INLINE write path -- runtime tested empirically (single-block scope)

First empirical validation of the BEAMFS INLINE write path. Until
this point, INLINE was read-only (Palier 3 recovery on the canary
fixture, validated 2026-04-29 morning session). This tag marks the
moment where INLINE write_begin / write_end / writepages produce a
durable byte-perfect on-disk encoding.

Test environment:
- Yocto qemuarm64 image hpc-arm64-research-beamfs (rebuilt with vim
  added to IMAGE_INSTALL for xxd availability)
- Kernel linux-mainline 7.0.1, beamfs-module from develop branch
- Test target: /var/volatile/beamfs-test (tmpfs RW workaround;
  full BEAMFS rootfs is the principal long-term objective per
  Documentation/roadmap.md)

Empirical results (10 phases, all PASS):
- Phase 1: mkfs.beamfs -s inline produces scheme=2 UNIVERSAL_INLINE
  volume with 40-byte journal entry, canary block 19 inode 2
- Phase 3: canary fixture SHA256 = ced446d9... (cross-platform
  reproducibility validated, Palier 3 sanity preserved)
- Phase 4: write 68-byte payload to /mnt/test.txt -- rc=0, sync rc=0
- Phase 5: stat inode 3 (distinct from canary inode 2), size 69
  (payload + newline), regular file mode 0644, durable timestamps
- Phase 6: cache read-back identical to payload
- Phase 7: drop_caches + read-back from disk SHA256 = 705aa689...
- Phase 8: dmesg clean, zero RS correction events (nothing to
  correct -- the encode produced a valid codeword), zero
  -EOPNOTSUPP, zero kernel BUG/WARN
- Phase 9 (durability): umount + remount + re-read, SHA256
  byte-identical post-remount = on-disk persistence verified
- Phase 10: cleanup clean (umount + losetup -d + rmmod)

Single-block scope (v2.0): files <= 3824 user bytes (one INLINE
disk block). write_begin rejects pos+len > 3824 with -EFBIG.
Multi-block INLINE write is part of the v2.x roadmap, prerequisite
for full POSIX coverage and ultimately for full BEAMFS rootfs (see
Documentation/roadmap.md, Principal long-term objective).

Branch: develop (release candidate). Will be merged to main with
v1.3.0 stable tag once xfstests subset (generic/001..010 + quick
group) passes and the saturation falsification protocol for
Theorem v2.2 is exercised.

Commit chain on develop:
- 4779581 wip(inline-write): initial implementation (had
  filemap_get_folios_tag type bug)
- 0fcd236 doc(roadmap): full BEAMFS rootfs as principal objective
- 39936fb chore(funding): PayPal sponsor button (cherry-picked
  to main as 4d1886a)
- efa064b fix(inline-write): writepages simplified single-block,
  removed unused inode in write_begin

WOW objective tracking: this is item 2 of the dependency chain
toward full BEAMFS rootfs + RadFI head-to-head bench. Items 3-9
(multi-block, full POSIX, xfstests, symlinks, xattr, dm-verity,
IMAGE_FSTYPES=beamfs) remain ahead.

Cross-DOI update -- both papers reference each other via Zenodo DOIs

After publication of both papers on Zenodo on 2026-04-29:
- BEAMFS v2 -> 10.5281/zenodo.19886192
- RadFI v1  -> 10.5281/zenodo.19885777
- FTRFS v1  -> 10.5281/zenodo.19824442 (lineage)

The bibliographic cross-references in both papers' refs.bib are
updated to use the permanent Zenodo DOIs instead of GitHub tags.
PDFs rebuilt and committed.

Antériorité scientifique now fully cross-linked through CERN-archived
DOIs in addition to GPG-signed GitHub tags.

RadFI v1 paper LaTeX port complete

Companion paper to BEAMFS v2 published as papers/2026-04-radfi-v1/.
PDF: aurelien-desbrieres-radfi-v1-20260429.pdf (9 pages, IEEEtran).

Theorem IV.1 (RadFI faithfulness as SEU emulator) proved via
Le Cam's binomial-to-Poisson total variation bound.

Empirical falsification of BEAMFS v1 Theorem IV.1 (2026-04-28),
empirical corroboration of BEAMFS v2 INLINE recovery (2026-04-29).

Companion to: roastercode/beamfs at v1.0.0-paper-zenodo-pending (52260d0).
RadFI implementation: roastercode/radfi (private, pending Zenodo upload).

Pending: Zenodo upload (both papers same day), DOI assignment,
cross-DOI update, retag v1.1.x-paper-zenodo-published.

BEAMFS v2 paper draft ready for Zenodo upload

Technical Report v2 published as papers/2026-04-beamfs-v2/.
PDF: aurelien-desbrieres-beamfs-v2-20260429.pdf (~13 pages, IEEEtran).

Theorem v2.1 + v2.2 replacing falsified Theorem IV.1.
EMR threat model. Stage 3 + Stage 4 empirical results.
2198 LOC kernel module (44% of 5000 audit budget).

Implementation: roastercode/beamfs at v0.4.0-palier4-validated (7c30934).
Companion paper: RadFI v1 at v0.3.0-radfi-paper-public (eff67b7).

Pending: Zenodo upload, DOI assignment, paper update with DOI.

BEAMFS v1: code + paper draft, ready for Zenodo finalization 2026-04-29

Palier 4 validated: perf evaluation + LOC audit + B3 INLINE sanity

LOC audit: kernel module 2198 LOC (44% of 5000 budget).
B1 read cold (10 runs): BEAMFS scheme=5 vs ext4 parity on p50/p95.
B2 write+fsync (10 runs): BEAMFS scheme=5 ~8x ext4 (RS encoding cost),
  tighter latency distribution (1.74x p50/p99 spread vs 5.00x ext4).
B3 INLINE read (10 runs, ~380k samples): p50 23.4 us, parity with B1
  and ext4. Zero RS correction events. Canary fixture stable.

Trade-off: BEAMFS trades write performance for autonomic EM-resilience
recovery. Read parity on median. Target deployments (OIV, aerospace,
defense) typically read-dominant.

INLINE write path (-EOPNOTSUPP) deferred to v2.x roadmap (post-Zenodo).

RadFI v1 methodology paper public

Companion to BEAMFS v0.2.0-palier3-validated. Establishes public
anteriorite scientifique for RadFI methodology (code remains
private pending peer review).

BEAMFS v0.2.0 -- palier 3 empirically validated

Milestone:
  - v2 INLINE read path complete
  - format-v4 spec normative with conformance fixture
  - EMR-first scientific framing
  - Empirical palier 3 victory: byte flip -> RS decode -> autonomic
    repair byte-perfect (cf. session 2026-04-29)

Status: paper-pending. Zenodo BEAMFS v2 paper to be published next.

Pre-fs-devel: write path + xfstests + LOC audit + cover letter remain.