feat: implement exponential-decay pioneer multiplier (fixes #240)

README.md

@@ -43,7 +43,7 @@ The result: a sustainable incentive layer that channels resources toward buildin
 # Quick start
 git clone https://github.com/entrius/gittensor.git
 cd gittensor
-cp env.example .env
+cp .env.example .env
 # Edit .env with proper values
 nano .env
 
@@ -60,7 +60,7 @@ See full guide **[here](https://docs.gittensor.io/miner.html)**
 # Quick start
 git clone https://github.com/entrius/gittensor.git
 cd gittensor
-cp env.example .env
+cp .env.example .env
 # Edit .env with proper values
 nano .env
 
@@ -83,7 +83,7 @@ _NOTE: don’t be afraid to provide recommendations for your favorite open sourc
 
 - Programming Language Weights
 
-A list of major file types/extensions, mostly related to programming languages, but also plenty of markdown, documentation, and other common files are included. Each extension has a weight for scoring. If the extension has a language full name then it code in those languages will be evaluated using token-based scoring.
+A list of major file types/extensions, mostly related to programming languages, but also plenty of markdown, documentation, and other common files are included. Each extension has a weight for scoring. If the extension has a language full name then code in those languages will be evaluated using token-based scoring.
 
 _NOTE: this list will also be dynamic. Additions, and weight changes will occur as the subnet matures._
 

docker-compose.miner.yml

@@ -3,6 +3,7 @@ services:
     image: entrius/gittensor:latest
     container_name: gt-miner
     restart: unless-stopped
+    network_mode: "host"
     entrypoint: /app/scripts/miner-entrypoint.sh
     env_file:
       - .env
@@ -11,5 +12,6 @@ services:
     volumes:
       # 'ro' = readonly
       - ${WALLET_PATH}:/root/.bittensor/wallets:ro
+    command: ["--subtensor.chain_endpoint", "${SUBTENSOR_CHAIN_ENDPOINT:-ws://172.17.0.1:9944}"]
     labels:
       - "com.centurylinklabs.watchtower.enable=true"

gittensor/classes.py

@@ -156,7 +156,7 @@ class PullRequest:
     base_score: float = 0.0
     issue_multiplier: float = 1.0
     open_pr_spam_multiplier: float = 1.0
-    repository_uniqueness_multiplier: float = 1.0
+    pioneer_multiplier: float = 1.0
     time_decay_multiplier: float = 1.0
     credibility_multiplier: float = 1.0
     raw_credibility: float = 1.0  # Before applying ^k scalar
@@ -194,7 +194,7 @@ def calculate_final_earned_score(self) -> float:
             'repo': self.repo_weight_multiplier,
             'issue': self.issue_multiplier,
             'spam': self.open_pr_spam_multiplier,
-            'unique': self.repository_uniqueness_multiplier,
+            'pioneer': self.pioneer_multiplier,
             'decay': self.time_decay_multiplier,
             'cred': self.credibility_multiplier,
         }

gittensor/constants.py

@@ -59,7 +59,8 @@
 DEFAULT_MAX_CONTRIBUTION_SCORE_FOR_FULL_BONUS = 2000
 
 # Boosts
-UNIQUE_PR_BOOST = 0.74
+PIONEER_PR_BOOST = 5.0  # Big reward for the first miner to merge a PR on a repo inactive for 90 days.
+PIONEER_LOOKBACK_DAYS = 90  # Repos with no merged PRs in this window are considered untouched.
 MAX_CODE_DENSITY_MULTIPLIER = 3.0
 
 # Issue boosts

gittensor/utils/github_api_tools.py

@@ -801,64 +801,71 @@ def load_miners_prs(
             page_info: Dict = pr_data.get('pageInfo', {})
 
             for pr_raw in prs:
-                repository_full_name = parse_repo_name(pr_raw['repository'])
-                pr_state = pr_raw['state']
-
-                # Stop querying once we hit PRs older than the tier incentive start date
-                pr_creation_time = datetime.fromisoformat(pr_raw['createdAt'].rstrip('Z')).replace(tzinfo=timezone.utc)
+                try:
+                    repository_full_name = parse_repo_name(pr_raw['repository'])
+                    pr_state = pr_raw['state']
 
-                if pr_creation_time < TIER_BASED_INCENTIVE_MECHANISM_START_DATE:
-                    bt.logging.info(
-                        f'Reached PR #{pr_raw["number"]} in {repository_full_name} created at {pr_creation_time}, '
-                        f'before tier incentive start date ({TIER_BASED_INCENTIVE_MECHANISM_START_DATE}). '
-                        f'Stopping PR fetch.'
+                    # Stop querying once we hit PRs older than the tier incentive start date
+                    pr_creation_time = datetime.fromisoformat(pr_raw['createdAt'].rstrip('Z')).replace(
+                        tzinfo=timezone.utc
                     )
-                    return
 
-                if repository_full_name not in master_repositories:
-                    bt.logging.info(f'Skipping PR #{pr_raw["number"]} in {repository_full_name} - ineligible repo')
-                    continue
+                    if pr_creation_time < TIER_BASED_INCENTIVE_MECHANISM_START_DATE:
+                        bt.logging.info(
+                            f'Reached PR #{pr_raw["number"]} in {repository_full_name} created at {pr_creation_time}, '
+                            f'before tier incentive start date ({TIER_BASED_INCENTIVE_MECHANISM_START_DATE}). '
+                            f'Stopping PR fetch.'
+                        )
+                        return
 
-                repo_config = master_repositories[repository_full_name]
+                    if repository_full_name not in master_repositories:
+                        bt.logging.info(f'Skipping PR #{pr_raw["number"]} in {repository_full_name} - ineligible repo')
+                        continue
 
-                # Check if repo is inactive
-                if repo_config.inactive_at is not None:
-                    inactive_dt = datetime.fromisoformat(repo_config.inactive_at.rstrip('Z')).replace(
-                        tzinfo=timezone.utc
-                    )
-                    # Skip PR if it was created after the repo became inactive
-                    if pr_creation_time >= inactive_dt:
-                        bt.logging.info(
-                            f'Skipping PR #{pr_raw["number"]} in {repository_full_name} - PR was created after repo became inactive (created: {pr_creation_time.isoformat()}, inactive: {inactive_dt.isoformat()})'
+                    repo_config = master_repositories[repository_full_name]
+
+                    # Check if repo is inactive
+                    if repo_config.inactive_at is not None:
+                        inactive_dt = datetime.fromisoformat(repo_config.inactive_at.rstrip('Z')).replace(
+                            tzinfo=timezone.utc
                         )
+                        # Skip PR if it was created after the repo became inactive
+                        if pr_creation_time >= inactive_dt:
+                            bt.logging.info(
+                                f'Skipping PR #{pr_raw["number"]} in {repository_full_name} - PR was created after repo became inactive (created: {pr_creation_time.isoformat()}, inactive: {inactive_dt.isoformat()})'
+                            )
+                            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)
                         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)
-                    continue
+                    should_skip, skip_reason = should_skip_merged_pr(
+                        pr_raw, repository_full_name, repo_config, lookback_date_filter
+                    )
 
-                should_skip, skip_reason = should_skip_merged_pr(
-                    pr_raw, repository_full_name, repo_config, lookback_date_filter
-                )
+                    if should_skip:
+                        bt.logging.debug(skip_reason)
+                        continue
 
-                if should_skip:
-                    bt.logging.debug(skip_reason)
-                    continue
+                    miner_eval.add_merged_pull_request(pr_raw)
 
-                miner_eval.add_merged_pull_request(pr_raw)
+                except Exception as e:
+                    pr_number = pr_raw.get('number', '?')
+                    bt.logging.warning(f'Error processing PR #{pr_number}, skipping: {e}')
 
             if not page_info.get('hasNextPage') or len(prs) == 0:
                 break
 
             cursor = page_info.get('endCursor')
 
-        bt.logging.info(
-            f'Fetched {len(miner_eval.merged_pull_requests)} merged PRs, {len(miner_eval.open_pull_requests)} open PRs, '
-            f'{len(miner_eval.closed_pull_requests)} closed'
-        )
-
     except Exception as e:
-        bt.logging.error(f'Error fetching PRs via GraphQL: {e}')
+        bt.logging.error(f'Unexpected error fetching PRs via GraphQL: {e}')
+
+    bt.logging.info(
+        f'Fetched {len(miner_eval.merged_pull_requests)} merged PRs, {len(miner_eval.open_pull_requests)} open PRs, '
+        f'{len(miner_eval.closed_pull_requests)} closed'
+    )
 
 
 def extract_pr_number_from_url(pr_url: str) -> Optional[int]:

gittensor/validator/evaluation/scoring.py

@@ -26,7 +26,7 @@
     TIME_DECAY_MIN_MULTIPLIER,
     TIME_DECAY_SIGMOID_MIDPOINT,
     TIME_DECAY_SIGMOID_STEEPNESS_SCALAR,
-    UNIQUE_PR_BOOST,
+    PIONEER_PR_BOOST,
 )
 from gittensor.utils.github_api_tools import (
     FileContentPair,
@@ -348,8 +348,8 @@ def finalize_miner_scores(miner_evaluations: Dict[int, MinerEvaluation]) -> None
 
         # Process merged PRs
         for pr in evaluation.merged_pull_requests:
-            pr.repository_uniqueness_multiplier = calculate_uniqueness_multiplier(
-                pr.repository_full_name, repo_counts, total_contributing_miners
+            pr.pioneer_multiplier = calculate_pioneer_multiplier(
+                pr, evaluation, miner_evaluations
             )
 
             # Apply spam multiplier (calculated once per miner based on unlocked tiers)
@@ -431,15 +431,52 @@ def finalize_miner_scores(miner_evaluations: Dict[int, MinerEvaluation]) -> None
     bt.logging.info('Finalization complete.')
 
 
-def calculate_uniqueness_multiplier(
-    repo_full_name: str, repo_counts: Dict[str, int], total_contributing_miners: int
+def calculate_pioneer_multiplier(
+    target_pr: PullRequest, target_miner_eval: MinerEvaluation, all_evaluations: Dict[int, MinerEvaluation]
 ) -> float:
-    """Calculate repository uniqueness multiplier based on how many miners contribute to a repo."""
-    if total_contributing_miners == 0:
-        return 1.0
-    repo_count = repo_counts.get(repo_full_name, 0)
-    uniqueness_score = (total_contributing_miners - repo_count + 1) / total_contributing_miners
-    return 1.0 + (uniqueness_score * UNIQUE_PR_BOOST)
+    """
+    Calculate pioneer multiplier for a merged PR.
+
+    Instead of a naive binary check, this uses an exponential decay model.
+    The bonus starts at its maximum for the very first PR in a repository
+    within the 90-day lookback window, and halves for each subsequent PR 
+    merged on that same repository, regardless of which miner merges it.
+
+    This creates a "gold rush" dynamic:
+    - 1st PR: 1.0 + MAX_BOOST (e.g., 6.0)
+    - 2nd PR: 1.0 + (MAX_BOOST * 0.5) (e.g., 3.5)
+    - 3rd PR: 1.0 + (MAX_BOOST * 0.25) (e.g., 2.25)
+    - ... and so on until it reaches ~1.0.
+    """
+    repo_name = target_pr.repository_full_name
+    target_time = target_pr.merged_at
+    previous_prs_count = 0
+
+    for uid, evaluation in all_evaluations.items():
+        if evaluation is None:
+            continue
+        for pr in evaluation.merged_pull_requests:
+            if pr.repository_full_name == repo_name and pr.merged_at:
+                # Count how many PRs were merged BEFORE the target PR
+                if pr.merged_at < target_time:
+                    previous_prs_count += 1
+
+                # Tie-breaker for PRs merged at the exact same second
+                elif pr.merged_at == target_time and pr.number != target_pr.number:
+                    if uid < target_miner_eval.uid:
+                        previous_prs_count += 1
+                    # If same miner merged two PRs at the same exact second, break tie by PR number
+                    elif uid == target_miner_eval.uid and pr.number < target_pr.number:
+                        previous_prs_count += 1
+
+    # Apply exponential decay: halving the boost for each prior PR
+    pioneer_bonus = PIONEER_PR_BOOST * (0.5 ** previous_prs_count)
+
+    if pioneer_bonus > 0.05:  # Only log significant bonuses to reduce noise
+        bt.logging.info(f"PIONEER BONUS: {repo_name} | PR #{target_pr.number} (Miner {target_miner_eval.uid}) "
+                        f"| Prior PRs: {previous_prs_count} | Multiplier: {1.0 + pioneer_bonus:.2f}")
+
+    return 1.0 + pioneer_bonus
 
 
 def calculate_issue_multiplier(pr: PullRequest) -> float:

gittensor/validator/storage/queries.py

@@ -37,7 +37,7 @@
     number, repository_full_name, uid, hotkey, github_id, title, author_login,
     merged_at, pr_created_at, pr_state,
     repo_weight_multiplier, base_score, issue_multiplier,
-    open_pr_spam_multiplier, repository_uniqueness_multiplier, time_decay_multiplier,
+    open_pr_spam_multiplier, pioneer_multiplier, time_decay_multiplier,
     credibility_multiplier, raw_credibility, credibility_scalar,
     earned_score, collateral_score,
     additions, deletions, commits, total_nodes_scored,
@@ -56,7 +56,7 @@
     base_score = EXCLUDED.base_score,
     issue_multiplier = EXCLUDED.issue_multiplier,
     open_pr_spam_multiplier = EXCLUDED.open_pr_spam_multiplier,
-    repository_uniqueness_multiplier = EXCLUDED.repository_uniqueness_multiplier,
+    pioneer_multiplier = EXCLUDED.pioneer_multiplier,
     time_decay_multiplier = EXCLUDED.time_decay_multiplier,
     credibility_multiplier = EXCLUDED.credibility_multiplier,
     raw_credibility = EXCLUDED.raw_credibility,

gittensor/validator/storage/repository.py

@@ -166,7 +166,7 @@ def store_pull_requests_bulk(self, pull_requests: List[PullRequest]) -> int:
                     pr.base_score,
                     pr.issue_multiplier,
                     pr.open_pr_spam_multiplier,
-                    pr.repository_uniqueness_multiplier,
+                    pr.pioneer_multiplier,
                     pr.time_decay_multiplier,
                     pr.credibility_multiplier,
                     pr.raw_credibility,

gittensor/validator/weights/master_repositories.json

@@ -498,7 +498,7 @@
     "weight": 4.25
   },
   "autoppia/autoppia_web_agents_subnet": {
-    "additional_acceptable_branches": ["dev"],
+    "additional_acceptable_branches": ["dev", "dev-gittensor"],
     "tier": "Silver",
     "weight": 6.46
   },
@@ -746,6 +746,7 @@
     "weight": 0.19
   },
   "bokeh/bokeh": {
+    "additional_acceptable_branches": ["branch-*.*"],
     "tier": "Bronze",
     "weight": 0.17
   },