docs: session 73 — record sessions 70-72 in CHANGELOG + README

Mirrors the post-v0.8.3 audit-coverage extension into the
canonical audit-pack documentation surface.

CHANGELOG.md
- New [Unreleased] subsection "Added — sessions 70-72" with
  per-session breakdowns:
  * Session 70 — audit-coverage runbook (new
    `docs/audit-coverage-runbook.md` for external review firms).
  * Session 71 — AUDIT.md release entries for v0.8.2 + v0.8.3,
    top-of-file pointer to the runbook.
  * Session 72 — `powers_of` 9th shared primitive + 5 proptest
    + first consumer migration in mosaic-hyperplonk::kzg.
- Running tally: shared-primitive count now 9 (was 8 at v0.8.3
  release).

README.md
- Property-test coverage row updated:
  * Session range bumped 36-66 → 36-72.
  * Lib test total 544 → 549 (+5 powers_of proptests in session 72).
  * "+147 proptest + shared-primitive coverage" → "+152 proptest +
    9 shared primitives lifted" reflecting the running primitive
    count.
- New row "Audit runbook" linking the session-70 runbook so
  reviewers find it from the README Status table directly,
  without going through AUDIT.md.

No code changes — documentation pass only, completes the audit-pack
mirror of sessions 70-72.

Author: kh0ra

CHANGELOG.md

@@ -7,6 +7,46 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Added — sessions 70-72 (post-v0.8.3)
+
+#### Session 70 — audit-coverage runbook
+`docs/audit-coverage-runbook.md` — entry point for an external
+review firm that wants to reproduce the Mosaic audit-coverage
+matrix locally and extend it with their own tests. Covers:
+- Coverage matrix at v0.8.3 with per-surface session ranges.
+- Local-reproduce recipes (property tests, BPF CU bench, host
+  criterion bench, fuzz harnesses).
+- Extension recipes (add a new property test, fuzz harness,
+  bench, shared primitive).
+- Three explicit caveats about what the coverage does NOT pin
+  (real prover output, full Phase-3 cryptographic soundness,
+  chunked dispatch integration).
+
+#### Session 71 — AUDIT.md release entries
+Recorded both v0.8.2 and v0.8.3 release milestones in AUDIT.md
+with per-release scope, findings, and lib-test counts. Top-of-
+file pointer to `docs/audit-coverage-runbook.md` so external
+reviewers find the local-reproduce workflow immediately.
+
+#### Session 72 — `powers_of` 9th shared primitive + first consumer
+`mosaic_zk_primitives::field::powers_of` lifts the multi-poly
+batched-opening ν-powers accumulator loop that surfaces in
+HyperPlonk session 3e, Halo2 session 17, and Nova Spartan
+session 22.
+
++5 proptest tests (mosaic-zk-primitives lib total 74 → 79):
+- prop_powers_of_length / prop_powers_of_first_is_one /
+  prop_powers_of_recurrence / prop_powers_of_one_is_all_ones
+- prop_powers_of_matches_pow — closed-form cross-check against
+  `fr_pow_u64`. The soundness invariant that justifies the lift.
+
+First consumer migration: `mosaic_hyperplonk::kzg` replaces an
+inline 11-step accumulator loop with `powers_of(&nu, 12)`. 82
+hyperplonk lib tests still pass; byte-identical refactor.
+
+After session 72 the shared-primitive count is **9 helpers** in
+mosaic-zk-primitives.
+
 ### Planned beyond v0.8.3-shared-primitive-lift
 
 - Fixture-driven differential testing for the four Phase-3 bodies

README.md

@@ -56,7 +56,8 @@ frozen CU budgets.
 | `mosaic-chunked` instruction handlers | ✅ Production | — |
 | Reference Solana program | ✅ 319 KB SBF ELF (30.4% of 1 MB cap; 12 cryptographic gates wired) | — |
 | Differential test harness (arkworks + snarkjs fixture) | ✅ Production (Groth16 + PLONK; Phase-3 extension tracked) | — |
-| Property-test coverage (proptest, sessions 36-66) | ✅ 544 lib tests across 12 crates (+147 proptest + shared-primitive coverage in audit-coverage sweep) | — |
+| Property-test coverage (proptest, sessions 36-72) | ✅ 549 lib tests across 12 crates (+152 proptest + 9 shared primitives lifted in audit-coverage sweep) | — |
+| Audit runbook | ✅ [`docs/audit-coverage-runbook.md`](docs/audit-coverage-runbook.md) — reproduce + extend recipes for external review firms | — |
 | BPF CU regression bench (`bpf-bench`) | ✅ 7 systems: Groth16 (single + batch), KZG-PLONK, HyperPlonk, Halo2, Nova, FRI-STARK | — |
 | Host criterion bench (wall-clock baseline) | ✅ 5 systems: Groth16, HyperPlonk, Halo2, Nova, FRI-STARK | — |
 | Fuzz harnesses (sessions 54-59) | ✅ 23 targets across all 6 production verifiers (5 proof-bytes + 5 vk-bytes + 5 public-inputs + 5 combined-slot for PLONK + 4 Phase-3, plus 3 original Groth16) | — |