docs: canonical API reference + anti-reinvention decision table by drewstone · Pull Request #281 · tangle-network/agent-runtime

drewstone · 2026-06-14T00:30:18Z

The senior-quality, file:line-verified canonical surface — the doc that makes reinvention impossible. Genome→run→optimize→gate spine, the "I want to X → use Y → NOT Z" decision table (names every trap: runConversation≈loopUntil, skill-optimizer≈improvementDriver, profile-seam≈AgentProfile, new-Sandbox-loop≈openSandboxRun), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions + the 3 named gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient ToC. Produced by a 5-reader→synthesize→verify workflow.

…n thesis, Mode 0 (intelligence-off), do-not-claim firewall, Verified→Gated The depth>breadth keystone retracted to a tie at n=72; the SDK sells cost + verification + transfer, not quality improvement. Adds the honest-claim firewall (normative do-not-claim block), the intelligence-OFF billing floor (sandbox-stream, inference+compute only, billing-line-on-the-spawn-line), and renames Verified PRs → Gated PRs. Full plan: docs/go-live-plan.md (5-lens rederivation, code-verified).

…not a ban on 'smarter') Corrects the over-rotation: POWER-16 killed within-run-cleverness-beats-blind-at- equal-compute, NOT the ability to make agents smarter. Smarter comes from gated search (spend compute, certify the winner) + cheap serving of the certified artifact. 'Smarter' is allowed with its gate+CI+n attached — that's the differentiation. Aligns the billing boundary to usage-classification + spawn-gating (not budget-pool surgery).

Replaces the minimal 4-sentence baseline in trata-gate.mts and trata-gepa.mts with the surface found by GEPA across 9 runs (+8.6pp holdout on deepseek-v4-flash, confirmed twice independently). Adds five sentences: labeled-section structure, named-peer benchmarking with EV/EBITDA/P/E/margin metrics, verbatim guidance citation, IRR computation with arithmetic, and explicit SYSTEM_PROMPT env override for future experimentation.

All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

…crash A 400 from /api/sql-runner means the verifier SQL references a column that doesn't exist in the seeded schema (version skew between gym_dbs.zip and the Docker container). Treat it as 0 points for that verifier rather than halting the entire judge run.

…_DIR path

…n table The senior-quality, file:line-verified canonical surface: the genome→run→optimize→gate spine, the 'I want to X → use Y → NOT Z' decision table (names every reinvention trap), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions+gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient-first ToC. Produced via a 5-reader→synthesize→verify workflow; the recursive-atom/analyst section added after.

tangletools

✅ Auto-approved PR — `e8e9371f`

Blanket team auto-approval is enabled for this reviewer service.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: blanket_auto_approve · 2026-06-14T00:30:24Z}

drewstone · 2026-06-14T00:30:18Z

The senior-quality, file:line-verified canonical surface — the doc that makes reinvention impossible. Genome→run→optimize→gate spine, the "I want to X → use Y → NOT Z" decision table (names every trap: runConversation≈loopUntil, skill-optimizer≈improvementDriver, profile-seam≈AgentProfile, new-Sandbox-loop≈openSandboxRun), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions + the 3 named gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient ToC. Produced by a 5-reader→synthesize→verify workflow.

…n thesis, Mode 0 (intelligence-off), do-not-claim firewall, Verified→Gated The depth>breadth keystone retracted to a tie at n=72; the SDK sells cost + verification + transfer, not quality improvement. Adds the honest-claim firewall (normative do-not-claim block), the intelligence-OFF billing floor (sandbox-stream, inference+compute only, billing-line-on-the-spawn-line), and renames Verified PRs → Gated PRs. Full plan: docs/go-live-plan.md (5-lens rederivation, code-verified).

…not a ban on 'smarter') Corrects the over-rotation: POWER-16 killed within-run-cleverness-beats-blind-at- equal-compute, NOT the ability to make agents smarter. Smarter comes from gated search (spend compute, certify the winner) + cheap serving of the certified artifact. 'Smarter' is allowed with its gate+CI+n attached — that's the differentiation. Aligns the billing boundary to usage-classification + spawn-gating (not budget-pool surgery).

Replaces the minimal 4-sentence baseline in trata-gate.mts and trata-gepa.mts with the surface found by GEPA across 9 runs (+8.6pp holdout on deepseek-v4-flash, confirmed twice independently). Adds five sentences: labeled-section structure, named-peer benchmarking with EV/EBITDA/P/E/margin metrics, verbatim guidance citation, IRR computation with arithmetic, and explicit SYSTEM_PROMPT env override for future experimentation.

All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

…crash A 400 from /api/sql-runner means the verifier SQL references a column that doesn't exist in the seeded schema (version skew between gym_dbs.zip and the Docker container). Treat it as 0 points for that verifier rather than halting the entire judge run.

…_DIR path

…n table The senior-quality, file:line-verified canonical surface: the genome→run→optimize→gate spine, the 'I want to X → use Y → NOT Z' decision table (names every reinvention trap), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions+gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient-first ToC. Produced via a 5-reader→synthesize→verify workflow; the recursive-atom/analyst section added after.

tangletools

✅ Auto-approved PR — `e8e9371f`

Blanket team auto-approval is enabled for this reviewer service.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: blanket_auto_approve · 2026-06-14T00:30:24Z}

drewstone · 2026-06-14T00:30:18Z

The senior-quality, file:line-verified canonical surface — the doc that makes reinvention impossible. Genome→run→optimize→gate spine, the "I want to X → use Y → NOT Z" decision table (names every trap: runConversation≈loopUntil, skill-optimizer≈improvementDriver, profile-seam≈AgentProfile, new-Sandbox-loop≈openSandboxRun), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions + the 3 named gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient ToC. Produced by a 5-reader→synthesize→verify workflow.

…n thesis, Mode 0 (intelligence-off), do-not-claim firewall, Verified→Gated The depth>breadth keystone retracted to a tie at n=72; the SDK sells cost + verification + transfer, not quality improvement. Adds the honest-claim firewall (normative do-not-claim block), the intelligence-OFF billing floor (sandbox-stream, inference+compute only, billing-line-on-the-spawn-line), and renames Verified PRs → Gated PRs. Full plan: docs/go-live-plan.md (5-lens rederivation, code-verified).

…not a ban on 'smarter') Corrects the over-rotation: POWER-16 killed within-run-cleverness-beats-blind-at- equal-compute, NOT the ability to make agents smarter. Smarter comes from gated search (spend compute, certify the winner) + cheap serving of the certified artifact. 'Smarter' is allowed with its gate+CI+n attached — that's the differentiation. Aligns the billing boundary to usage-classification + spawn-gating (not budget-pool surgery).

Replaces the minimal 4-sentence baseline in trata-gate.mts and trata-gepa.mts with the surface found by GEPA across 9 runs (+8.6pp holdout on deepseek-v4-flash, confirmed twice independently). Adds five sentences: labeled-section structure, named-peer benchmarking with EV/EBITDA/P/E/margin metrics, verbatim guidance citation, IRR computation with arithmetic, and explicit SYSTEM_PROMPT env override for future experimentation.

All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

…crash A 400 from /api/sql-runner means the verifier SQL references a column that doesn't exist in the seeded schema (version skew between gym_dbs.zip and the Docker container). Treat it as 0 points for that verifier rather than halting the entire judge run.

…_DIR path

…n table The senior-quality, file:line-verified canonical surface: the genome→run→optimize→gate spine, the 'I want to X → use Y → NOT Z' decision table (names every reinvention trap), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions+gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient-first ToC. Produced via a 5-reader→synthesize→verify workflow; the recursive-atom/analyst section added after.

tangletools

✅ Auto-approved PR — `e8e9371f`

Blanket team auto-approval is enabled for this reviewer service.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: blanket_auto_approve · 2026-06-14T00:30:24Z}

drewstone · 2026-06-14T00:30:18Z

The senior-quality, file:line-verified canonical surface — the doc that makes reinvention impossible. Genome→run→optimize→gate spine, the "I want to X → use Y → NOT Z" decision table (names every trap: runConversation≈loopUntil, skill-optimizer≈improvementDriver, profile-seam≈AgentProfile, new-Sandbox-loop≈openSandboxRun), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions + the 3 named gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient ToC. Produced by a 5-reader→synthesize→verify workflow.

…n thesis, Mode 0 (intelligence-off), do-not-claim firewall, Verified→Gated The depth>breadth keystone retracted to a tie at n=72; the SDK sells cost + verification + transfer, not quality improvement. Adds the honest-claim firewall (normative do-not-claim block), the intelligence-OFF billing floor (sandbox-stream, inference+compute only, billing-line-on-the-spawn-line), and renames Verified PRs → Gated PRs. Full plan: docs/go-live-plan.md (5-lens rederivation, code-verified).

…not a ban on 'smarter') Corrects the over-rotation: POWER-16 killed within-run-cleverness-beats-blind-at- equal-compute, NOT the ability to make agents smarter. Smarter comes from gated search (spend compute, certify the winner) + cheap serving of the certified artifact. 'Smarter' is allowed with its gate+CI+n attached — that's the differentiation. Aligns the billing boundary to usage-classification + spawn-gating (not budget-pool surgery).

Replaces the minimal 4-sentence baseline in trata-gate.mts and trata-gepa.mts with the surface found by GEPA across 9 runs (+8.6pp holdout on deepseek-v4-flash, confirmed twice independently). Adds five sentences: labeled-section structure, named-peer benchmarking with EV/EBITDA/P/E/margin metrics, verbatim guidance citation, IRR computation with arithmetic, and explicit SYSTEM_PROMPT env override for future experimentation.

All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

…crash A 400 from /api/sql-runner means the verifier SQL references a column that doesn't exist in the seeded schema (version skew between gym_dbs.zip and the Docker container). Treat it as 0 points for that verifier rather than halting the entire judge run.

…_DIR path

…n table The senior-quality, file:line-verified canonical surface: the genome→run→optimize→gate spine, the 'I want to X → use Y → NOT Z' decision table (names every reinvention trap), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions+gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient-first ToC. Produced via a 5-reader→synthesize→verify workflow; the recursive-atom/analyst section added after.

tangletools

✅ Auto-approved PR — `e8e9371f`

Blanket team auto-approval is enabled for this reviewer service.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: blanket_auto_approve · 2026-06-14T00:30:24Z}

drewstone · 2026-06-14T00:30:18Z

The senior-quality, file:line-verified canonical surface — the doc that makes reinvention impossible. Genome→run→optimize→gate spine, the "I want to X → use Y → NOT Z" decision table (names every trap: runConversation≈loopUntil, skill-optimizer≈improvementDriver, profile-seam≈AgentProfile, new-Sandbox-loop≈openSandboxRun), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions + the 3 named gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient ToC. Produced by a 5-reader→synthesize→verify workflow.

…n thesis, Mode 0 (intelligence-off), do-not-claim firewall, Verified→Gated The depth>breadth keystone retracted to a tie at n=72; the SDK sells cost + verification + transfer, not quality improvement. Adds the honest-claim firewall (normative do-not-claim block), the intelligence-OFF billing floor (sandbox-stream, inference+compute only, billing-line-on-the-spawn-line), and renames Verified PRs → Gated PRs. Full plan: docs/go-live-plan.md (5-lens rederivation, code-verified).

…not a ban on 'smarter') Corrects the over-rotation: POWER-16 killed within-run-cleverness-beats-blind-at- equal-compute, NOT the ability to make agents smarter. Smarter comes from gated search (spend compute, certify the winner) + cheap serving of the certified artifact. 'Smarter' is allowed with its gate+CI+n attached — that's the differentiation. Aligns the billing boundary to usage-classification + spawn-gating (not budget-pool surgery).

Replaces the minimal 4-sentence baseline in trata-gate.mts and trata-gepa.mts with the surface found by GEPA across 9 runs (+8.6pp holdout on deepseek-v4-flash, confirmed twice independently). Adds five sentences: labeled-section structure, named-peer benchmarking with EV/EBITDA/P/E/margin metrics, verbatim guidance citation, IRR computation with arithmetic, and explicit SYSTEM_PROMPT env override for future experimentation.

All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

…crash A 400 from /api/sql-runner means the verifier SQL references a column that doesn't exist in the seeded schema (version skew between gym_dbs.zip and the Docker container). Treat it as 0 points for that verifier rather than halting the entire judge run.

…_DIR path

…n table The senior-quality, file:line-verified canonical surface: the genome→run→optimize→gate spine, the 'I want to X → use Y → NOT Z' decision table (names every reinvention trap), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions+gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient-first ToC. Produced via a 5-reader→synthesize→verify workflow; the recursive-atom/analyst section added after.

tangletools

✅ Auto-approved PR — `e8e9371f`

Blanket team auto-approval is enabled for this reviewer service.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: blanket_auto_approve · 2026-06-14T00:30:24Z}

drewstone · 2026-06-14T00:30:18Z

The senior-quality, file:line-verified canonical surface — the doc that makes reinvention impossible. Genome→run→optimize→gate spine, the "I want to X → use Y → NOT Z" decision table (names every trap: runConversation≈loopUntil, skill-optimizer≈improvementDriver, profile-seam≈AgentProfile, new-Sandbox-loop≈openSandboxRun), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions + the 3 named gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient ToC. Produced by a 5-reader→synthesize→verify workflow.

…n thesis, Mode 0 (intelligence-off), do-not-claim firewall, Verified→Gated The depth>breadth keystone retracted to a tie at n=72; the SDK sells cost + verification + transfer, not quality improvement. Adds the honest-claim firewall (normative do-not-claim block), the intelligence-OFF billing floor (sandbox-stream, inference+compute only, billing-line-on-the-spawn-line), and renames Verified PRs → Gated PRs. Full plan: docs/go-live-plan.md (5-lens rederivation, code-verified).

…not a ban on 'smarter') Corrects the over-rotation: POWER-16 killed within-run-cleverness-beats-blind-at- equal-compute, NOT the ability to make agents smarter. Smarter comes from gated search (spend compute, certify the winner) + cheap serving of the certified artifact. 'Smarter' is allowed with its gate+CI+n attached — that's the differentiation. Aligns the billing boundary to usage-classification + spawn-gating (not budget-pool surgery).

Replaces the minimal 4-sentence baseline in trata-gate.mts and trata-gepa.mts with the surface found by GEPA across 9 runs (+8.6pp holdout on deepseek-v4-flash, confirmed twice independently). Adds five sentences: labeled-section structure, named-peer benchmarking with EV/EBITDA/P/E/margin metrics, verbatim guidance citation, IRR computation with arithmetic, and explicit SYSTEM_PROMPT env override for future experimentation.

All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

…crash A 400 from /api/sql-runner means the verifier SQL references a column that doesn't exist in the seeded schema (version skew between gym_dbs.zip and the Docker container). Treat it as 0 points for that verifier rather than halting the entire judge run.

…_DIR path

…n table The senior-quality, file:line-verified canonical surface: the genome→run→optimize→gate spine, the 'I want to X → use Y → NOT Z' decision table (names every reinvention trap), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions+gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient-first ToC. Produced via a 5-reader→synthesize→verify workflow; the recursive-atom/analyst section added after.

tangletools

✅ Auto-approved PR — `e8e9371f`

Blanket team auto-approval is enabled for this reviewer service.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: blanket_auto_approve · 2026-06-14T00:30:24Z}

drewstone · 2026-06-14T00:30:18Z

The senior-quality, file:line-verified canonical surface — the doc that makes reinvention impossible. Genome→run→optimize→gate spine, the "I want to X → use Y → NOT Z" decision table (names every trap: runConversation≈loopUntil, skill-optimizer≈improvementDriver, profile-seam≈AgentProfile, new-Sandbox-loop≈openSandboxRun), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions + the 3 named gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient ToC. Produced by a 5-reader→synthesize→verify workflow.

…n thesis, Mode 0 (intelligence-off), do-not-claim firewall, Verified→Gated The depth>breadth keystone retracted to a tie at n=72; the SDK sells cost + verification + transfer, not quality improvement. Adds the honest-claim firewall (normative do-not-claim block), the intelligence-OFF billing floor (sandbox-stream, inference+compute only, billing-line-on-the-spawn-line), and renames Verified PRs → Gated PRs. Full plan: docs/go-live-plan.md (5-lens rederivation, code-verified).

…not a ban on 'smarter') Corrects the over-rotation: POWER-16 killed within-run-cleverness-beats-blind-at- equal-compute, NOT the ability to make agents smarter. Smarter comes from gated search (spend compute, certify the winner) + cheap serving of the certified artifact. 'Smarter' is allowed with its gate+CI+n attached — that's the differentiation. Aligns the billing boundary to usage-classification + spawn-gating (not budget-pool surgery).

Replaces the minimal 4-sentence baseline in trata-gate.mts and trata-gepa.mts with the surface found by GEPA across 9 runs (+8.6pp holdout on deepseek-v4-flash, confirmed twice independently). Adds five sentences: labeled-section structure, named-peer benchmarking with EV/EBITDA/P/E/margin metrics, verbatim guidance citation, IRR computation with arithmetic, and explicit SYSTEM_PROMPT env override for future experimentation.

All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

…crash A 400 from /api/sql-runner means the verifier SQL references a column that doesn't exist in the seeded schema (version skew between gym_dbs.zip and the Docker container). Treat it as 0 points for that verifier rather than halting the entire judge run.

…_DIR path

…n table The senior-quality, file:line-verified canonical surface: the genome→run→optimize→gate spine, the 'I want to X → use Y → NOT Z' decision table (names every reinvention trap), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions+gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient-first ToC. Produced via a 5-reader→synthesize→verify workflow; the recursive-atom/analyst section added after.

tangletools

✅ Auto-approved PR — `e8e9371f`

Blanket team auto-approval is enabled for this reviewer service.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: blanket_auto_approve · 2026-06-14T00:30:24Z}

drewstone · 2026-06-14T00:30:18Z

The senior-quality, file:line-verified canonical surface — the doc that makes reinvention impossible. Genome→run→optimize→gate spine, the "I want to X → use Y → NOT Z" decision table (names every trap: runConversation≈loopUntil, skill-optimizer≈improvementDriver, profile-seam≈AgentProfile, new-Sandbox-loop≈openSandboxRun), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions + the 3 named gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient ToC. Produced by a 5-reader→synthesize→verify workflow.

…n thesis, Mode 0 (intelligence-off), do-not-claim firewall, Verified→Gated The depth>breadth keystone retracted to a tie at n=72; the SDK sells cost + verification + transfer, not quality improvement. Adds the honest-claim firewall (normative do-not-claim block), the intelligence-OFF billing floor (sandbox-stream, inference+compute only, billing-line-on-the-spawn-line), and renames Verified PRs → Gated PRs. Full plan: docs/go-live-plan.md (5-lens rederivation, code-verified).

…not a ban on 'smarter') Corrects the over-rotation: POWER-16 killed within-run-cleverness-beats-blind-at- equal-compute, NOT the ability to make agents smarter. Smarter comes from gated search (spend compute, certify the winner) + cheap serving of the certified artifact. 'Smarter' is allowed with its gate+CI+n attached — that's the differentiation. Aligns the billing boundary to usage-classification + spawn-gating (not budget-pool surgery).

Replaces the minimal 4-sentence baseline in trata-gate.mts and trata-gepa.mts with the surface found by GEPA across 9 runs (+8.6pp holdout on deepseek-v4-flash, confirmed twice independently). Adds five sentences: labeled-section structure, named-peer benchmarking with EV/EBITDA/P/E/margin metrics, verbatim guidance citation, IRR computation with arithmetic, and explicit SYSTEM_PROMPT env override for future experimentation.

All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

…crash A 400 from /api/sql-runner means the verifier SQL references a column that doesn't exist in the seeded schema (version skew between gym_dbs.zip and the Docker container). Treat it as 0 points for that verifier rather than halting the entire judge run.

…_DIR path

…n table The senior-quality, file:line-verified canonical surface: the genome→run→optimize→gate spine, the 'I want to X → use Y → NOT Z' decision table (names every reinvention trap), the recursive atom (persona=driver, spawnChild=worker|sub-driver, isolated|Workspace artifact, conserved sub-budgets, analyst dimensions+gaps), the two substrates, the intelligence-SDK note. Linked first in the CLAUDE.md orient-first ToC. Produced via a 5-reader→synthesize→verify workflow; the recursive-atom/analyst section added after.

tangletools

✅ Auto-approved PR — `e8e9371f`

Blanket team auto-approval is enabled for this reviewer service.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: blanket_auto_approve · 2026-06-14T00:30:24Z}

drewstone added 7 commits June 13, 2026 12:04

chore(bench): delete legacy standalone gate scripts
a491abc
All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

docs(bench): eops setup — gym_dbs.zip download + correct EOPS_GYM_DBS…
b628c8b
…_DIR path

tangletools approved these changes Jun 14, 2026
View reviewed changes

drewstone mentioned this pull request Jun 14, 2026
docs: canonical API reference + anti-reinvention decision table #283
Merged

drewstone closed this Jun 14, 2026

drewstone added 7 commits June 13, 2026 12:04

chore(bench): delete legacy standalone gate scripts
a491abc
All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

docs(bench): eops setup — gym_dbs.zip download + correct EOPS_GYM_DBS…
b628c8b
…_DIR path

tangletools approved these changes Jun 14, 2026
View reviewed changes

drewstone mentioned this pull request Jun 14, 2026
docs: canonical API reference + anti-reinvention decision table #283
Merged

drewstone closed this Jun 14, 2026

drewstone added 7 commits June 13, 2026 12:04

chore(bench): delete legacy standalone gate scripts
a491abc
All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

docs(bench): eops setup — gym_dbs.zip download + correct EOPS_GYM_DBS…
b628c8b
…_DIR path

tangletools approved these changes Jun 14, 2026
View reviewed changes

drewstone mentioned this pull request Jun 14, 2026
docs: canonical API reference + anti-reinvention decision table #283
Merged

drewstone closed this Jun 14, 2026

drewstone added 7 commits June 13, 2026 12:04

chore(bench): delete legacy standalone gate scripts
a491abc
All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

docs(bench): eops setup — gym_dbs.zip download + correct EOPS_GYM_DBS…
b628c8b
…_DIR path

tangletools approved these changes Jun 14, 2026
View reviewed changes

drewstone mentioned this pull request Jun 14, 2026
docs: canonical API reference + anti-reinvention decision table #283
Merged

drewstone closed this Jun 14, 2026

drewstone added 7 commits June 13, 2026 12:04

chore(bench): delete legacy standalone gate scripts
a491abc
All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

docs(bench): eops setup — gym_dbs.zip download + correct EOPS_GYM_DBS…
b628c8b
…_DIR path

tangletools approved these changes Jun 14, 2026
View reviewed changes

drewstone mentioned this pull request Jun 14, 2026
docs: canonical API reference + anti-reinvention decision table #283
Merged

drewstone closed this Jun 14, 2026

drewstone added 7 commits June 13, 2026 12:04

chore(bench): delete legacy standalone gate scripts
a491abc
All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

docs(bench): eops setup — gym_dbs.zip download + correct EOPS_GYM_DBS…
b628c8b
…_DIR path

tangletools approved these changes Jun 14, 2026
View reviewed changes

drewstone mentioned this pull request Jun 14, 2026
docs: canonical API reference + anti-reinvention decision table #283
Merged

drewstone closed this Jun 14, 2026

drewstone added 7 commits June 13, 2026 12:04

chore(bench): delete legacy standalone gate scripts
a491abc
All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

docs(bench): eops setup — gym_dbs.zip download + correct EOPS_GYM_DBS…
b628c8b
…_DIR path

tangletools approved these changes Jun 14, 2026
View reviewed changes

drewstone mentioned this pull request Jun 14, 2026
docs: canonical API reference + anti-reinvention decision table #283
Merged

drewstone closed this Jun 14, 2026

drewstone added 7 commits June 13, 2026 12:04

chore(bench): delete legacy standalone gate scripts
a491abc
All superseded by BENCH=<name> tsx bench/src/rsi.ts — the one flow.

docs(bench): eops setup — gym_dbs.zip download + correct EOPS_GYM_DBS…
b628c8b
…_DIR path

tangletools approved these changes Jun 14, 2026
View reviewed changes

drewstone mentioned this pull request Jun 14, 2026
docs: canonical API reference + anti-reinvention decision table #283
Merged

drewstone closed this Jun 14, 2026

Uh oh!

Conversation

drewstone commented Jun 14, 2026

Uh oh!

tangletools left a comment

Choose a reason for hiding this comment

✅ Auto-approved PR — e8e9371f

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

drewstone commented Jun 14, 2026

Uh oh!

tangletools left a comment

Choose a reason for hiding this comment

✅ Auto-approved PR — e8e9371f

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

drewstone commented Jun 14, 2026

Uh oh!

tangletools left a comment

Choose a reason for hiding this comment

✅ Auto-approved PR — e8e9371f

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

drewstone commented Jun 14, 2026

Uh oh!

tangletools left a comment

Choose a reason for hiding this comment

✅ Auto-approved PR — e8e9371f

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

drewstone commented Jun 14, 2026

Uh oh!

tangletools left a comment

Choose a reason for hiding this comment

✅ Auto-approved PR — e8e9371f

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

✅ Auto-approved PR — `e8e9371f`

✅ Auto-approved PR — `e8e9371f`

✅ Auto-approved PR — `e8e9371f`

✅ Auto-approved PR — `e8e9371f`

✅ Auto-approved PR — `e8e9371f`

✅ Auto-approved PR — `e8e9371f`

✅ Auto-approved PR — `e8e9371f`