Head to head
Same brief, same rubric, same scale. Put up to three models side by side: dimension scores, how many of their claims held up, and which issues each review raised.
Right about the chip, wrong about the CPU.
Same brief, with UNIONALPHA’s findings handed over as a baseline, which the review acknowledges.
B UNIONALPHAAccurate to the line number. Silent on what stops the radio decoding.
Blind. The model’s identity was withheld from the auditor; the repository’s earlier reviews were in the working tree.
More useful, less careful
UNIONALPHA is the more careful review: almost every claim holds, but it spends them on tooling. DeepSeek finds the chip-side problems UNIONALPHA missed and proposes the best experiment either offers, but makes more mistakes, and one of them clears a real blocker. Neither found the vocoder CPU wall or the circular RRC test oracle.
From the DeepSeek V4.1 Flash audit, 17 September 2026.
Weighted score
Out of 100, on the grade scale.
Earned and lost, by rubric slot
Each bar is a model’s 100-point grade split into the rubric’s weighted slots; filled is earned, empty is lost. The slots line up, so a gap in one bar against a full slot in another is exactly where the models part ways.
- 1Accuracy & evidence30 pts
- 2Coverage of decode problems25 pts
- 3Root cause & prioritisation15 pts
- 4Fix plan & acceptance gates15 pts
- 5Originality & attribution10 pts
- 6Clarity & calibration5 pts
Scores on each rubric dimension
Out of 100 per dimension; the weight is that dimension’s share of the grade. Hover or focus a row for the exact scores.
- ADeepSeek V4.1 Flash
- BUNIONALPHA
Scores and weighted points
| Dimension | Weight | A · DeepSeek V4.1 Flash | B · UNIONALPHA | A − B |
|---|---|---|---|---|
| Accuracy & evidence | 30% | 68 · 20.4 pts | 90 · 27.0 pts | −22 |
| Coverage of decode problems | 25% | 66 · 16.5 pts | 45 · 11.2 pts | +21 |
| Root cause & prioritisation | 15% | 64 · 9.6 pts | 55 · 8.2 pts | +9 |
| Fix plan & acceptance gates | 15% | 74 · 11.1 pts | 74 · 11.1 pts | ±0 |
| Originality & attribution | 10% | 70 · 7.0 pts | 40 · 4.0 pts | +30 |
| Clarity & calibration | 5% | 68 · 3.4 pts | 76 · 3.8 pts | −8 |
| Weighted total | 100% | 68 · C− | 65 · C− | +3 |
Every claim, checked
One cell per claim in each audit’s claim table, checked at the lines it cites. Hover a cell for the claim and the auditor’s note. The reviews made different numbers of claims, so the bar under each grid gives the shares.
DeepSeek V4.1 Flash’s claim table · UNIONALPHA’s claim table
- Holds
- Qualifiedoverstated, miscounted or doubtful
- Wrong
17 hold · 8 qualified · 5 wrong · 57% hold
21 hold · 2 qualified · 0 wrong · 91% hold
What each review raised
DeepSeek V4.1 Flash found 3 of the 4 decode-critical issues; UNIONALPHA found 0 of the 4 decode-critical issues. Every tracked issue below, with the reviews that were already in the repository for reference.
Issue signatures
Each review as a trace across the tracked issues. Where traces part is where the reviews differ; hover a column for every review’s entry. Column numbers match the table below.
- raised
- partly
- absent
- Decode-critical
| # | Issue | P25-HARDWARE-REVIEWEarlier repo review · Sep 8 | P25-PROJECT-REVIEWEarlier repo review · Sep 10 | UNIONALPHAModel · Sep 16 | DeepSeek V4.1 FlashModel · Sep 16–17 |
|---|---|---|---|---|---|
| 1 | Capture export stages the wrong I2S halfwords | absent | raisedF1 | raisedF1 | raisedB5 |
| 2 | Capture parser certifies an incomplete stream | absent | raisedF2 | raisedF2 | partlyin its plan |
| 3 | pcm_starve never increments | partlyasks for underrun counts | raisedF3 | raisedF3 | raisedB9 |
| 4 | No static RAM margin | partly648 bytes then | raisedF4 | raisedF4 | raisedB6 |
| 5 | The I2S stream is most likely microphone audio Decode-critical | raisedcites sound.c | raised | partly“either direction” | raisedB1 |
| 6 | HR-C6000 de-emphasis on the capture path Decode-critical | partlycode comment | partlycode comment | absent | raisedB2, register 0x34 |
| 7 | AT1846S FM filters, low-frequency bit, 25 kHz Decode-critical | raised | raised | absent | partlyevidence file only |
| 8 | 0x10=0x6E hybrid state; 0x36 dual role | absent | absent | absent | raisednew |
| 9 | Manual: I2S frame clock “must be 8KHz” Decode-critical | absent | absent | absent | raisednew |
| 10 | One-layer 4FSK test mode as a P25 tap | absent | absent | absent | raisednew |
| 11 | ±10% health gate versus ±1% timing clamp | absent | absent | raisedC2 | raisedB3, executed |
| 12 | Fail-closed muting at LDU cadence | absent | partlylate entry | absent“keep fail-closed” | raisedB8 |
| 13 | Non-standard MFID mutes clear calls | absent | absent | absent | raisedB9 |
| 14 | Test waveform shares the receiver’s RRC filter missed | raised | raisedF6 | absent | absentreuses it |
| 15 | MCU runs at 72 MHz | raised | raised | absent | partlyPhase 2 gate only |
| 16 | Vocoder needs 11–16× the 72 MHz CPU Decode-criticalmissed | absent | partlycycles per frame | partly“to measure” | absent“not the blocker” |
| 17 | Direct discriminator tap (M17 mod) | raisedwith pins | raised | partlyno pins | raisedwith pins |
| 18 | Phase 2 architecture and scope | partly | partly | raised | raisedwith errors |
From the provenance table in the DeepSeek V4.1 Flash audit (17 Sep 2026), the newest audit on this task. Shaded columns are reviews that were already in the repository.