pdurlej/patchwarden

Fork 0

kill]") #11

New issue

Closed

opened 2026-05-15 19:18:57 +02:00 by claude · 1 comment

claude commented

2026-05-15 19:18:57 +02:00

Collaborator

TL;DR (PL)

Explicit kill criteria: "Jeśli do [data] zobaczymy [sygnał], to [pivot/park/kill]". NIE kierunki ("jeśli będzie źle, pivot"). NIE wishlist ("by Q4 mamy 100 users"). Dates + signals + commitment. Wave 5 sequential po #2 Product Strategy + #6 Four-Risks + #7 KPI Tree.

TL;DR (EN)

Explicit kill criteria: "If by [date] we see [signal], we [pivot/park/kill]". NOT directions ("if it gets bad, pivot"). NOT aspirational ("by Q4 we'll have 100 users"). Dates + signals + commitment. Wave 5 sequential after #2 Strategy + #6 4-Risks + #7 KPI Tree.

Scope

Kill criteria categories (mapped to Cagan 4 risks):

Value risk kill criteria:

If by 2026-09-30 (M2 + 3 months) discovery synthesis shows <30% of 7 interviews surface Patchwarden's primary functional job as "must-have" → pivot wedge or park
If by 2026-12-31 OSS install count (telemetry opt-in) <30 → reassess channel mix, possibly pivot to dev-focused wedge

Usability risk kill criteria:

If by 2026-12-31 any 3 of 5 dogfood operators (pdurlej + 4 interview followups) report Patchwarden config schema is "too engineering-heavy" → revise schema, freeze code
If false-blocker rate on dogfood repos > 10% across first 50 PRs → pause v0, redesign policy decision engine

Feasibility risk kill criteria:

If by 2026-09-30 Codex BMADX Waves 0-5 not complete in iskra-openclaw → Patchwarden v0 delayed accordingly (NOT killed, but acknowledge dependency)
If fallow-ts implementation requires >40h operator time post-Codex Wave 5 → defer fallow-ts to Y2

Viability risk kill criteria:

If by 2026-12-31 USPTO trademark search reveals collision → rename or defer launch
If by 2027-03-31 Cloud beta has 0 paying conversions from 50 OSS installs → kill Cloud (L3), maintain OSS only
If by 2027-06-30 AGPLv3 Kodus ships Forgejo adapter + policy DSL → reassess Patchwarden moat; possibly pivot to "Patchwarden as Kodus adapter"
If legal review (Q3 2026) flags policy-engine liability → restructure as advisory-only product

Cross-cutting "Cagan would weep" rules:

Every criterion names date (not "soon", not "eventually")
Every criterion names observable signal (not "if it doesn't feel right")
Every criterion names commitment (pivot / park / kill — operator's pre-registered choice, not "we'll decide then")

What this artifact does NOT cover:

Engineering deadlines (anti-roadmap)
Generic "kill switch" for production (anti-roadmap)
Burndown / velocity metrics (that's docs/roadmap.md)

Inputs (feedstock)

#6 Four-Risks Assessment (each risk → kill criterion)
#7 KPI Tree (which metrics signal failure)
#2 Product Strategy (each bet has wrongness condition → kill criterion derivation)
Cagan Inspired — outcomes-of-failure framework
Andy Grove High Output Management — pre-registered commitment

Skill draft

Anthropic Skill: operations:risk-assessment (kill-criterion mode)
Anti-prompt: "NOT generic risk mitigation. NOT 'we'll monitor X'. EVERY criterion must have date + signal + action. If criterion uses words like 'soon', 'eventually', 'if it feels wrong', REJECT. Pre-register decisions BEFORE evidence arrives — Andy Grove's commitment principle."
Prime with: Cagan outcomes-over-output + Grove pre-commitment + #6 + #7 outputs

Dependencies

Blocked by: #2 Product Strategy, #6 Four-Risks, #7 KPI Tree
Blocks: M1 milestone (without kill criteria, package fails Cagan rigor)

Cagan-grade acceptance criteria

Universal (must pass all 5):

Evidence cited — every signal threshold links to risk register (#6) or KPI tree (#7) baseline
Operator-readable <10 min — 4-category list + cross-cutting rules ≤2000w
Names the wrongness condition — meta: "these kill criteria are wrong if all of them are triggered AND we still feel committed to Patchwarden" (cognitive-bias check)
Single concrete decision — moves docs/decisions.md by adding D17: "kill criteria pre-committed; revisit only at M3 quarterly refresh"
Cousin-family attribution — credits skill, #6, #7, Grove pre-commitment framework

Tier-0 anchor extras:

3-year horizon explicit — criteria span 2026-2029
Leadership change test — successor PM can execute kill decisions without operator input

"Cagan would weep" stretch test:

Names dates + signals, NOT directions — every criterion: explicit date + observable signal + pre-committed action. No "we'll see".
If ANY criterion is "we'll decide based on how we feel" → FAIL.

Operator decision points

This artifact gates on:

OD10 — Kill criteria honesty level (park-only / pivot-only / both/full kill option) — default: full kill option (most rigorous)

Status flow

status:cagan-grade-review-pending → operator merges → close → tick epic Wave 5 #10

Parent epic: #1 (this is artifact 10 of 11)
Plan: /Users/pd/.claude/plans/deep-pondering-popcorn.md § Wave 5

Drafted by Prof Kong (claude Opus 4.7) on behalf of pdurlej, 2026-05-15.

## TL;DR (PL) Explicit kill criteria: *"Jeśli do [data] zobaczymy [sygnał], to [pivot/park/kill]"*. **NIE kierunki** ("jeśli będzie źle, pivot"). **NIE wishlist** ("by Q4 mamy 100 users"). Dates + signals + commitment. Wave 5 sequential po #2 Product Strategy + #6 Four-Risks + #7 KPI Tree. ## TL;DR (EN) Explicit kill criteria: "If by [date] we see [signal], we [pivot/park/kill]". **NOT directions** ("if it gets bad, pivot"). **NOT aspirational** ("by Q4 we'll have 100 users"). Dates + signals + commitment. Wave 5 sequential after #2 Strategy + #6 4-Risks + #7 KPI Tree. ## Scope **Kill criteria categories** (mapped to Cagan 4 risks): **Value risk kill criteria**: - **If by 2026-09-30 (M2 + 3 months) discovery synthesis shows <30% of 7 interviews surface Patchwarden's primary functional job as "must-have"** → pivot wedge or park - **If by 2026-12-31 OSS install count (telemetry opt-in) <30** → reassess channel mix, possibly pivot to dev-focused wedge **Usability risk kill criteria**: - **If by 2026-12-31 any 3 of 5 dogfood operators (pdurlej + 4 interview followups) report Patchwarden config schema is "too engineering-heavy"** → revise schema, freeze code - **If false-blocker rate on dogfood repos > 10% across first 50 PRs** → pause v0, redesign policy decision engine **Feasibility risk kill criteria**: - **If by 2026-09-30 Codex BMADX Waves 0-5 not complete in iskra-openclaw** → Patchwarden v0 delayed accordingly (NOT killed, but acknowledge dependency) - **If fallow-ts implementation requires >40h operator time post-Codex Wave 5** → defer fallow-ts to Y2 **Viability risk kill criteria**: - **If by 2026-12-31 USPTO trademark search reveals collision** → rename or defer launch - **If by 2027-03-31 Cloud beta has 0 paying conversions from 50 OSS installs** → kill Cloud (L3), maintain OSS only - **If by 2027-06-30 AGPLv3 Kodus ships Forgejo adapter + policy DSL** → reassess Patchwarden moat; possibly pivot to "Patchwarden as Kodus adapter" - **If legal review (Q3 2026) flags policy-engine liability** → restructure as advisory-only product **Cross-cutting "Cagan would weep" rules**: - Every criterion names **date** (not "soon", not "eventually") - Every criterion names **observable signal** (not "if it doesn't feel right") - Every criterion names **commitment** (pivot / park / kill — operator's pre-registered choice, not "we'll decide then") **What this artifact does NOT cover**: - Engineering deadlines (anti-roadmap) - Generic "kill switch" for production (anti-roadmap) - Burndown / velocity metrics (that's `docs/roadmap.md`) ## Inputs (feedstock) - #6 Four-Risks Assessment (each risk → kill criterion) - #7 KPI Tree (which metrics signal failure) - #2 Product Strategy (each bet has wrongness condition → kill criterion derivation) - Cagan *Inspired* — outcomes-of-failure framework - Andy Grove *High Output Management* — pre-registered commitment ## Skill draft - **Anthropic Skill**: `operations:risk-assessment` (kill-criterion mode) - **Anti-prompt**: *"NOT generic risk mitigation. NOT 'we'll monitor X'. EVERY criterion must have date + signal + action. If criterion uses words like 'soon', 'eventually', 'if it feels wrong', REJECT. Pre-register decisions BEFORE evidence arrives — Andy Grove's commitment principle."* - **Prime with**: Cagan outcomes-over-output + Grove pre-commitment + #6 + #7 outputs ## Dependencies Blocked by: #2 Product Strategy, #6 Four-Risks, #7 KPI Tree Blocks: M1 milestone (without kill criteria, package fails Cagan rigor) ## Cagan-grade acceptance criteria **Universal (must pass all 5)**: 1. **Evidence cited** — every signal threshold links to risk register (#6) or KPI tree (#7) baseline 2. **Operator-readable <10 min** — 4-category list + cross-cutting rules ≤2000w 3. **Names the wrongness condition** — meta: "these kill criteria are wrong if all of them are triggered AND we still feel committed to Patchwarden" (cognitive-bias check) 4. **Single concrete decision** — moves `docs/decisions.md` by adding D17: "kill criteria pre-committed; revisit only at M3 quarterly refresh" 5. **Cousin-family attribution** — credits skill, #6, #7, Grove pre-commitment framework **Tier-0 anchor extras**: - **3-year horizon explicit** — criteria span 2026-2029 - **Leadership change test** — successor PM can execute kill decisions without operator input **"Cagan would weep" stretch test**: - **Names dates + signals, NOT directions** — every criterion: explicit date + observable signal + pre-committed action. No "we'll see". - If ANY criterion is "we'll decide based on how we feel" → FAIL. ## Operator decision points This artifact gates on: - **OD10** — Kill criteria honesty level (park-only / pivot-only / both/full kill option) — default: full kill option (most rigorous) ## Status flow `status:cagan-grade-review-pending` → operator merges → close → tick epic Wave 5 #10 --- Parent epic: #1 (this is artifact 10 of 11) Plan: `/Users/pd/.claude/plans/deep-pondering-popcorn.md` § Wave 5 *Drafted by Prof Kong (claude Opus 4.7) on behalf of pdurlej, 2026-05-15.*

claude added this to the M1: Cagan Package v1.0 milestone

2026-05-15 19:18:57 +02:00

claude added the

kind:artifact

area:metrics

status:cagan-grade-review-pending

wave:5-operating

tier:0-anchor

labels

2026-05-15 19:18:57 +02:00

claude referenced this issue

2026-05-15 19:19:30 +02:00

#11 Empowered Team Operating Principles (solo-operator variant, cousin-family delegation) #12

claude referenced this issue

2026-05-15 19:21:36 +02:00

Epic: Cagan-grade product strategy package (11 artifacts × 5 waves) #1

claude referenced this issue from a commit

2026-05-15 21:11:49 +02:00

docs/product: Cagan-grade product strategy package v1.0 (11 artifacts batch)

claude referenced this issue

2026-05-15 21:12:42 +02:00

Epic: Cagan-grade product strategy package (11 artifacts × 5 waves) #1

claude referenced this issue

2026-05-15 21:37:34 +02:00

#1 Product Vision Narrative (3-year motivational prose) #2

claude referenced this issue

2026-05-15 22:03:45 +02:00

#12 Strategic Thesis — defense-in-depth + Goldratt constraint-redirect + emergent selection pressure #13

claude referenced this issue

2026-05-15 22:04:31 +02:00

#6 Four-Risks Assessment (Cagan: value / usability / feasibility / viability) #6

claude commented

2026-05-26 12:45:39 +02:00

Author

Collaborator

Closed via operator voice ack 2026-05-26.

Operator highlights:

13 triggers wszystkie ack
Threshold precision — relax: "triggery są good enough, dla mnie to jest mało ważne w tym momencie, gadam jak dużo to gadania i moje głupoty"
F1 trigger update: claude (Patchwarden dedicated thread) przejmuje codex executor role dla Patchwarden ship — "ty teraz ciśniesz, mam nadzieję że ty teraz tutaj pociśniesz"
B3 (legal liability → advisory-only recommend_* escape hatch) — ack
D17 (only-quarterly review, no panic editing) — ack
Forgejo functionality dla pisania milestone progress — "może wiesz będziemy używać funkcjonalności na foregio żeby sobie pisać o tym jak rzeczy nam iżdżą"
Premortem insight (largest personal kill trigger): "najgorsze by było jeśli bym nie skończył i bym nadal marnował czas i ładował coś co nie działa i nie pomaga mi — to były chyba największe kryteria"

Modifications queued in claude/patchwarden-tactical-updates PR:

Premortem insight dodać jako meta-check § "operator's most-binding personal trigger": wasted time on non-working, non-helpful output
F1 trigger update: claude→Patchwarden executor role reflected

D17 (kill criteria pre-committed, quarterly review only, no panic editing) → claude/patchwarden-decisions-batch PR.

Status: cagan-grade-approved. — claude (Patchwarden dedicated thread)

**Closed via operator voice ack 2026-05-26.** Operator highlights: - 13 triggers wszystkie ack - Threshold precision — relax: *"triggery są good enough, dla mnie to jest mało ważne w tym momencie, gadam jak dużo to gadania i moje głupoty"* - **F1 trigger update**: claude (Patchwarden dedicated thread) przejmuje codex executor role dla Patchwarden ship — *"ty teraz ciśniesz, mam nadzieję że ty teraz tutaj pociśniesz"* - B3 (legal liability → advisory-only `recommend_*` escape hatch) — ack - D17 (only-quarterly review, no panic editing) — ack - Forgejo functionality dla pisania milestone progress — *"może wiesz będziemy używać funkcjonalności na foregio żeby sobie pisać o tym jak rzeczy nam iżdżą"* - **Premortem insight** (largest personal kill trigger): *"najgorsze by było jeśli bym nie skończył i bym nadal marnował czas i ładował coś co nie działa i nie pomaga mi — to były chyba największe kryteria"* **Modifications queued** in `claude/patchwarden-tactical-updates` PR: - Premortem insight dodać jako meta-check § "operator's most-binding personal trigger": **wasted time on non-working, non-helpful output** - F1 trigger update: claude→Patchwarden executor role reflected **D17** (kill criteria pre-committed, quarterly review only, no panic editing) → `claude/patchwarden-decisions-batch` PR. Status: `cagan-grade-approved`. — claude (Patchwarden dedicated thread)

claude closed this issue

2026-05-26 12:46:10 +02:00

claude referenced this issue

2026-05-26 12:53:29 +02:00

docs(v0): record 11 M1 close decisions D8/D8b/D9/D10/D11/D12-mod/D15/D16/D17/D18/D19 #40