Make target-contributed-psl-blocks an explicit step in the running order

TML-2537 was only referenced in passing inside the other steps; it is
Will-owned, in flight, and on the critical path (gates postgres-rls's
PSL policy {} surface), so it now reads as its own numbered step.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: Will Madden <madden@prisma.io>

Author: wmadden

projects/supabase-integration/README.md

@@ -86,12 +86,13 @@ This separation also answers "how do we test RLS before the runtime exists": **p
 The foundation + `control-policy` are done, so the remaining work is highly parallel. Capacity: Will runs ~2 concurrent lanes alongside [target-contributed-psl-blocks](../target-contributed-psl-blocks/spec.md) (TML-2537) in flight; Serhii owns `explicit-namespace-dsl`.
 
 1. **Lane 1 — `extension-supabase` M1 + walking skeleton.** Stands up `examples/supabase` on the stock Postgres runtime and authors `bootstrapSupabaseShim`. Unblocked by the done foundation + control-policy alone.
-2. **Lane 2 — `cross-contract-refs`.** Chosen *before* `postgres-rls` despite RLS being the longer/riskier pole: within a single serial lane, order doesn't change total wall-clock, and cross-contract-refs is **fully unblocked end-to-end** whereas RLS's must-have PSL surface is gated on TML-2537. Doing cross-contract-refs first lets TML-2537 finish in parallel, so `postgres-rls` can then run TS + PSL in one clean pass instead of stalling mid-project. It also lands the contract-aggregate/brand machinery the rest leans on, and delivers the headline `Profile → auth.User` FK early — validating the skeleton with something real before wading into the riskier RLS work.
-3. **Then — `postgres-rls`** (TS + PSL in one pass) the moment TML-2537 lands / a lane frees. **`runtime-target-layer`** is a cheap interleave (~50–100 LOC core) slotted whenever convenient; it's the substrate `SupabaseRuntime` extends in `extension-supabase` M2.
-4. **In parallel throughout — `explicit-namespace-dsl` (Serhii).** The launch blocker; depends only on the landed TML-2605.
-5. **Integration — `extension-supabase` M2→M4.** Role binding (`asUser`/`asAnon`/`asServiceRole`), the live-query RLS e2e, real-Supabase acceptance, close-out.
+2. **Lane 2 — `cross-contract-refs`.** Chosen *before* `postgres-rls` despite RLS being the longer/riskier pole: within a single serial lane, order doesn't change total wall-clock, and cross-contract-refs is **fully unblocked end-to-end** whereas RLS's must-have PSL surface is gated on TML-2537 (step 3). Doing cross-contract-refs first lets TML-2537 finish in parallel, so `postgres-rls` can then run TS + PSL in one clean pass instead of stalling mid-project. It also lands the contract-aggregate/brand machinery the rest leans on, and delivers the headline `Profile → auth.User` FK early — validating the skeleton with something real before wading into the riskier RLS work.
+3. **In flight (Will) — [`target-contributed-psl-blocks`](../target-contributed-psl-blocks/spec.md) (TML-2537).** Not a constituent but on the critical path: it's the PSL-block extensibility substrate that lets the Postgres pack contribute the `policy {}` keyword, so it **gates `postgres-rls`'s PSL surface** (PSL authoring is must-have for launch). Landing it is the pivot that lets step 4 run as a single TS + PSL lane rather than stalling.
+4. **Then — `postgres-rls`** (TS + PSL in one pass) the moment TML-2537 (step 3) lands / a lane frees. **`runtime-target-layer`** is a cheap interleave (~50–100 LOC core) slotted whenever convenient; it's the substrate `SupabaseRuntime` extends in `extension-supabase` M2.
+5. **In parallel throughout — `explicit-namespace-dsl` (Serhii).** The launch blocker; depends only on the landed TML-2605.
+6. **Integration — `extension-supabase` M2→M4.** Role binding (`asUser`/`asAnon`/`asServiceRole`), the live-query RLS e2e, real-Supabase acceptance, close-out.
 
-"Longest pole first" is the right call for *parallel* lanes, so `postgres-rls` should move onto its own lane as early as one frees — which coincides with TML-2537 landing (also unblocking RLS-PSL).
+"Longest pole first" is the right call for *parallel* lanes, so `postgres-rls` should move onto its own lane as early as one frees — which coincides with TML-2537 (step 3) landing.
 
 ## What's in the umbrella directory