docs(v0): decisions D21 — M2 gate amendment for dogfood wiring #57

Merged

pdurlej merged 1 commit from claude/patchwarden-decisions-d21-m2-amendment into main

2026-05-27 15:18:25 +02:00

claude commented

2026-05-27 15:17:52 +02:00

Collaborator

What

Records operator decision (2026-05-27) że M2 gate jest amended: defense hardening + end-to-end wiring of existing capabilities are permitted; new core features remain parked.

This is PR D1 of 4 in the Wizji-Wartości Wave. D21 must merge first to legalize PRs D2-D4 of this wave.

Why this matters (operator-side discovery this session)

M2 gate (D16 + M2 milestone) originally read as "park all v0 code work until M2 closes 5-7 discovery interviews." Operator's mental-model conversation today revealed that reading blocks legitimate value delivery:

Real situation	Strict M2 reading
Swiss Cheese hardening (#54+#55+#56) closes existing gaps	"code change → blocked"
Wire Ollama into platform workflow (capability already in `main` from #49)	"any wiring is feature → blocked"
Document actual-vs-aspirational state	"any change is change → blocked"

Per Cagan kill criteria: when a gate stops delivering value, kill the gate, not the value.

What D21 changes

✅ Permitted under amended M2

Defense hardening of existing v0 (Swiss Cheese fixes, lint, soft-fail handling)
End-to-end wiring of existing CLI capabilities into platform workflow
Documentation of actual-vs-aspirational state
ADRs/decisions records

❌ Still parked under M2 gate

New core product features (webhook listener, audit storage, multi-tenant Cloud)
Pyfallow/fallow-ts runtime dep in Patchwarden (plan doc permitted; binding the dep is not)
Expansion of dogfood lane beyond safe_docs_status classification
Schema-version bumps on spec/schemas/*.schema.json

Hard manual classes (force operator decision regardless of amendment)

New runtime dep in pyproject.toml → parked
New CLI subcommand with fundamentally new behavior (e.g. serve for webhook) → parked
Schema-version bump → parked
New policy class in policies/*.toml schema → parked

Mitigation against amendment-drift

Hard manual classes codify exactly what stays parked.
M3 quarterly review (2026-10-15) re-evaluates amendment. If wiring PRs grew into capability creep, amendment rolls back.
D20 operator merge gate still binding: amendment changes what's permitted to propose, not what merges unilaterally.

Test impact

None — docs-only. 152/152 tests still green.

Atomic per ADR-0017

1 file edited (docs/decisions.md), +39 lines, 0 deletions.
0 src changes, 0 test changes.
base=main, no stacking.

Wizji-Wartości Wave sequence

PR	What	Gate effect
D1 (this PR)	D21 amendment	legalizes D2-D4
D2	`docs/operations/dogfood-actual-vs-mental-model.md`	docs
D3	`pdurlej/platform` workflow wiring (Ollama + post-findings)	wiring of existing capability
D4	`docs/operations/pyfallow-integration-plan.md` (PLAN, not code)	architectural decision doc

Merge order is sequential (D1 first to legalize the rest).

## What Records operator decision (2026-05-27) że M2 gate jest **amended**: defense hardening + end-to-end wiring of existing capabilities are **permitted**; new core features remain **parked**. This is **PR D1 of 4** in the Wizji-Wartości Wave. **D21 must merge first** to legalize PRs D2-D4 of this wave. ## Why this matters (operator-side discovery this session) M2 gate (D16 + M2 milestone) originally read as "park all v0 code work until M2 closes 5-7 discovery interviews." Operator's mental-model conversation today revealed that reading **blocks legitimate value delivery**: | Real situation | Strict M2 reading | |---|---| | Swiss Cheese hardening (#54+#55+#56) closes existing gaps | "code change → blocked" | | Wire Ollama into platform workflow (capability already in `main` from #49) | "any wiring is feature → blocked" | | Document actual-vs-aspirational state | "any change is change → blocked" | Per **Cagan kill criteria**: when a gate stops delivering value, kill the gate, not the value. ## What D21 changes ### ✅ Permitted under amended M2 - Defense hardening of existing v0 (Swiss Cheese fixes, lint, soft-fail handling) - **End-to-end wiring** of existing CLI capabilities into platform workflow - Documentation of actual-vs-aspirational state - ADRs/decisions records ### ❌ Still parked under M2 gate - New core product features (webhook listener, audit storage, multi-tenant Cloud) - Pyfallow/fallow-ts **runtime dep** in Patchwarden (plan doc permitted; binding the dep is not) - Expansion of dogfood lane beyond `safe_docs_status` classification - Schema-version bumps on `spec/schemas/*.schema.json` ### Hard manual classes (force operator decision regardless of amendment) - New runtime dep in `pyproject.toml` → parked - New CLI subcommand with fundamentally new behavior (e.g. `serve` for webhook) → parked - Schema-version bump → parked - New policy class in `policies/*.toml` schema → parked ## Mitigation against amendment-drift 1. Hard manual classes codify exactly what stays parked. 2. M3 quarterly review (2026-10-15) re-evaluates amendment. If wiring PRs grew into capability creep, amendment rolls back. 3. D20 operator merge gate still binding: amendment changes *what's permitted to propose*, not *what merges unilaterally*. ## Test impact **None — docs-only.** 152/152 tests still green. ## Atomic per ADR-0017 - 1 file edited (`docs/decisions.md`), +39 lines, 0 deletions. - 0 src changes, 0 test changes. - `base=main`, no stacking. ## Wizji-Wartości Wave sequence | PR | What | Gate effect | |---|---|---| | **D1 (this PR)** | D21 amendment | **legalizes D2-D4** | | D2 | `docs/operations/dogfood-actual-vs-mental-model.md` | docs | | D3 | `pdurlej/platform` workflow wiring (Ollama + post-findings) | wiring of existing capability | | D4 | `docs/operations/pyfallow-integration-plan.md` (PLAN, not code) | architectural decision doc | Merge order is sequential (D1 first to legalize the rest).

claude added 1 commit

2026-05-27 15:17:52 +02:00

docs(v0): decisions D21 — M2 gate amendment for dogfood wiring 5afc0c2abb

Records operator decision (2026-05-27) that M2 gate is amended:
defense hardening + end-to-end wiring of existing capabilities are
permitted; new core features remain parked.

## Why

M2 gate (D16 + M2 milestone) originally read as "park all v0 code work
until M2 closes 5-7 discovery interviews." Operator-side discovery this
session showed that reading blocks legitimate value delivery:

1. Defense hardening (Swiss Cheese PRs #54 + #55 + #56) doesn't expand
   scope — it closes real gaps in existing v0.
2. Workflow wiring on pdurlej/platform doesn't add new Patchwarden
   capability — it activates capability already in main (Ollama via
   review-run, post-findings --execute).
3. Operator's mental model assumed end-to-end flow that doesn't exist
   in current dogfood loop; closing those gaps is "wire what we have",
   not "build what we don't have".

Cagan kill criteria: when a gate stops delivering value, kill the gate,
not the value. M2 amendment formalizes the pattern already validated by
the Swiss Cheese wave.

## What changes (D21 body)

- Permitted under amended M2: defense hardening; end-to-end wiring of
  existing capabilities into platform workflow; documentation; ADRs.
- Still parked: new core features (webhook, audit storage, Cloud);
  pyfallow runtime dep binding; expansion of dogfood lane beyond
  safe_docs_status classification; schema-version bumps.

Hard manual classes codify exactly what stays parked (new runtime dep,
new schema, new policy class, new fundamentally-different subcommand).

## Mitigation against amendment-drift

- Hard manual classes pause anything that touches parked surface.
- M3 quarterly review (2026-10-15) re-evaluates whether wiring PRs
  stayed wiring or grew into capability creep. If creep, amendment is
  rolled back.
- D20 operator merge gate still binding: amendment changes what's
  permitted to propose, not what can be merged unilaterally.

## Test impact

None — docs-only. 152/152 tests still green.

## Atomic per ADR-0017

- 1 file edited, +49 lines, 0 deletions.
- 0 src changes, 0 test changes.
- base=main, no stacking.

## Wave context

This is PR D1 of 4 in the Wizji-Wartości Wave:
- D1 (this PR): D21 amendment — LEGALIZES the wave
- D2: docs/operations/dogfood-actual-vs-mental-model.md
- D3: platform workflow wiring (Ollama + post-findings)
- D4: docs/operations/pyfallow-integration-plan.md (PLAN, not code)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>