Related glossary terms: No-op dominance, structural action, viability retention
Related blueprint: No-op decision record

Team exercise: Turn the failure into one observable test before widening rollout.

Review prompt: What evidence proves the fix, what does not prove it, and who owns the next human decision?

Last updated UTC: 2026-06-11T22:21:35Z