Repository navigation
Cross-harness model discovery reads for #792 - #794
Open
flip-engineer wants to merge 13 commits into
Open
flip-engineer wants to merge 13 commits into
flip-engineer wants to merge 13 commits into
Conversation
The models read reports the session recorded harness command, model and workspace. The muse profile path resolved one effort tier without recording it. Thread the recorded effort through the models path and include it as requestedEffort in the muse text answer. The spawned argv, the codex/omp JSON path and the unknown/failure paths are unchanged. modelExistence stays unknown for profile text.
Pin the recorded effort tier in the profile answer alongside the requested model. The fixture sets effort high and the answer must carry it as requestedEffort while modelExistence stays unknown.
models SESSION --all reports one model observation per distinct harness, command and model triple recorded in sessions. Unsupported harnesses and missing commands report unknown with their causes. The read changes no seat identity, route or pending work.
Two focused cases over recorded sessions: one per-route observation across omp and codex routes, and unknown unsupported and missing command routes reported without gating. Fixture helper accepts a target session for the second route.
Each --all entry now reads as its source session's own models read: the entry carries the source session, harness, command, model, recorded effort and recorded workspace, and the harness command runs in that workspace. Route rows whose fields contain the separator are excluded and named in the limits. Thread recorded effort through the shared models path with the same signature as spark-b 07310369a so the merge carries effort per route. Cover the fan-out, unknown routes and the empty route set in observed-usage fixtures.
A native session reads recorded harness/model/effort identities across sessions with operator account profiles and limits in one descriptive answer for issue 792. The read runs no harness command, writes nothing, and decides nothing.
The Bend compiler refuses a match over a local binder in inventory_record_json (remote run 38989401787, candidate evidence complete-native.stderr). Split the parsed-record rendering into inventory_record_parsed, following record_current_parsed in the same file. No behavior change.
Remote run 38989938340 candidate evidence reports code consumed more than once in inventory_profile_result: the code is read in the Bool.pick condition and in the failure answer. Use the +code binder the existing Fail patterns use (agent-command.bend, turn.bend) in both inventory result functions, and +rest in inventory_session_tail where both pick branches consume the tail, matching profiles_for. No behavior change.
Remote run 38990445184 candidate evidence reports expected Data, observed Type at inventory_session_next. Duplicable list parameters in this codebase carry List<&2,T>; Tx.split_lines returns List<String>, and its existing consumers (last_line, profile_lines) take the list plain. Take rest and lines plain, and move the empty-line choice into inventory_session_pick over a Bool, following the source_decide shape in usage.bend. No behavior change.
Both new folds read their head and tail results in the Bool.pick condition and in a branch. The existing models_command marks that shape with +answer; apply the same annotation to the inventory and models --all folds. No behavior change.
Remote run 38990709912 candidate evidence reports inventory_session_tail used before its definition. This codebase keeps definitions before their first caller, so the mutually recursive next/tail and observed/answers pairs cannot compile. Each fold is now one self-recursive definition whose cons case runs the step in its own do block; models_all_entry moves before its caller. No behavior change.
models_all calls usage_registered, which sat below it under the command heading. Definitions precede their first caller in these files, so move the unchanged definition above the cross-harness section; its existing callers in observed_usage and models stay below it. No behavior change.
Remote run 38991157042 candidate evidence reports db consumed more than once in models_all_answers: the entry read and the recursive call both use it. Annotate the parameter +db, as the other discovery definitions do. No behavior change.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Implements native cross-harness model discovery for issue #792 at base e18f003 (current main).
Changes integrated by the Principal Conductor from three Muse Spark 1.3 Contributor branches:
Constraints from #683 and #598: quota, budget, catalog membership, freshness and probe outcomes gate nothing. Per-session baton2-model-discovery-v1 answers keep their shape. No Artificial Analysis or Design Arena data is read; the answers state that absence.
Lineage: spark-a 095ab3b6e/a06f90e36/ecbba5a00, spark-b 07310369a/937316c6b, spark-c d885f8359 (final tree restored by fe573d676). Spark-b 51f1b6f6e (a sessions-only harness-inventory) stays preserved on its branch and is not integrated because spark-c defines the same verb with account coverage.
Validation: no local build or test was run. This pull request requests the remote bend2-native workflow, which builds the native coordinator and runs the bend2/test Python suites including observed-usage.py and harness-inventory.py. Results will be reported on #792 with run identities. Live provider reads are separate observations and have not been run for this change.