Front door · Scout app · APIs · provenance · settlement
The public site becomes the front door to the complete PGF implementation: GrappleMap, PGF Scout, live scoring surfaces, FastAPI services, and ContextOS-for-PGF provenance. One experience for leadership, athletes, fans, operators, and technical diligence.
The provenance layer keeps settlement and officiating decisions tied to receipts, overlays, policy, and replayable state.
The companies winning sports technology sell the same core promise: trusted live data that powers fan engagement, media products, integrity programs, and regulated betting markets. PGF now has a grappling-native version of that stack.
Any tape — pro card or amateur open call — is ingested and graded by a three-agent vision pipeline into six talent metrics plus an XP rating. Unknown prospects arrive with a scouting record from the first touch, not a highlight reel.
Positions, grips, and submission threats are classified against a closed registry and mapped to PGF's official rulebook — Kill 6, Break 3, DQ 2, sub-under-1:00 +1 — deterministically. Same rules, same math, every match, replayable forever.
Each exchange can become a verifiable market event: takedown attempt, control window, submission threat, finish type, or specific technique. Each outcome seals into a tamper-evident packet that a licensed operator can verify before settlement.
The platform is not a highlight generator. It is a compiled cognitive pipeline that parses mat video, constrains model output to PGF-native schemas, applies the ruleset, seals the result through ContextOS-for-PGF, and converts athlete performance into scouting, fantasy, draft, tryout, and settlement products.
Three independent agents split the fight into non-overlapping domains: positional dominance, grip fighting, and submission probability. Outputs are constrained to strict Pydantic v2 snake_case identifiers mapped to the GrappleMap taxonomy.
Before a score becomes productized data, the active scoring code and rubric state are checked against the expected digest. SettlementProof envelopes bind the scoring receipt to video-derived evidence, rubric digest, timestamp state, and signature.
{
"match_id": "match_s10_pro_jett_thompson",
"winner_id": "THOMPSON",
"payout_points": 4.0,
"video_hash": "a4b7f94...",
"rubric_digest": "e7080d31...",
"signature": "c98c2bd9..."
}
Amateur uploads become six-metric fighter cards: Submission Efficiency, Control Dominance, Escape Rate, Transition Speed, Technique Diversity, and Explosiveness. The platform compares that vector to the pro roster and assigns the closest pro analogue.
The Global Talent Ledger flags amateurs whose XP reaches the top-quintile professional baseline. Franchise owners then use the draft bid advisor to avoid emotional overpays and find value arbitrage in underpriced athletes.
pgf_scout_platform.py executable platform
pgf_scout_analysis_viz.png draft economics visualization
The repo is not just a static website. It is a complete product stack: the public front door, the PGF Scout product app, and the FastAPI backend that carries Scout, GrappleMap, recognition, auth, markets, and ContextOS-for-PGF provenance surfaces.
pgf.iaib.ai becomes the executive and public entry point: the PGF Kill Guide, the thesis page, doc vault, and browser GrappleMap experience.
The React app carries the actual product surfaces: login, register, scoreboard, recognition views, athlete cards, leaderboard, admin, and fight-night UX.
The backend mounts the operational APIs: Scout, GrappleMap, recognition, auth, billing, admin, BKFC, and ContextOS-for-PGF settlement/provenance routes.
“pgf.iaib.ai is the front door. Behind it is the unified PGF platform: Scout, GrappleMap, recognition, scoring, and ContextOS-for-PGF provenance running as one fight-night data system.”
ContextOS-for-PGF is not the vision model, not a chatbot, not a vector database, and not a blockchain pitch. In this implementation, it is the verifiable settlement and officiating trust kernel: receipts, correction overlays, policy decisions, rubric epochs, proof packs, and replayable provenance bound to PGF Scout outputs.
PGF Scout produces the athlete and match analysis. ContextOS-for-PGF binds the relevant outputs into envelopes, receipts, overlays, policy results, ledger entries, proofs, and market-settlement surfaces that can be verified and replayed.
The system preserves what was believed at the time of analysis instead of silently overwriting it later.
Corrections suppress or supersede stale outputs while preserving the original state for audit, replay, and model improvement.
Receipts, digests, rubric epochs, approvals, and ledger entries make the settlement/officiating trail replayable instead of anecdotal.
Click each stage to see what happens inside the machine. Nothing here is a mockup — every stage maps to a schema-locked interface already living in the private repo.
Match video enters through quota-controlled onboarding — a pro card feed or an amateur open-call upload. Each bout becomes a scoped analysis order defining exactly what evidence, model plan, and actions are allowed for that single match.
MatchAnalysisOrder → EvidenceClip (frame/timestamp-bounded)
Three specialized agents process the same stream concurrently — and each is barred from the others' domain, so signals never cross-contaminate. Positional dominance, grip fighting, and submission probability are graded in parallel.
PositionalDominance · GripFighting · SubmissionProbability
Every agent output is validated against a closed registry of 160+ grappling transitions. Anything off-registry is rejected — not guessed. The active rulebook is sealed into a chained RubricEpoch so historical scores stay anchored to the rules in force on match day.
Pydantic v2 enforcement · prev_rubric_digest → rubric_digest
When a human official corrects the AI, the original prediction is never deleted. A timestamped overlay suppresses the stale belief for live queries while preserving the original for auditors and model retraining — forensic-grade lineage a regulator can replay.
ResultCorrection → non-destructive overlay (point-in-time)
Validated scores publish six talent metrics plus XP to a live board where seeded athletes meet newly scouted prospects. Top-quintile performers automatically earn real-world tryout invitations — the digital-to-physical talent bridge.
xp = min(round(talent_score × 24), 2400)
The terminal artifact: four independent data layers — event, video, model, ruleset — are hashed into a single root digest in the same process frame, timestamped, and signed. Any mismatch on any layer rejects the settlement. This closes the past-posting fraud window.
root = SHA-256(H1‖H2‖H3‖H4) · ECDSA-P256 signed
Signed proofs broadcast to sportsbook, regulatory, and anchoring endpoints. A scoped human-consent token is required before any live-mode settlement executes — so PGF keeps a hand on the switch while the machine does the work.
SettlementProof → Sportsbook · Regulator · SCITT anchor
Built to explain the machine in plain language. Tap any card to play.
The big money does not stop at “who wins?” It moves into the moments inside the match: throws, submission chains, control time, first attacks, and finish probabilities. PGF's data layer makes those moments verifiable enough for fantasy products and licensed betting partners to trust.
That question is only commercially useful if the answer is not a commentator’s opinion or a disputed judging call. PGF can bind the video window, technique classification, confidence trail, rulebook, correction history, and final outcome into a replayable proof packet. No proof, no settlement.
Draftable athlete cards with live XP — the funnel that turns fans into repeat data contributors.
Uchi mata in 30 seconds, first back take, first deep submission threat, first guard pass — markets around the action fans already argue about.
Deterministic in-play scores fans and operators can inspect against the same sealed rulebook.
One signed proof per outcome, binding video + model + ruleset — the artifact books need to auto-settle.
Non-destructive corrections and chained epochs give regulators a fully replayable history.
Signed proofs stream to book, regulator, and anchor endpoints — a settlement oracle without human intermediaries.
Human authority is preserved, but every call is tied to evidence and a correction trail — the machine proposes, PGF disposes, and everyone can replay the basis.
The proof is sealed at the moment of the physical event, closing the fraud window between a real submission and the ledger.
This is a data and proof layer — not a live gambling product. Any wagering deployment requires a licensed operator and full regulatory / gaming-commission integration in the applicable jurisdiction. PGF supplies the verified scoring, audit trail, and settlement packet that makes combat-sports markets more accountable.
On July 3, 2026 Dustin filed an omnibus provisional patent application covering five interlocking targets. Copy one piece and you still can't ship the system. That's the point.
Four-layer cryptographic binding of event, video, model, and ruleset into one signed root — the lock on the regulated wagering market.
Non-destructive correction overlays + chained rubric epochs + code-drift guard — the forensic lineage gaming commissions require.
Schema-constrained layer isolates vision from scoring — rulesets hot-swap with no model retraining. Ruleset-agnostic licensing.
Exclusion-bounded parallel agents deliver 30 matches in a 5.5-hour window with zero cross-contamination.
Every uploader is a paying customer and an unpaid labeler — a compounding data moat late entrants can't buy.
Each target defends the next. Together they form a design-around barrier that is prohibitively expensive to engineer past.
Status: provisional patent application filed (patent-pending infrastructure). No claims are issued patents. Priority date secured for all five targets; non-provisional filings roll out within the 12-month window.
The rails are built. PGF can roll them out as a talent product, a fan-engagement product, or a licensed-operator proof layer — each option compounds into the next.
Roll out the scouting pipeline as PGF's talent-identification standard. Every bout becomes a scored, replayable record; the athlete database becomes proprietary and compounding.
Launch the quarterly open-call Fantasy Week. Flood the funnel, audit every prospect, seed the leaderboard, and convert fans into a self-reinforcing data engine.
Partner with a licensed operator to pilot settlement proofs as the trust layer for regulated grappling markets — from match outcomes to action-based micro-markets like technique windows, control time, and submission threats.
This is the private conversation after the reveal: what PGF wants to own first, how public to make the patent-pending moat, and which commercial partner category to approach.
A → B → C is the natural sequence: scouting proves the data, Fantasy Week scales it, settlement monetizes the trust. But any single path stands alone. Where does PGF want its first public win?
The critical patent claims target the sportsbook market directly. Pursuing it means engaging a licensed operator and a gaming commission early. Is that a 2026 conversation or a 2027 one?
With the provisional filed, PGF can begin signaling "patent-pending settlement infrastructure" to partners and sponsors. How much do we reveal, and to whom?
Options: a media/broadcast data deal, a fantasy platform integration, or an operator pilot. Which partner conversation does PGF want Dustin to prepare for?
The two critical targets (SettlementProof, Honesty Kernel) should convert to non-provisionals first. That's a budget and counsel decision due before July 2027.
A live Fantasy Week dry-run? A settlement-proof verification against a real bout? A talent-board with seeded + scouted athletes? Name the artifact that would move the decision forward.