Roadmap: useful GHS ↔ Dig/Hunt + fuzzengine v2.7 depth (from FounderB)

[FounderB]

Making real hashrate feed real Hunt/Dig work

Hey team — thank you again for having me in the circle. I’ve been living in Hunt/engine land because it’s the part that feels alive: fleet ASAN, honest reports, and a pool where GPU power isn’t just a scoreboard.

This issue is a clean proposal of what we already built, what we proved locally, and what I’d love to ship next with you — at a pace that stays fun (interest-driven, no rigid schedule).

---

Why this matters

HackMe’s story is useful Proof-of-Work. Today that means:

Rail	What it does	What it pays
GPU / PoH	Real GH/s on the public pool	HMC / SUP
Dig / Hunt	WASM guards + ASAN shards for customers	Escrow (20/80 · 50/50)

If GH/s and Dig/Hunt stay strangers, the slogan is half-true.
If we couple them honestly (no “GPU ASAN” fairy tales), the pool becomes something miners and customers can feel.

---

Already in flight — PR #12

PR: #12
Branch: FounderB:feat/ghs-dig-coupling

1) Real GHS ↔ Dig/Hunt coupling

• Healthy hybrid GH/s → dig claim gap tightens (more shards without killing the GPU)
• Dig-only soft-defers when live hybrid capacity is online (~25% admit share)
• Fleet capacity + ETA on /api/fuzz/pool/stats and campaign progress
  (fleet_hashrate_gh_s, est_shards_per_hour, eta_sec_fleet)
• Doc: docs/GHS_DIG_COUPLING.md

2) fuzzengine v2.7 — deeper mutation for Dig/Hunt

• Havoc ops 16 → 64, stack ≤32, denser crossover
• Format footguns (HTTP/gzip/zip/UTF-16/protobuf/NaN/frames/nesting…)
• Deterministic for coordinator replay
• Docs: docs/HUNT_ENGINE_V26.md · docs/HUNT_ENGINE_V27.md

---

Local proof (my machine today)

Host: Linux · 4 CPU · Go 1.26.8 · clang 18

Mutation A/B (same seeds/dict)

Samples	Unique (v2.7 / baseline)	Lengths	Unique gain	Length gain
5k	4987 / 4794	249 / 55	+4.0%	+353%
20k	19867 / 18761	256 / 61	+5.9%	+320%
50k	49527 / 45789	256 / 63	+8.2%	+306%

Gates: bash scripts/tests/fuzz_engine_local_stress.sh · bash scripts/tests/ghs_dig_coupling_bench.sh ✅

Live Hunt ASAN smoke (cjson / libucl, ~45s wall)

Target	Hunt local	Notes
cjson	CLEAN · 3000 iter · ~112 exec/s	Honest quiet outcome
libucl	INFORMATIONAL · 1785 iter · 6 crashes · ~39 exec/s	Sanitizer signal under Hunt Standard path
libFuzzer (ref)	cjson ~41k exec/s; libucl hit known UBSan class fast	Confirms Hunt ≠ raw exec/s race — depth + deliverable

Artifact dir: reports/hunt-benchmark/20260920T100010Z/

Honesty check: mutation uniqueness ≠ CVE. Hunt value remains verified sanitizer work + report + escrow, not “we mint CVEs.”

---

Proposed next steps (when you’re ready)

Ordered by impact / risk — happy to take any slice:

A. LIVE pool canary of PR #12

1. Merge / deploy GHS coupling + engine v2.7 to a canary coordinator
2. Watch hybrid logs: dig_boosted vs backpressure pauses
3. Compare eta_sec_fleet vs real campaign wall time on one Hunt Lite/Standard
4. Confirm dig-only Sybils no longer monopolize leases while GPUs are online

B. Make capacity customer-visible

• Surface fleet_capacity / ETA in marketplace / progress UI (not only JSON)
• Optional: tiny Capacity add-on SKU (prepaid shard throughput tied to pledged GHS)

C. Hunt Watch honesty 2.0

• Next public ledger series stamps finding_families, rarity / corpus health — cite families, not raw crash artifact counts

D. Customer-repo Hunt edge cases

• Fail-closed paths for weird crates (no fuzz_target!, broken manifests) — keep Phase B quality bar

E. (Later) GPU Dig mutators only

• CUDA/OpenCL generate mutants → CPU WASM/ASAN eval
• Explicitly not GPU ASAN — same honesty rule

---

How I’d like to work

Same agreement we already have: interest-driven, no fixed hours.
My bias is Hunt/engine + pool truthfulness — the parts that make HackMe feel like a real security product, not a hashrate cosplay.

If this roadmap resonates, tell me which letter (A–E) you want first after #12 — I’ll open PRs in that order.

— FounderB
(engine depth · GHS coupling · local Hunt smoke)

[jokeez]

Shipped to main (GHS coupling + engine v2.7 + honesty fixes). Hub canary next: HACKME_FUZZ_CLAIM_GHS_PRIORITY=0 then =1. Thanks — A is in progress.

[jokeez]

Hey FounderB — thank you again for the clean write-up and the PR. Really glad you’re in the circle.

A is done on the live hub:

• Merged to main (GHS ↔ Dig/Hunt coupling + engine v2.7), with honesty fixes before ship (ghost hybrid / PoH-only not throttling dig fleets / boost >100 escape hatch).
• Canary: HACKME_FUZZ_CLAIM_GHS_PRIORITY=0 first, then flipped to =1.
• Looking good in the wild: a few real hybrids online, dig-only soft-defer firing (ghs_priority_defer), fleet ETA capacity honest (live PoH + fuzz, not stale GH/s ghosts).

So the useful-PoW loop you described — real GH/s steering Dig/Hunt without fairy tales — is live.

Next when we pick it up: B (surface fleet_capacity / ETA in marketplace UI) is the natural follow-on. C–E stay on the roadmap as written — interest-driven, no rigid schedule.

Thanks for the local A/B + Hunt smoke proof too; that made the merge call easy.

[jokeez]

B update: marketplace now shows live fleet capacity + per-campaign eta_sec_fleet (campaign board + mining panel). API: /api/fuzz/pool/campaigns/list and /api/fuzz/marketplace include fleet_capacity. Finding HTML reports unchanged (no GH/s in artifacts). Shipped ff55b14 on main / hub.

[FounderB]

Awesome! Glad to hear the live metrics and canary tests are looking solid in the wild. Thanks for the quick review and deployment of A and B.

[FounderB]

FounderB follow-through on remaining #13 slices

Hub already shipped A (GHS↔Dig live) + B (marketplace fleet ETA) + most of C (Hunt Watch honesty 2.0 / finding_families). Thanks for that.

I took the leftover / adjacent pieces on FounderB:feat/ghs-dig-coupling:

Slice	Status
D customer-repo fail-closed	Negative gates: refuse Rust package stub without fuzz_target!; refuse C inventory without LLVMFuzzerTestOneInput when template_accept=false. Gate: bash scripts/tests/hunt_issue13_cde_gate.sh
E GPU Dig mutators	Scaffold internal/gpudig — accelerator may propose mutants; CPU WASM/ASAN eval only (explicitly not GPU ASAN). Opt-in dig_gpu_mutators.
Engine	Merged rc17.2 DeepHavocV28 + FounderB v2.9 weights; v2.10 burst is opt-in only (havoc_deep_v210) — measured unique can dip vs deep-v28 at fixed maxLen, so default Hunt stays deep-v28.

Local compare artifact: reports/hunt-marathon/20260926T104926Z-v210-vs-rc172/COMPARE.md (mutation A/B + ~12m libucl soak vs tip).

Happy to open a focused PR when you want D/E reviewed separately from the older GHS branch history.