Context
On E-IV a sample record is located and sized by its E3S1 directory entry — a big-endian record-length field — because the record's own in-record length/pointer block is explicitly not usable on E-IV: the EIII extent rule (+34 plus a bias of two) matches the next-record distance on 0 of 522, 0 of 3893 and 0 of 934 consecutive pairs across the three E-IV discs, and no other offset survives all three (emu3.md, "The record's own length field is not usable"; ADR-0020).
The loop, however, is still decoded from that same in-record eight-pointer block. _eiv_samples in fs/emu3.py hands sample_pointers(window) to the shared _loops() in sample/emu3.py, which gates the loop on that block's own start (must open at DATA_START = 92) and end pointers. So on E-IV the loop is read from the one structure the format otherwise distrusts, and its end gate leans on a field the doc has shown does not size the record here — with the directory-derived frame count as the only reliable backstop.
What the evidence gap is
The loop-validation oracle in emu3.md ("Loop points") — the splice/join and shape correlation, with a wrong start at the same end as the control — confirms loops on the EIII/ESI discs. On E-IV it can lack the power to say anything, because E-IV loops are overwhelmingly the whole-extent "no loop" (bounds inset a fixed few bytes from frame 0 and the last frame), leaving almost no audio before the loop start or after the loop end for the windowed correlation.
That is not hypothetical on eiv-analogia: 441 of its 449 records pass every structural gate, but only 34 carry audio loud enough at both ends to score, and those 34 show nothing — a lack of power, not a refutation (ADR-0025 records this explicitly). After ADR-0030/D23 refused the whole-extent form at both ends, eiv-analogia drops from 449 loops to 6. Those 6 surviving loops are emitted entirely on the rule the other reference discs establish, with no independent per-record confirmation on the disc itself.
Not this, and why
- Not
eiv-vitous or eiv-studio. Their loops were validated by the shape test — at +0.68 and +0.86 — and are kept on that evidence (ADR-0025, ADR-0030). The gap is the disc where the oracle had no power, and the general risk it names for any future E-IV disc dominated by whole-extent loops.
- Not #44 (E-IV banks that are named, declare a length and read nothing — an extent/binding problem, upstream of any loop).
- Not #41 (the whole-extent "no loop" refused only at frame 0), which D23/ADR-0030 closed. That refusal is why only 6 loops remain to worry about here; it did not confirm them.
- Not the origin fix in #49. That corrected where E-IV records are found at a non-zero filesystem origin and does not touch the pointer decode.
What closing this would establish
An independent measurement with power on the residual E-IV loop population — the non-whole-extent loops eiv-analogia keeps after D23 — that says they splice where a control (the same end, a wrong start) does not, the way the EIII/ESI loops do. The doc already notes the forward-windowed oracle is the wrong instrument for near-whole-extent loops and that a period-based / pre-start window is what carries the discs it has no forward power on; the question is whether that (or another metric) can score eiv-analogia's survivors at all. If nothing can, the honest outcome is to state plainly that E-IV loop emission on that disc rests on borrowed evidence — which is currently recorded in ADR-0025 but not surfaced as a limitation the way the extent gaps are.
Context
On E-IV a sample record is located and sized by its
E3S1directory entry — a big-endian record-length field — because the record's own in-record length/pointer block is explicitly not usable on E-IV: the EIII extent rule (+34plus a bias of two) matches the next-record distance on 0 of 522, 0 of 3893 and 0 of 934 consecutive pairs across the three E-IV discs, and no other offset survives all three (emu3.md, "The record's own length field is not usable"; ADR-0020).The loop, however, is still decoded from that same in-record eight-pointer block.
_eiv_samplesin fs/emu3.py handssample_pointers(window)to the shared_loops()in sample/emu3.py, which gates the loop on that block's ownstart(must open atDATA_START = 92) andendpointers. So on E-IV the loop is read from the one structure the format otherwise distrusts, and itsendgate leans on a field the doc has shown does not size the record here — with the directory-derived frame count as the only reliable backstop.What the evidence gap is
The loop-validation oracle in emu3.md ("Loop points") — the splice/join and shape correlation, with a wrong start at the same end as the control — confirms loops on the EIII/ESI discs. On E-IV it can lack the power to say anything, because E-IV loops are overwhelmingly the whole-extent "no loop" (bounds inset a fixed few bytes from frame 0 and the last frame), leaving almost no audio before the loop start or after the loop end for the windowed correlation.
That is not hypothetical on
eiv-analogia: 441 of its 449 records pass every structural gate, but only 34 carry audio loud enough at both ends to score, and those 34 show nothing — a lack of power, not a refutation (ADR-0025 records this explicitly). After ADR-0030/D23 refused the whole-extent form at both ends,eiv-analogiadrops from 449 loops to 6. Those 6 surviving loops are emitted entirely on the rule the other reference discs establish, with no independent per-record confirmation on the disc itself.Not this, and why
eiv-vitousoreiv-studio. Their loops were validated by the shape test — at +0.68 and +0.86 — and are kept on that evidence (ADR-0025, ADR-0030). The gap is the disc where the oracle had no power, and the general risk it names for any future E-IV disc dominated by whole-extent loops.What closing this would establish
An independent measurement with power on the residual E-IV loop population — the non-whole-extent loops
eiv-analogiakeeps after D23 — that says they splice where a control (the same end, a wrong start) does not, the way the EIII/ESI loops do. The doc already notes the forward-windowed oracle is the wrong instrument for near-whole-extent loops and that a period-based / pre-start window is what carries the discs it has no forward power on; the question is whether that (or another metric) can scoreeiv-analogia's survivors at all. If nothing can, the honest outcome is to state plainly that E-IV loop emission on that disc rests on borrowed evidence — which is currently recorded in ADR-0025 but not surfaced as a limitation the way the extent gaps are.