feat: add cross-run accumulation ledger

[dopexthrone]

Summary

• Closes Ada's open self-referential loop: the compiler now reads AND writes its own compile state across runs
• Governor Write Hook: on ACCEPT, writes a CompileRecord to the accumulation ledger with intent hash, gate deltas, per-stage entropy readings, and blueprint postcode (PostcodeAddress: ML.ACC.{hash}/v1)
• Intent Read Hook: on next compile, reads recent ledger entries and injects them as auxiliary context into the Intent stage, surfacing prior bottlenecks and high-entropy stages

No new agents. No new pipeline stages. Two hooks, one new artifact type.

Files changed

Package	File	Change
@ada/provenance	postcode.ts	Add ACC stage code
@ada/compiler	types.ts	Add CompileRecord, GateDelta, AccumulationContext types
@ada/compiler	engine.ts	accumulationContext option + buildCompileRecord() / buildAccumulationContext() static methods
@ada/compiler	index.ts	Export new types
@ada/storage	index.ts	compile_ledger table + recordCompile() / getRecentCompileRecords() / getRecordsByIntentHash()
@ada/cli	init.ts	Read ledger before compile, write ledger after ACCEPT
@ada/cli	compile.ts	Same read/write hooks for standalone compile command

What this closes

• The self-referential loop is now fully closed — Ada reads AND writes its own state
• PostcodeAddress on ML.ACC.{hash}/v{n} — provenance system already handles this
• The SSOT remains hand-managed (frozen, authoritative) — the ledger is a separate, live artifact

What this does NOT do (deferred)

• Governor criteria evolution — still fixed. The ledger enables evidence accumulation but does not change the gate
• Domain transfer weighting — the ledger captures domain-agnostic patterns, but routing based on them is a future epoch

Test plan

• Build all modified packages: pnpm --filter @ada/provenance --filter @ada/storage --filter @ada/compiler build
• Verify compile_ledger table is created on first AdaStorage instantiation
• Run ada init "test" → confirm CompileRecord written to ledger on ACCEPT
• Run ada init "test" again → confirm accumulation context injected into intent
• Verify ledger entries carry valid ML.ACC.{hash}/v1 postcodes

https://claude.ai/code/session_012aDqyuDBYMws5ZF2DuHeKs

[cloudflare-workers-and-pages]

Deploying motherlabs with    Cloudflare Pages

Latest commit:	1cb7805
Status:	✅  Deploy successful!
Preview URL:	https://c80c0cb5.motherlabs.pages.dev
Branch Preview URL:	https://claude-add-accumulation-ledg.motherlabs.pages.dev

View logs

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1cb78051e6

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Write compile ledger entries only after ACCEPT

This hook persists a CompileRecord unconditionally before the command checks gov.decision, so REJECT/ITERATE runs (including --strict failures) still enter the accumulation ledger and are then fed back into later intent context. That contradicts the accept-only contract used elsewhere (e.g., init path) and can pollute cross-run memory with runs that were explicitly rejected by governance.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Avoid replacing ledger rows by non-unique run IDs

Using INSERT OR REPLACE on compile_ledger keyed by run_id can silently overwrite prior records when two compiles share the same generated run ID (callers use second-level timestamps like ML-${Math.floor(Date.now()/1000)}). That loses historical accumulation data and skews the context read hook toward fewer/incorrect prior runs.

Useful? React with 👍 / 👎.