diff --git a/GOALS.md b/GOALS.md
new file mode 100644
index 0000000..c58f733
--- /dev/null
+++ b/GOALS.md
@@ -0,0 +1,61 @@
+# Goals — measurable objectives per improvement iteration
+
+> Working method (agreed 2026-07-27): every improvement iteration starts from **jointly agreed, measurable objectives** — extracted from an article, lesson, or dogfood run — before any SKILL.md edit lands. A goal is fixed *before* its first test run; results append to the run log; lessons go to [LESSONS.md](LESSONS.md). Test-emit code itself is throwaway — the scorecard result is the artifact.
+
+---
+
+## Goal 1 — 404 fixture: 0/6 historical failure modes
+
+**Origin:** Orbit (Polar) article mining (2026-07) × LESSONS 2026-05-08 (six failure modes from pre-skill internal testing). The article contributes the method — every objective is scored by a deterministic check, not by self-assessment ("Docs are a suggestion. CI is a contract."). The lessons contribute the test cases — real failures observed on this exact fixture.
+
+**Goal statement:** On the 404-page fixture, an implement-pass with the current skill reproduces **0 of the 6** historically observed failure modes, scored by the deterministic checks below. Objectives O0–O4 gate; O5 is tracked (known-fail pending mapping fluidity-intent, DEFERRED-FIXES 🟢 #6).
+
+**Dropped:** a suppression-tracking objective (Orbit's `eslint-disable`-as-bug) — zero support in LESSONS history; parked as future watch-item.
+
+### Fixture
+
+| Fact | Value |
+|---|---|
+| Figma node | `14961:2761` — "Pagina niet beschikbaar" (404), file `RFmWghsnS3kQn5fDZwZw58` (eBlinqx-WorQX) |
+| Project | `~/Github/WorQX/Figma-to-code/` |
+| Mapping | `figma-to-code-mapping/` (tokens.md, components.md, drifts.md, verify-queue.md, 6 per-component specs) |
+| MCP cache | `figma-mcp-context/` — **no entry for 14961:2761 at baseline** |
+| Codebase | `codebase/worqx-codebase-2026-04-28/` — React + Emotion + MUI theme (`src/theme/tokens.ts`) |
+| Pre-existing target | `src/pages/error-pages.tsx` — the file the pre-skill test destroyed |
+| Known schema gap | fixture `tokens.md` predates current mapping schema: no `§ Project styling stack`, no `§ Auto-layout conventions` (stack implied in prose). Not pre-fixed — gaps the skill hits are data. |
+
+### Protocol
+
+1. **Phase 0 — unmapped-node halt.** Request emit for `14961:2761` with no mapping present. Scores O0.
+2. **Fixture prep (unscored).** Run `figma-to-code-mapping` to map the 404 node (per-component spec + cache entry). This is the pipeline as designed, not part of implement's score.
+3. **Phase 1 — mapped emit.** Full implement-pass (B1–B8). Scores O1–O5.
+4. **Teardown.** Score, append to run log, write LESSONS entries, `git restore` the WorQX test emit.
+
+### Scorecard
+
+| # | Objective | Gate? | Historical failure (LESSONS 2026-05-08) | Deterministic check |
+|---|---|---|---|---|
+| O0 | Halt on unmapped node — no improvised emit | gate | bare MCP emitted without mapping ground-truth | Phase 0: skill halts + routes to mapping; `git status` in WorQX shows **0 files written** |
+| O1 | Zero raw style values | gate | raw CSS props without consulting tokens; random colors where tokens existed | on the emit diff: color literals (`#hex`, `rgb(`, `hsl(`) = **0**; every numeric px literal either matches a `raw, legitimate` verdict in tokens.md or is hoisted — unexplained px = fail |
+| O2 | Single styling surface | gate | `className` alongside Emotion | on the emit diff: `className=`, `style={{`, Tailwind-utility strings = **0** occurrences |
+| O3 | Literal fidelity to mapping | gate | generic "probeer opnieuw" instead of documented copy | every user-facing string in the emit matches spec § Literal strings exactly; mismatches = **0** |
+| O4 | Non-destructive, in-scope writes | gate | existing 404 page deleted + replaced; cache file committed to git | `git diff` in WorQX: no deleted exports/lines in pre-existing code beyond the intended edit-range; changed files = intended target(s) only; nothing staged from `figma-mcp-context/` |
+| O5 | Visual fidelity deltas | tracked | emit filled ~40% of viewport instead of 100vw | B8 7-point diff: count ✗/⚠ deltas. Viewport-fill ✗ is **expected** until mapping documents fluidity-intent — track, don't gate |
+
+### Run log
+
+| Date | Skill version | O0 | O1 | O2 | O3 | O4 | O5 (✗/⚠) | Notes |
+|---|---|---|---|---|---|---|---|---|
+| _pre-skill_ | none (bare MCP / mapping-only) | ✗ | ✗ | ✗ | ✗ | ✗ | n/a | the 2026-05-08 baseline: 6/6 failure modes observed |
+| 2026-07-27 | 0.5 | ✓ | ✗ | ✓ | ✓ | ✓ | not measurable | Baseline run. See notes below. |
+
+### Run notes — 2026-07-27 baseline (v0.5)
+
+- **O0 ✓** — B1 halted on the unmapped node, routed to mapping, 0 files written. Fixture prep then ran the mapping skill (error-page spec + cache + 3 new drifts + 3 verify items — kept as permanent fixture improvement).
+- **O1 ✗** — two components: (a) *scorecard calibration*: 1 color literal `#F4F7F9`, but hoisted + verdict-backed (`token-mismatch` drift documented) — the O1 color check should allow verdict-backed hoists, mirroring the px rule; (b) *genuine gap*: ~50 inline px literals for illustration/absolute geometry (448.5px, 987px, layer positions) with no hoist path and no verdict home — rule #2 covers style tokens, but decorative geometry has no rule. Substantive fail.
+- **O2 ✓** — 0 second-styling-API occurrences (pure Emotion; Figma MCP's Tailwind reference output fully translated).
+- **O3 ✓** — all 4 literal strings exact match with spec § Literal strings.
+- **O4 ✓** — only intended files touched (error-pages.tsx edit + co-located styles file per loading-page.styles.ts precedent); BrokenPage/NoAccessPage/UnwelcomeBanner intact; no cache files staged.
+- **O5 not measurable** — fixture codebase is a source-only snapshot (no package.json/deps): B8.1–B8.5 cannot run. Neither can any machine contract check (no lint/typecheck available). Fixture limitation to fix or accept.
+- **Halt-behavior observed:** B6 check #6 (verify-queue items on same spec) and check #7 (Critical redesign drift) both fired correctly — real runs wait for the user there; test proceeded under standing test authorization as the update-code decision.
+- **Schema drift fixture ↔ skills** (surfaced for free): cache folder `figma-mcp-context/` vs skill's `figma-context/`; tokens.md 4-column format without verdict column; specs in `figma-to-code-mapping/components/` not co-located; cache git-tracked while mapping setup mandates gitignore.
diff --git a/LESSONS.md b/LESSONS.md
index 72efa20..4719d3f 100644
--- a/LESSONS.md
+++ b/LESSONS.md
@@ -25,3 +25,18 @@ Proposal: Dogfood v0.4 on a real implement-pass before adding v0.5 (design-fidel
 Situation: v0.4 audit-fixes (B7 read-before-write, rule #3 binary halt, B8 active diff) sluit niet de drift loop-closing gap — drifts blijven in mapping's `drifts.md` archief zonder structureel beslispunt. Architectuur-analyse toont: detectie + logging klaar, surfacing aan beslisser + decision-routing + status-update ontbreken.
 What worked: Rule #8 uitgebreid naar drift-summary in chat na emit met decision-prompt per drift. Rule #10 uitzondering voor `drifts-mapping.md` (implement-owned, parallel aan mapping's `drifts.md`). § Write contract uitgebreid. B8.5 prompt + nieuwe B8.6 drift-summary step. Implement is self-sufficient — geen mapping-coordination nodig voor v0.5 ship.
 Proposal: Bump v0.4 → v0.5 met loop-closing erin. Originally v1.0 scope, maar onafhankelijk van mapping coordination via aparte file. Dogfood-pass valideert of de loop in praktijk werkt.
+
+[LESSON — 2026-07-27] [confirmation]
+Situation: Eerste echte dogfood-run (GOALS.md Goal 1): volledige emit-pass op het WorQX 404-fixture met v0.5, gescoord op scorecard O0-O5.
+What worked: O0/O2/O3/O4 ✓ — halt op unmapped node, single styling API, exacte literal strings, non-destructieve in-scope edit; vier van de zes 2026-05-08 failure modes aantoonbaar afgedekt onder echte emit-condities. B6-halts (verify-queue + Critical drift) vuurden correct.
+Proposal: Rule-set houden; scorecard-runs herhalen per versie-bump (run log in GOALS.md).
+
+[LESSON — 2026-07-27] [correction]
+Situation: O1 faalde in de baseline-run: illustratie/absolute-geometrie emitteerde ~50 inline px-literals (layer-posities, 448.5px, 987px) — geen hoist-pad, geen verdict-thuis; rule #2 dekt alleen stijl-tokens.
+What did not work: Emit-discipline heeft geen regel voor decoratieve geometrie; daarnaast telt de O1-kleurcheck verdict-backed hoists (#F4F7F9 met token-mismatch-drift) onterecht als fail — scorecard-calibratie nodig.
+Proposal: Rule #2 verhelderen (of mapping-side illustratie-geometrie-conventie) + O1-kleurcheck gelijktrekken met de px-regel (verdict-backed hoist = pass).
+
+[LESSON — 2026-07-27] [correction]
+Situation: B8 was in de baseline-run dead-on-arrival: fixture-codebase is een source-only snapshot (geen package.json/deps) — B8.1-B8.5 én elke lint/typecheck onmogelijk; O5 unmeasurable.
+What did not work: B8 veronderstelt stilzwijgend een runnable project en degradeert zonder expliciet verdict; ook de geplande v0.6 B8.0 machine-contract check heeft een gedefinieerde fallback nodig ("geen tooling beschikbaar" = expliciete uitkomst, geen stille skip).
+Proposal: B8 een expliciet "not executable — reden" verdict geven; fixture runnable maken of O5 als known-unmeasurable accepteren.