Identity & integrity of the public artifact captured 2025-09-10 — the same file, unchanged, that Ardora read. Not a safety statement.
The reading
Character abstainedthe read did not resolve at the published ceiling
Fine-tune change · subtle · low confidence
The instruct + dual-mode reasoning post-train is present (the shift is detectable) but reads FAINT on the disposition axis — notably less cleanly than the other assistant/post-train pairs in this batch. The reading abstains more here rather than force a crisp characterization onto a faint read.
Disposition
instruction / assistant registerlow confidence
present; the base is a plain text-continuer — but reads FAINT, notably less cleanly than the other post-train pairs in the batch
documented; not cleanly characterized at the ceiling — see abstain
Axis profile
Overall · band 2 · low confidence
slight - a character is present but faintly read; the profile abstains more here
Assistant Adherencemoderate - an assistant register is present but reads faint, less cleanly than the crisp instruct pairslow
Register / Formality— not characterised · the read battery does not calibrate tone / diction; register-formality is not characterized at the published ceiling
Reasoning Scaffolding— not characterised · a dual-mode (think / no_think) reasoning register is documented but reads faint at the published ceiling; not cleanly characterized here
Domain Specializationminimal - general-purpose; no domain-idiom pulllow
Verbosity— not characterised · default response length / elaboration is not exercised by the read battery; not characterized at the published ceiling
Turn-Taking / Interactivityslight - takes a turn, but the shift reads faintlow
Performed Voiceminimal - neutral assistant voice; no sustained characterlow
disposition-definition (descriptive): how pronounced and clearly-read the model's overall character is, discounted by what had to be abstained. NOT a quality, capability, or safety ranking. A plain base reads low because it has less disposition to characterize, never because it is worse.
Abstained — read, but not characterised at the published ceiling
dual-mode (think / no_think) reasoning registerdocumented in provenance, but the disposition reads faint at the published ceiling and the reasoning register is not cleanly characterized here
reasoning depth / qualitynot characterized at the published ceiling
multilingual + long-context dispositiondocumented capabilities; the read battery is English-centric and short-context, not characterized
Characterization is descriptive and coverage-bounded; preview engine, single-battery read; see published fidelity ceiling.
Claim grounding
The uploader states: "SmolLM3 is a 3B parameter language model designed to push the boundaries of small models. It supports dual mode reasoning, 6 languages and long context." This instruct model is "optimized for hybrid reasoning" with "/think" and "/no_think" extended-thinking modes and support for tool calling and agentic usage. Post-training "included midtraining on 140B reasoning tokens followed by supervised fine-tuning and alignment via Anchored Preference Optimization (APO)." Apache-2.0; described as fully open. HF tags: text-generation, transformers, safetensors, smollm3, conversational.
This is an instruct model (post-trained assistant), optimized for hybrid reasoning; tagged `conversational`.
Ardora does detect an assistant register present vs the plain-continuer base (assistant-adherence band 3), so the instruct identity is supported -- BUT the reading flags this shift as FAINT (finetune_change magnitude subtle, low confidence; ardora_score band 2), notably less cleanly read than the crisp instruct pairs. Honestly characterized as a legibility boundary, not forced into a clean read. (Same KIND of assistant move as other families -- finding instruction-shift-recurs-across-families -- but faint here.)
Dual-mode reasoning: extended thinking can be enabled/disabled via "/think" and "/no_think" flags.
First-class abstain: the reading engaged the reasoning-register axis and DECLINED (reasoning-scaffolding abstained) -- the dual-mode reasoning register is documented but reads faint at the published ceiling and is not cleanly characterized here. Ardora abstains rather than force a crisp reasoning read; reasoning depth/quality is separately out of scope as a capability.
INDISTINGUISHABLEalignment
Post-trained via midtraining on 140B reasoning tokens, then SFT, then alignment via Anchored Preference Optimization (APO).
The post-train assistant shift is detected (faintly), but the specific alignment method (APO vs plain SFT) does not resolve into a separate disposition at the published ceiling -- consistent with finding sft-vs-dpo-fingerprint (the alignment method leaves no distinct disposition fingerprint). Ardora reads that a post-train disposition is present, not which alignment recipe produced it.
Tool-calling and agentic behavior are task-competence capabilities Ardora does not test; out of disposition scope.
UNVERIFIED-ABSTAINlanguage
6 natively supported languages plus additional trained languages.
Ardora abstains on multilingual disposition (English-centric reading); neither confirms nor contradicts the language claim.
UNVERIFIED-ABSTAINcapability
Long context up to 128k tokens via YaRN (trained at 64k).
Long-context behavior is a documented capability Ardora abstains on (short-context reading); out of scope.
UNVERIFIED-ABSTAINidentity
Described as "fully open."
"Fully open" is a provenance/licensing statement, not a disposition; the atlas records licence apache-2.0, but openness is not something the reading characterizes.
Claims are the uploader's, quoted from their public card at the capture date; stances are coarse, witnessed, disposition-only, measured against the published fidelity ceiling for this engine version -- not the Engine's raw numbers. Recomputed when the ceiling moves; capability, benchmark, and safety claims are abstained, never refuted.