feat(scoring): gate PR scoring by PAT first-registration date

Author: HadesHappy

gittensor/constants.py

@@ -75,6 +75,7 @@
 # =============================================================================
 DEFAULT_REPO_WEIGHT = 0.01  # fallback weight for repos not in master_repositories.json
 PR_LOOKBACK_DAYS = 35  # rolling window for scoring
+REGISTRATION_GRACE_DAYS = 7
 MERGED_PR_BASE_SCORE = 25
 MIN_TOKEN_SCORE_FOR_BASE_SCORE = 5  # PRs below this get 0 base score
 MAX_CONTRIBUTION_BONUS = 25

gittensor/utils/github_api_tools.py

@@ -728,6 +728,7 @@ def try_add_open_or_closed_pr(
     pr_raw: Dict,
     pr_state: str,
     lookback_date_filter: datetime,
+    registration_cutoff: Optional[datetime] = None,
 ) -> None:
     """
     Attempts to add an OPEN or CLOSED PR to miner_eval if eligible.
@@ -737,6 +738,7 @@ def try_add_open_or_closed_pr(
         pr_raw: Raw PR data from GraphQL
         pr_state: GitHub PR state (OPEN, CLOSED, MERGED)
         lookback_date_filter: Date filter for lookback period
+        registration_cutoff: Skip closed PRs whose closed_at is earlier than this cutoff
     """
     # Ignore all maintainer contributions
     if not os.environ.get('DEV_MODE') and pr_raw.get('authorAssociation') in MAINTAINER_ASSOCIATIONS:
@@ -764,6 +766,9 @@ def try_add_open_or_closed_pr(
         if created_dt < lookback_date_filter:
             return
 
+        if registration_cutoff is not None and closed_dt < registration_cutoff:
+            return
+
         if closed_dt >= lookback_date_filter:
             miner_eval.add_closed_pull_request(pr_raw)
 
@@ -773,6 +778,7 @@ def should_skip_merged_pr(
     repository_full_name: str,
     repo_config: RepositoryConfig,
     lookback_date_filter: datetime,
+    registration_cutoff: Optional[datetime] = None,
 ) -> tuple[bool, Optional[str]]:
     """
     Validate a merged PR against all eligibility criteria.
@@ -782,6 +788,7 @@ def should_skip_merged_pr(
         repository_full_name (str): Full repository name (owner/repo)
         repo_config (RepositoryConfig): Repository configuration
         lookback_date_filter (datetime): Date filter for lookback period
+        registration_cutoff (Optional[datetime]): Skip merged PRs whose merged_at is earlier than this cutoff
 
     Returns:
         tuple[bool, Optional[str]]: (should_skip, skip_reason) - True if PR should be skipped with reason
@@ -799,6 +806,12 @@ def should_skip_merged_pr(
             f'Skipping PR #{pr_raw["number"]} in {repository_full_name} - merged before {PR_LOOKBACK_DAYS}-day lookback window',
         )
 
+    if registration_cutoff is not None and merged_dt < registration_cutoff:
+        return (
+            True,
+            f'Skipping PR #{pr_raw["number"]} in {repository_full_name} - merged before miner registration on this validator',
+        )
+
     # Skip if PR author is a maintainer
     author_association = pr_raw.get('authorAssociation')
     if not os.environ.get('DEV_MODE') and author_association in MAINTAINER_ASSOCIATIONS:
@@ -854,7 +867,10 @@ def should_skip_merged_pr(
 
 
 def load_miners_prs(
-    miner_eval: MinerEvaluation, master_repositories: Dict[str, RepositoryConfig], max_prs: int = 1000
+    miner_eval: MinerEvaluation,
+    master_repositories: Dict[str, RepositoryConfig],
+    max_prs: int = 1000,
+    registration_cutoff: Optional[datetime] = None,
 ) -> None:
     """
     Fetches user PRs via GraphQL API and categorize them by state.
@@ -864,6 +880,7 @@ def load_miners_prs(
         miner_eval: The MinerEvaluation object containing github details + more
         master_repositories: Repository metadata (name -> RepositoryConfig)
         max_prs: Maximum merged PRs to fetch
+        registration_cutoff: Skip PRs (merged or closed) whose effective timestamp predates this cutoff
     """
     bt.logging.info('*****Fetching PRs*****')
     miner_eval.github_pr_fetch_failed = False
@@ -945,11 +962,17 @@ def load_miners_prs(
                             continue
 
                     if pr_state in (PRState.OPEN.value, PRState.CLOSED.value):
-                        try_add_open_or_closed_pr(miner_eval, pr_raw, pr_state, lookback_date_filter)
+                        try_add_open_or_closed_pr(
+                            miner_eval, pr_raw, pr_state, lookback_date_filter, registration_cutoff=registration_cutoff
+                        )
                         continue
 
                     should_skip, skip_reason = should_skip_merged_pr(
-                        pr_raw, repository_full_name, repo_config, lookback_date_filter
+                        pr_raw,
+                        repository_full_name,
+                        repo_config,
+                        lookback_date_filter,
+                        registration_cutoff=registration_cutoff,
                     )
 
                     if should_skip:

gittensor/validator/oss_contributions/mirror/load.py

@@ -34,13 +34,15 @@ def load_mirror_miner_prs(
     mirror_eval: MirrorMinerEvaluation,
     mirror_repos: Dict[str, RepositoryConfig],
     client: Optional[MirrorClient] = None,
+    registration_cutoff: Optional[datetime] = None,
 ) -> None:
     """Populate mirror_eval with PRs fetched from the mirror service.
 
     Args:
         mirror_eval: container to populate; must already have github_id set
         mirror_repos: repo configs to filter against (only mirror_enabled entries)
         client: optional MirrorClient for dependency injection in tests
+        registration_cutoff: skip merged/closed PRs whose effective timestamp predates this cutoff
     """
 
     bt.logging.info('*****Fetching PRs from mirror*****')
@@ -65,7 +67,7 @@ def load_mirror_miner_prs(
 
     for pr in response.pull_requests:
         try:
-            _maybe_add_pr(mirror_eval, pr, mirror_repos, lookback_date)
+            _maybe_add_pr(mirror_eval, pr, mirror_repos, lookback_date, registration_cutoff=registration_cutoff)
         except Exception as e:
             bt.logging.warning(f'Error processing mirror PR #{pr.pr_number} ({pr.repo_full_name}): {e}')
 
@@ -80,6 +82,7 @@ def _maybe_add_pr(
     pr: MirrorPullRequest,
     mirror_repos: Dict[str, RepositoryConfig],
     lookback_date: datetime,
+    registration_cutoff: Optional[datetime] = None,
 ) -> None:
     """Apply load-time filters and bucket pr by state if it passes."""
 
@@ -112,12 +115,19 @@ def _maybe_add_pr(
         # closing an old PR shouldn't trigger a fresh credibility penalty).
         if pr.created_at < lookback_date:
             return
+        if registration_cutoff is not None and pr.closed_at is not None and pr.closed_at < registration_cutoff:
+            return
         mirror_eval.closed_prs.append(ScoredMirrorPR(pr=pr))
     elif pr.state == 'MERGED':
         # Apply the merge-eligibility gate at LOAD time (matches legacy parity —
         # should_skip_merged_pr runs inside load_miners_prs before adding). If we
         # deferred to scoring, rejected PRs would remain in mirror_merged_prs and
         # inflate the merged_count used in check_eligibility, distorting credibility.
+        if registration_cutoff is not None and pr.merged_at is not None and pr.merged_at < registration_cutoff:
+            bt.logging.debug(
+                f'Skipping mirror PR #{pr.pr_number} in {pr.repo_full_name} - merged before miner registration on this validator'
+            )
+            return
         candidate = ScoredMirrorPR(pr=pr)
         should_skip, reason = _should_skip_merged_mirror_pr(candidate, repo_config)
         if should_skip:

gittensor/validator/oss_contributions/reward.py

@@ -2,6 +2,7 @@
 # Copyright © 2025 Entrius
 from __future__ import annotations
 
+from datetime import datetime
 from typing import TYPE_CHECKING, Dict, Optional, Set, Tuple
 
 import bittensor as bt
@@ -39,6 +40,7 @@ async def evaluate_miners_pull_requests(
     programming_languages: Dict[str, LanguageConfig],
     token_config: TokenConfig,
     stale_hotkey: Optional[str] = None,
+    registration_cutoff: Optional[datetime] = None,
 ) -> MinerEvaluation:
     """
     Entry point from taking a miners response -> Get PRs -> Score PRs
@@ -51,6 +53,7 @@ async def evaluate_miners_pull_requests(
         programming_languages: The programming languages and their weights
         token_config: Token-based scoring weights configuration
         stale_hotkey: If set, the UID has a stored PAT from this old hotkey (re-registration detected)
+        registration_cutoff: If set, skip merged/closed PRs whose effective timestamp predates this cutoff
 
     Returns:
         MinerEvaluation: The object containing scores, valid_prs, etc.
@@ -75,7 +78,7 @@ async def evaluate_miners_pull_requests(
 
     if legacy_repos:
         with session_scope():
-            load_miners_prs(miner_eval, legacy_repos)
+            load_miners_prs(miner_eval, legacy_repos, registration_cutoff=registration_cutoff)
             score_miner_prs(miner_eval, legacy_repos, programming_languages, token_config)
 
     if mirror_repos:
@@ -85,7 +88,9 @@ async def evaluate_miners_pull_requests(
             github_id=miner_eval.github_id,
         )
         with MirrorClient() as mirror_client:
-            load_mirror_miner_prs(mirror_eval, mirror_repos, client=mirror_client)
+            load_mirror_miner_prs(
+                mirror_eval, mirror_repos, client=mirror_client, registration_cutoff=registration_cutoff
+            )
             score_mirror_miner_prs(mirror_eval, mirror_repos, programming_languages, token_config, client=mirror_client)
         combine(miner_eval, mirror_eval)
 
@@ -133,6 +138,8 @@ async def get_rewards(
             else:
                 stale_hotkey = pat_entry.get('hotkey')
 
+        registration_cutoff = pat_storage.get_registration_cutoff(uid, hotkey)
+
         # Calculate score
         miner_evaluation = await evaluate_miners_pull_requests(
             uid,
@@ -142,6 +149,7 @@ async def get_rewards(
             programming_languages,
             token_config,
             stale_hotkey=stale_hotkey,
+            registration_cutoff=registration_cutoff,
         )
         miner_evaluations[uid] = miner_evaluation
 

gittensor/validator/pat_storage.py

@@ -11,10 +11,12 @@
 import os
 import tempfile
 import threading
-from datetime import datetime, timezone
+from datetime import datetime, timedelta, timezone
 from pathlib import Path
 from typing import Optional
 
+from gittensor.constants import REGISTRATION_GRACE_DAYS
+
 PATS_FILE = Path(__file__).resolve().parents[2] / 'data' / 'miner_pats.json'
 
 _lock = threading.Lock()
@@ -37,14 +39,25 @@ def save_pat(uid: int, hotkey: str, pat: str, github_id: str) -> None:
     """Upsert a PAT entry by UID. Creates the file if needed."""
     with _lock:
         entries = _read_file()
+        now_iso = datetime.now(timezone.utc).isoformat()
+
+        prior = next((e for e in entries if e.get('uid') == uid), None)
+
+        first_registered_at: Optional[str]
+        if prior is not None and prior.get('hotkey') == hotkey:
+            first_registered_at = prior.get('first_registered_at') if 'first_registered_at' in prior else None
+        else:
+            first_registered_at = now_iso
 
         entry = {
             'uid': uid,
             'hotkey': hotkey,
             'pat': pat,
             'github_id': github_id,
-            'stored_at': datetime.now(timezone.utc).isoformat(),
+            'stored_at': now_iso,
         }
+        if first_registered_at is not None:
+            entry['first_registered_at'] = first_registered_at
 
         for i, existing in enumerate(entries):
             if existing.get('uid') == uid:
@@ -56,6 +69,21 @@ def save_pat(uid: int, hotkey: str, pat: str, github_id: str) -> None:
         _write_file(entries)
 
 
+def get_registration_cutoff(uid: int, hotkey: str) -> Optional[datetime]:
+    """Return the registration scoring cutoff for this miner, or None when no gate applies."""
+    entry = get_pat_by_uid(uid)
+    if entry is None or entry.get('hotkey') != hotkey:
+        return None
+    raw = entry.get('first_registered_at')
+    if not raw:
+        return None
+    try:
+        first_registered_at = datetime.fromisoformat(raw)
+    except ValueError:
+        return None
+    return first_registered_at - timedelta(days=REGISTRATION_GRACE_DAYS)
+
+
 def get_pat_by_uid(uid: int) -> Optional[dict]:
     """Look up a single PAT entry by UID. Returns None if not found."""
     with _lock: