Inside Sika Agent
Right, per task and language
Share of matched episodes the agent got right; ▲ = episodes that harmed the user. Same seeds, wallets and expected outcomes in every column. Flips: twins right in English but not here / right here but not in English.
Read-outs
Share of reasoning words tagged English, Twi or mixed, by request language.
Mean vertical score of the top-20 receiver heads, per sentence, by source and request language (first turns). Right: Twi–English and Twi minus the English twin, 95% bootstrap interval over twins.
Accuracy reading held-out seeds at the end of the request, by layer. Dashed: bag of words on the request text.
20 classes; chance 5%. Dashed: bag of words.
Share of Twi–English requests whose state is nearer the English twin's than the Twi twin's (centred cosine), by layer.
Mean placement margin (cos to English twin − cos to Twi twin, centred within task type) of failed minus right episodes. Below 0: failures sit nearer the Twi twin. ● p < 0.001 (within-type permutation).
English-trained scam probe at the end of the request, as written and with words masked (…), by layer.
Token probe trained on English vs Twi requests; mean over a word's tokens, by layer. Mixed: English stem in Twi grammar (a-send-i).
1 − cosine between consecutive tokens' residual states in Twi–English requests: at a language switch vs elsewhere.
Best layer at the last reasoning token. Scam base rate 65%.
Steering at every decision
Each cell: one neuron's write direction added at the positions where the model chooses its move (± dose × residual norm), at one decision. Colour: shift of the margin toward the right move (lime) or the wrong/harmful one (coral), in nats; or its percentile among 32 random directions at the same layer, decision, sign and dose. Select a cell, a circuit node or a point on the map; the inspector follows.
Whole-layer sweep
Every MLP neuron of one layer pushed at the first move of 10 tasks × 4 seeds × 3 languages (dose 0.4, both signs). Per neuron: share of decisions where its push beats 97% of the 32 random directions measured at that layer and decision (better sign). By chance about 3%. Lime: neurons on the map, picked by their scores.
Steered replays
Episodes replayed from a decision with one direction added after its layer at every token for the rest of the episode (dose × residual norm), against the same simulator and scripted user; graded like the original episodes. Unsteered: the same replays with nothing added.
Outcomes
Replays ending right / harmful, per push and request language; Δ against unsteered in percentage points. Random: a random direction at the same layer. Select a row to compare its transcripts.
Thought anchors
Filled ring: model turn, size = change in mean outcome replaying from it vs from the next turn (10 replays per turn); colour = what it did. Hollow ring: input, size = largest KL its removal causes in a later turn. Arcs: step → later turn, width = mean KL over that turn's tokens. Select a step; the inspector shows its text shaded by receiver-head score.