codeburn models shows raw ids (gpt-5.6-*, accounts/fireworks/models/*) and duplicate rows that report resolves correctly

[therickfactr]

Problem

On 0.9.20, codeburn models and codeburn report disagree about the same models. report resolves friendly names; models shows raw ids for some families and also emits duplicate rows.

Evidence (0.9.20, npm)

Same three models, same run:

codeburn models	codeburn report
gpt-5.6-sol	GPT-5.6 Sol
gpt-5.6-terra	GPT-5.6 Terra
accounts/fireworks/models/kimi-k2p6	Kimi K2.6

report also resolves gpt-5.6-luna → GPT-5.6 Luna, which models does not surface under a friendly name either.

This is not a missing-names problem — the names clearly exist, since report produces them from the same data in the same binary.

It is also not "models deliberately shows raw ids". The same table mixes both conventions: GLM-5.2, Kimi K2.7 Code, DeepSeek v4 Pro, Qwen 3.7 Plus and Grok 4.5 all render friendly, while the gpt-5.6-* family and the full accounts/fireworks/models/* path form stay raw. So the resolution path runs in models, but misses those two shapes.

Second symptom, same view: duplicate rows

codeburn models lists Cline CLI | GLM-5.2 twice:

│ Cline CLI  │ GLM-5.2   │ Testing (75%)      │ 867.3K │ $0.246 │
│ Cline CLI  │ GLM-5.2   │ Conversation (43%) │  15.7K │ $0.019 │

Two distinct raw ids collapse to one display name and are not merged, so one model is reported as two, splitting its tokens and cost across rows. Reproducible with:

codeburn models | awk -F'│' 'NF>3 {gsub(/^ +| +$/,"",$2); gsub(/^ +| +$/,"",$3); print $2" | "$3}' | sort | uniq -d

That row-merge already exists — aggregateModelTotals(), keyed by display name, added in #787 for the By Model panel. It looks like models does not go through it.

Why it matters

codeburn models is the natural place to look up an id before running codeburn model-alias, and it is the view where a raw id is most likely to be copied into a command or a report. Two rows for one model also understates that model's real share when scanning the list by eye.

Suggested direction

Route the models command through the same name resolution and the same aggregateModelTotals() merge that report uses, rather than adding more entries to the names table — the entries are already there.

Happy to send the PR if you'd like it; wanted to check the approach first per CONTRIBUTING.

Environment: codeburn 0.9.20 (npm), macOS 26.6.1 (25G76) arm64, Node v26.5.0.

[iamtoruk]

Confirmed on both counts, and thank you for the precise diagnosis. report resolves through the global short-name table, while models asks each provider, and several providers fall back to the raw id when their local table misses. The duplicate rows are the same root: buckets are keyed by raw id and display names resolve after bucketing with no merge.

One adjustment to the suggested direction: models should keep provider-first resolution, since providers carry semantics the global table does not (Cursor marks estimated pricing, provider tables intentionally override globals). The fix we would take: providers fall back to the global resolver instead of the raw id, and models merges rows that resolve to the same provider plus display name after resolution, the same display-name merge the By Model panel uses.

A PR along those lines is very welcome. Tests to include: one provider-fallback case (a gpt-5.6-* id through a provider with a local table), one path-form case, and a merge case where two raw ids collapse into one row with summed tokens and cost.