fix(nonce): correct TICKN candidate freeze and η_ph hash

[wcatz]

Summary

• Bug 1: FreezeCandidate off-by-one — candidate nonce included one extra block's contribution because ProcessBlock evolved before freeze. Moved freeze inside ProcessBlock before evolution.
• Bug 2: TICKN η_ph used last block's own hash instead of prevHash (= second-to-last block hash). Added GetPrevHashOfLastBlock store method.
• Bug 3: RecomputeCurrentEpochNonce trusted stored nonce_value instead of recomputing from raw VRF output. Added GetVrfOutputsForEpoch.

Bugs 1+2 compound to produce completely wrong nonces on the auto-leaderlog trigger at the stability window. CLI/manual commands were unaffected (fall through to Koios).

Post-deploy

After deploying, clear cached wrong TICKN nonces:

DELETE FROM epoch_nonces WHERE source = 'computed';
DELETE FROM leader_schedules WHERE epoch >= (SELECT MAX(epoch)-2 FROM leader_schedules);

Test plan

• Deploy to goduckbot-test (latest tag)
• Wait for next epoch stability window trigger
• Compare auto-triggered leaderlog nonce against Koios
• Run /nonce <epoch> and verify match

🤖 Generated with Claude Code

Summary by CodeRabbit

• Refactor
  • Restructured nonce computation to derive epoch values directly from VRF outputs, replacing prior storage-based approach for enhanced consistency across epoch transitions.
  • Nonce freezing now automatically occurs at the epoch stability window, eliminating manual operations and improving epoch synchronization efficiency.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

Configuration used: defaults

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 02502a7 and 59f29b8.

📒 Files selected for processing (4)

• db.go
• main.go
• nonce.go
• store.go

---

📝 Walkthrough

Walkthrough

This change refactors nonce computation to rely on raw VRF outputs instead of stored nonce values. New database methods retrieve VRF outputs per epoch and the previous block's hash. Nonce candidate freezing moves from an explicit API call into ProcessBlock at the stability window. The NonceTracker's public FreezeCandidate method is removed in favor of internal epoch-driven logic.

Changes

Cohort / File(s)	Summary
Store & Database Layer store.go, db.go	Introduced VrfBlock type and extended Store interface with GetVrfOutputsForEpoch and GetPrevHashOfLastBlock methods; added corresponding PgStore and SqliteStore implementations using raw SQL queries ordered by slot and leveraging LIMIT/OFFSET for previous block retrieval.
Nonce Computation Rewrite nonce.go	Switched from stored nonce_value sequences to recomputation from raw VRF outputs via GetVrfOutputsForEpoch; integrated candidate nonce freezing into ProcessBlock at stability window; updated GetNonceForEpoch to compute η_ph using GetPrevHashOfLastBlock; removed public FreezeCandidate method; updated RecomputeCurrentEpochNonce and ComputeEpochNonce to derive nonces directly from VRF data with revised logging and state management.
Block Processing Flow main.go	Removed explicit nonceTracker.FreezeCandidate(epoch) call from checkLeaderlogTrigger, relocating freeze behavior into ProcessBlock logic; trigger activation and leaderlog calculation flows remain unchanged.

Sequence Diagram(s)

sequenceDiagram
    actor Client as Client
    participant PB as ProcessBlock
    participant NT as NonceTracker
    participant Store as Store/Database
    
    Client->>PB: ProcessBlock(slot, epoch, blockHash, vrfOutput)
    activate PB
    
    alt At Stability Window
        PB->>NT: Freeze candidate nonce internally
        activate NT
        NT->>NT: Log & persist η_c = evolving nonce
        deactivate NT
    end
    
    PB->>PB: Continue with normal processing
    deactivate PB
    
    Client->>NT: RecomputeCurrentEpochNonce(ctx, epoch)
    activate NT
    
    NT->>Store: GetVrfOutputsForEpoch(ctx, epoch)
    activate Store
    Store-->>NT: []VrfBlock (ordered by slot)
    deactivate Store
    
    loop For each VrfBlock
        NT->>NT: Recompute nonce from VRF output
    end
    
    NT->>Store: GetPrevHashOfLastBlock(ctx, epoch-1)
    activate Store
    Store-->>NT: Previous epoch's second-to-last block hash (η_ph)
    deactivate Store
    
    NT->>NT: Persist recomputed nonce & update in-memory state
    deactivate NT
    
    Client->>NT: GetNonceForEpoch(epoch)
    activate NT
    NT->>Store: GetPrevHashOfLastBlock(ctx, epoch-1)
    activate Store
    Store-->>NT: η_ph for epoch transition
    deactivate Store
    NT->>NT: Compute η = H(η_c || η_ph)
    NT-->>Client: []byte (derived nonce)
    deactivate NT

Loading

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

• fix(nonce): Koios fallback for η_ph in TICKN computation #80: Introduces Koios fallback for η_ph retrieval, directly complementing this PR's new GetPrevHashOfLastBlock database method for previous-block-hash sourcing.
• feat(chainsync): full chainsyc history #20: Establishes the Store/PgStore/SqliteStore abstractions that this PR extends with two new retrieval methods.
• feat: nonce integrity checker + schedule history backfill #56: Adds StreamBlockVrfOutputs and NonceIntegrityCheck utilities for VRF-based nonce verification, aligning with this PR's shift to raw VRF output recomputation.

Poem

🐰 Freeze the nonce when stability blooms,
Recompute from VRF's truthful rooms,
Hash the last block, derive the new way,
No stored values lead us astray! ✨

✨ Finishing Touches

• 📝 Generate docstrings (stacked PR)
• 📝 Generate docstrings (commit on current branch)

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch fix/nonce-tickn-bugs

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}