Model review audits
Compare · Silent Carrier

Head to head

Same brief, same rubric, same scale. Every model on the task side by side, or just the ones you pick: dimension scores, how many of their claims held up, and which issues each review raised.

Models 2 of 7
All 7 models →
UNIONALPHA F D C B A C 74 / 100
rank 1 of 7 · graded 17 Sep 2026

Accurate to the byte. Hesitant about what stops decoding.

 Qwen3.8 Max F D C B A D 63 / 100
rank 5 of 7 · graded 17 Sep 2026

Best case yet for the microphone. Never looks at the CPU.

Every model here ran under the same conditions. How the runs were set up →

From the audits

Deepest on the source, thinnest on the tools

Qwen3.8 Max and Grok 4.6 make the same microphone argument, and Qwen3.8 Max makes it in more detail and adds the stop rule. It also spent five hours and 598K output tokens to do it, and returned nothing about the capture, the parser, starvation or memory, where UNIONALPHA and DeepSeek V4.1 Flash found real defects.

From the Qwen3.8 Max audit, 17 September 2026.

Weighted score

Out of 100, on the grade scale.

1 UNIONALPHAStealth model · graded 17 Sep 2026 74C 5 Qwen3.8 MaxAlibaba · graded 17 Sep 2026 63D
Where the points went

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.

UNIONALPHA
74/ 100
Qwen3.8 Max
63/ 100
  • 1Accuracy & evidence30 pts
  • 2Coverage of decode problems25 pts
  • 3Root cause & prioritisation15 pts
  • 4Fix plan & acceptance gates15 pts
  • 5Originality & attribution10 pts
  • 6Clarity & calibration5 pts
Dimension profile

Scores on each rubric dimension

One panel per model: its score on each rubric dimension and the weighted total over the grade zones, in its own colour, with every other model on this task as a grey dot for scale. Hover a row for every model’s score and the auditor’s reasoning.

  • The panel’s model, joined down the dimensions
  • Every other model on this task

UNIONALPHA74 C

AccuracyCoverageRoot causeFix planOriginalityClarityTotalFDCBA AccuracyCoverageRoot causeFix planOriginalityClarityTotalFDCBA

Qwen3.8 Max63 D

AccuracyCoverageRoot causeFix planOriginalityClarityTotalFDCBA AccuracyCoverageRoot causeFix planOriginalityClarityTotalFDCBA
Scores and weighted points
Dimension Weight UNIONALPHAQwen3.8 MaxGrok 4.6Muse Spark 1.3 ContributorDeepSeek V4.1 FlashQwen3.8 FlashGLM 5.3 Flash Difference
Accuracy & evidence 30% 90 · 27.0 pts 70 · 21.0 pts 74 · 22.2 pts 74 · 22.2 pts 73 · 21.9 pts 62 · 18.6 pts 64 · 19.2 pts +20
Coverage of decode problems 25% 65 · 16.3 pts 39 · 9.8 pts 59 · 14.8 pts 54 · 13.5 pts 49 · 12.3 pts 40 · 10.0 pts 34 · 8.5 pts +26
Root cause & prioritisation 15% 60 · 9.0 pts 71 · 10.7 pts 70 · 10.5 pts 72 · 10.8 pts 66 · 9.9 pts 68 · 10.2 pts 62 · 9.3 pts −11
Fix plan & acceptance gates 15% 70 · 10.5 pts 71 · 10.7 pts 75 · 11.3 pts 71 · 10.7 pts 70 · 10.5 pts 70 · 10.5 pts 62 · 9.3 pts −1
Originality & attribution 10% 76 · 7.6 pts 66 · 6.6 pts 72 · 7.2 pts 70 · 7.0 pts 74 · 7.4 pts 64 · 6.4 pts 58 · 5.8 pts +10
Clarity & calibration 5% 80 · 4.0 pts 78 · 3.9 pts 78 · 3.9 pts 80 · 4.0 pts 74 · 3.7 pts 70 · 3.5 pts 68 · 3.4 pts +2
Weighted total 100% 74 · C63 · D70 · C68 · C−66 · C−59 · D56 · D +11
Claim check

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.

UNIONALPHA’s claim table · Qwen3.8 Max’s claim table

  • Holds
  • Qualifiedoverstated, miscounted or doubtful
  • Wrong
UNIONALPHA35 claims checked

33 hold · 2 qualified · 0 wrong · 94% hold

Qwen3.8 Max27 claims checked

21 hold · 4 qualified · 2 wrong · 78% hold

Issue coverage

What each review raised

UNIONALPHA found 1 of the 4 decode-critical issues; Qwen3.8 Max found 2 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.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Project docs earlier repo review · in the repository UNIONALPHA model · Sep 17 Qwen3.8 Max model · Sep 17
  • raised
  • partly
  • absent
  • Decode-critical
Issues raised by each review of P25 voice decoding on the DM-1701
# Issue Project docsEarlier repo review · in the repository UNIONALPHAModel · Sep 17 Qwen3.8 MaxModel · Sep 17
1 Capture export stages the wrong I2S halfwords absent raisedF3, fixed and verified absentcalls the capture validated
2 Capture parser certifies an incomplete stream absent absent absentrelies on the parser
3 pcm_starve never increments absentdocumented as working raisedF7 absent
4 No static RAM margin partlymargins still to measure raisedF6, byte-exact absent
5 The I2S stream is most likely microphone audio Decode-critical partlyopen, leaning sceptical partlyblocking, but never says microphone raisedA, the clearest case yet
6 HR-C6000 de-emphasis on the capture path Decode-critical absent partlycited, called unmeasured raisedC, with the 0x34 bit 5 fix
7 AT1846S FM filters, low-frequency bit, 25 kHz Decode-criticalmissed partly“require characterization” partlycited, called unmeasured absent
8 0x10=0x6E hybrid state; 0x36 dual role partlybring-up clock rules partlyF1, 0x6E against 0x80 partlyslot engine and eco only
9 Manual: I2S frame clock “must be 8KHz” Decode-criticalmissed absent raisedF4 absentcites the section, not the rule
10 One-layer 4FSK test mode as a P25 tap partlystock BER-test block only partlyworth a bounded test raisedstep 2, a symbol source
11 ±10% health gate versus ±1% timing clamp absent absent absent
12 Fail-closed muting at LDU cadence absent absent absent
13 Non-standard MFID mutes clear calls absent absent absent
14 Test waveform shares the receiver’s RRC filter partly“synthetic RRC/AWGN” caveat raisedF5, unquantified partlysynthetic only, not the circularity
15 MCU runs at 72 MHz raised raisedF7 absent
16 Vocoder needs 11–16× the 72 MHz CPU Decode-criticalmissed partlydecode timing unmeasured partlydeadlines “unproven” absent“not the problem”
17 Direct discriminator tap (M17 mod) raised raisedfallback raisedstep 5, the likely answer
18 Phase 2 architecture and scope partlynot implemented raisedmost accurate section raisedAMBE+2 present but uncalled
19 Two unverified SPI writes per decoded 20 ms frame absent absent absent
20 Capture sessions lack epochs absent raisedF9, 8 kHz under a 24 kHz header absent
21 Ring and tick real-time budget partlydeadlines unproven raisedF7, 1 ms is a minimum absent
22 Clock config 3 assumes 12,288 Hz; the codec formula gives 12,000 absent absent absent
23 Clock-config writes bypass the verified SPI writer partlySPI retry note raisedF9 absent
24 Stock squelch re-arms FM audio (0x10=0x80) during monitoring absentassumes it can’t re-arm raisedF1, new partlynames the squelch path as a writer
25 Stale clear-call state releases a new call’s first frames absent raisedF8, probe absent

From the provenance table in the Qwen3.8 Max audit (17 Sep 2026), the newest audit on this task. Shaded columns are reviews that were already in the repository.

Run facts

How each run went

Descriptive, not graded, and in grade order rather than ranked: one run per model. Run time is wall-clock, so it includes the provider’s speed, rate-limit waits and retries, and token counts vary with each model’s tokenizer. Every run followed Clean Room 1.0, which is what makes them comparable.

Run facts for each compared model, in grade order. Descriptive only: not graded or ranked.
Model Run time Agent steps Tool calls Output tokens Route Launch
UNIONALPHAThird attempt; the first two stopped on provider rate limits 36 min 83 95 63K OpenRouter API one-shot
Qwen3.8 Max 5 h 10 min 71 111 598K Alibaba Cloud Token Plan interactive
Every run figure
Model Started (UTC) Output tokens Reasoning tokens Context tokens Messages Sessions Retry budget
UNIONALPHA 17 Sep 2026, 12:57 63,278 not reported 9,988,183 179 1 10
Qwen3.8 Max 17 Sep 2026, 14:52 597,788 582,689 12,033,596 183 1 10

Context tokens are input plus cache reads and writes: mostly the conversation re-sent at every step, so they grow with steps × context size. The retry budget is how many API retries the harness allowed.