Static analysis on today's run (fetchers blocked — known CCR network policy). Main finding: the is_significant_release() filter has coverage gaps for orgs that are already in KNOWN_ORGS/MAJOR_HF_ORGS but missing from significant_orgs.
Finding 1: significant_orgs gaps in is_significant_release()
These KNOWN_ORGS members are not in the significant_orgs list inside is_significant_release(), meaning their models only pass the filter if downloads >= 100_000 — a threshold that new releases often don't reach in the first 24–48 hours.
Verified via direct function test against models from the last 2 days of digests:
| Model |
Author |
is_significant_release result |
Why it fails |
cohere/command-a-plus |
cohere |
False (dl=50K simulated) |
Not in significant_orgs, not in significant_families |
inclusionai/ring-2.6-1t |
inclusionAI |
False (dl=80K simulated) |
Not in significant_orgs |
bytedance-research/lance |
bytedance-research |
False (dl=30K simulated) |
Not in significant_orgs |
sapientinc/hrm-text-1b |
sapientinc |
False |
Not in significant_orgs |
All four of these appeared in the May 21 digest — meaning the curator's manual pending-file approach caught them, but the auto-monitor fallback would have dropped them silently.
Impact: On days when no curator prepares a pending file, Cohere Command A+ (218B, Apache 2.0) and inclusionAI Ring-2.6-1T (1T param, MIT) would not appear in the channel. These are not marginal models.
Proposed fix (PR-level): Add to significant_orgs inside is_significant_release():
significant_orgs = [
...,
"cohere", # Command A+ and future Command series
"inclusionai", # Ring-2.6-1T and future releases
# bytedance-seed is already in KNOWN_ORGS; bytedance-research needs evaluation (see below)
]This is a PR-level change (logic modification to is_significant_release). Flagging here for Sov to review scope before I open a PR.
Finding 2: bytedance-research — new ByteDance HF org
bytedance-research/Lance (any-to-any multimodal, Apache 2.0) appeared in the May 21 digest. The HF org bytedance-research is distinct from bytedance-seed (which is already in KNOWN_ORGS).
- Only 1 model visible in current data → does not yet meet the supervisor's KNOWN_ORGS criteria (≥3 models with likes≥300, or ≥1 with likes≥1000).
- Will re-evaluate on next run when live data is available (or when the network allowlist is configured).
- Worth noting: if ByteDance continues releasing under
bytedance-research, it should eventually earn a KNOWN_ORGS entry.
Finding 3: sapientinc (Sapient Intelligence) — new architecture org to watch
sapientinc/HRM-Text-1B (non-transformer hierarchical recurrence model, Apache 2.0, trained for ~$1K) appeared in the May 21 digest. Novel architecture, not a fine-tune.
- Only 1 model visible → not yet KNOWN_ORGS eligible.
- Org is not in
significant_orgs or significant_families → auto-monitor blind to future releases.
- Watching for future releases; will surface if engagement criteria are met.
Production health
Digest quality is healthy — May 21 (9 models) and May 22 (8 models) both showed full tier coverage with editorial commentary. No 0-model posts, no missing tiers, no timing anomalies. The curator-prepared pending files are covering the auto-monitor gaps well.
Action requested
- Review Finding 1 — approve supervisor opening a PR to add
cohere and inclusionai to significant_orgs, or adjust the list to your preferred scope.
- Note Finding 2 —
bytedance-research is on the watch list; no action needed yet.
- Note Finding 3 —
sapientinc is on the watch list; no action needed yet.
Static analysis on today's run (fetchers blocked — known CCR network policy). Main finding: the
is_significant_release()filter has coverage gaps for orgs that are already inKNOWN_ORGS/MAJOR_HF_ORGSbut missing fromsignificant_orgs.Finding 1:
significant_orgsgaps inis_significant_release()These KNOWN_ORGS members are not in the
significant_orgslist insideis_significant_release(), meaning their models only pass the filter ifdownloads >= 100_000— a threshold that new releases often don't reach in the first 24–48 hours.Verified via direct function test against models from the last 2 days of digests:
is_significant_releaseresultcohere/command-a-pluscoheresignificant_orgs, not insignificant_familiesinclusionai/ring-2.6-1tinclusionAIsignificant_orgsbytedance-research/lancebytedance-researchsignificant_orgssapientinc/hrm-text-1bsapientincsignificant_orgsAll four of these appeared in the May 21 digest — meaning the curator's manual pending-file approach caught them, but the auto-monitor fallback would have dropped them silently.
Impact: On days when no curator prepares a pending file, Cohere Command A+ (218B, Apache 2.0) and inclusionAI Ring-2.6-1T (1T param, MIT) would not appear in the channel. These are not marginal models.
Proposed fix (PR-level): Add to
significant_orgsinsideis_significant_release():This is a PR-level change (logic modification to
is_significant_release). Flagging here for Sov to review scope before I open a PR.Finding 2:
bytedance-research— new ByteDance HF orgbytedance-research/Lance(any-to-any multimodal, Apache 2.0) appeared in the May 21 digest. The HF orgbytedance-researchis distinct frombytedance-seed(which is already in KNOWN_ORGS).bytedance-research, it should eventually earn a KNOWN_ORGS entry.Finding 3:
sapientinc(Sapient Intelligence) — new architecture org to watchsapientinc/HRM-Text-1B(non-transformer hierarchical recurrence model, Apache 2.0, trained for ~$1K) appeared in the May 21 digest. Novel architecture, not a fine-tune.significant_orgsorsignificant_families→ auto-monitor blind to future releases.Production health
Digest quality is healthy — May 21 (9 models) and May 22 (8 models) both showed full tier coverage with editorial commentary. No 0-model posts, no missing tiers, no timing anomalies. The curator-prepared pending files are covering the auto-monitor gaps well.
Action requested
cohereandinclusionaitosignificant_orgs, or adjust the list to your preferred scope.bytedance-researchis on the watch list; no action needed yet.sapientincis on the watch list; no action needed yet.