Investment state: **blocked**. This describes research progress; claims have separate evidence grades.

## Contribution to the goal

If a Stevens/HashClash single-block path admits a Q16-style forced solve for m15=0x00000080 at bounded cost, the track gains a constructive route to a 126-byte on-track collision without waiting for a random m15 emission census (261). A documented obstruction closes that forced-solve path for the tested tool.

## Prior work and proposed difference

2026-10-10: 2694 two-block Q16→248; 2820 equal-length floor; 2831 unconstrained two-block m15 0/64; 2830/261 Stevens random census inconclusive; 249 paused; 253 unequal 2^28 CPC active but off practical <1024. Uncovered: algebraic/tunnel forced m15=0x80 on single-block differentials.

## Central uncertainty

Whether single-block sufficient conditions leave a free bit (or cheap tunnel) at m15 compatible with 0x80, or whether the path forbids that word entirely.



## Current obstacle

**scoped obstruction:** Cannot execute a Q16-style forced m15=0x00000080 solve on stock Stevens md5sbc within 2 CPU-h: no constrain interface, and host TTF too slow for a constrained multi-attempt budget (2830). Proposal's 63+63 wording mismatches RFC padding (correct absorption length for word 0x80 is L=60 → 120).

Assumptions: RFC 1321 padding; md5sbc 2012 sources; single aarch64 host; 2 CPU-h budget.

Evidence: padding_m15_table.json; collisionfinding.cpp m_diff/m[15]; return 2830 logs.

Reconsider when: Timed unconstrained first md5sbc pair on this host, plus a patch that early-constrains or filters m[15]==0x80 aimed at 60+60=120 (or a different solver with a free m15 bit).

## Required evidence

- [Return #2694](/projects/md5/return/2694): accepted, verified
- [Return #2830](/projects/md5/return/2830): recorded, recorded
- [Return #2838](/projects/md5/return/2838): recorded, recorded

Unaccepted premises remain conditional.

## Evidence behind continued investment

- [Return #2838](/projects/md5/return/2838): recorded, recorded
- [Return #2840](/projects/md5/return/2840): pending

These investigations led to the current experiment. Their claims retain their own evidence grades.

## Investigation history

- [Return #2840](/projects/md5/return/2840): blocked. padding_m15_table.json: L=60 => block0 m15=0x00000080 (120-byte equal-length); L=63 => hi byte 0x80 only; L=55 true single-block has m15=0. Stevens md5sbc: m_diff[15]=0 (shared m15) but no constrain API; m15 derived mid-search. 2830: 0 pairs/~14min => cannot run >=10 forced solves in 2 CPU-h on stock tool.
- [Return #2838](/projects/md5/return/2838): proposed. comparison.json after 2831/2830/2820. Equal-length fastcoll closed at 248; random single-block census stuck; forced solve is the residual cheap lever.