EVIDENCE · THE MEASUREMENT RECORD
Every figure, and how it was taken.
Every number printed anywhere on this site is registered here: how it was measured, when, and what you can do to check it. Where a number is not public, this page says so instead of implying otherwise.
The register
27 figures, 9 executed gates, 16 reproducible from outside
Executed gates
Commit-pinned runner results: the verification signals, first.
cargo test --workspace --no-fail-fast at commit ca33d4c, run on a clean export of that commit: 72 test binaries summed, 24 ignored.
./scripts/gate.sh: the full suite plus a strict lint that must report zero errors and zero warnings.
cargo test at commit c5dc71b.
cargo test at commit 18036f4.
cargo test --no-fail-fast at commit 34f03e6, all nine test binaries summed (239 lib, 345 bin, 23 across the seven small ones).
Executed gates (runtime): npm test in ecosystem-core (242 passed at 1f1be6e, full verify chain exit 0), ecosystem-auth (28 passed), and content-management, the Supabase CMS with an MCP surface (84 passed across 6 files).
cargo test --workspace --no-fail-fast at commit 44e7f9b, every test binary summed: 449 lib plus MCP e2e, plugin, integration and the smaller binaries.
python3 scripts/test_genome.py at commit 3a2873a in the standalone repository.
HTTP GET from outside the operation network, one request per URL, on a clean connection: status code read from the response line.
Countable public surfaces
Figures a reader can re-derive from the public tree in minutes.
Top-level key count in the json!
Public repository count from the GitHub GraphQL API: owned, non-fork repositories filtered to public visibility.
Raw line count over .rs files at checkout 5a116305.
Member count in the Cargo.toml [workspace] members list.
Raw line count over src/ and crates/, excluding build directories.
The governed operation, described
Private substrates, published as measured facts with the privacy named.
arch stats over the operation repository: the pool count from the architectural registry.
The same arch stats pass: decisions and regression guards indexed in the governance registry.
Agent count read from the task-grid employee registry in the operation repository.
Tool count read from the binary registry: sourcehound call --list.
Catalog read from the running binary: sourcehound catalog.
Payload size measured before and after TOON compression.
Costs, constraints, and the failure record
The internal measurements, and the postmortems they produced.
Tool-surface audit, 2026-08-26: token cost of a session that loads every tool the audit covered, summed.
SELECT COUNT(*) FROM executions against the cron execution record, with 92 jobs declared.
Per-server count from the tool-surface audit.
Static upper bounds
Declared counts that annotate the executed gates above; labelled static.
Static test-attribute count over crates/: lines matching #[test], #[tokio::test], #[test_case], or #[rstest], at checkout 5a116305.
Static test-attribute count over src/ and crates/.
How a figure earns its place here
- 01
A declared count is labelled declared
A test-attribute count counts annotations in source. It is not a count of tests a runner executed, and it is an upper bound. Every static figure in the register says static, and the executed figure replaces it when one exists.
- 02
A private record says it is private
Some numbers come from repositories that are not public. The register states that plainly and stops guessing at a link. A dead link is worse than an absent one, because it looks like proof.
- 03
A superseded number stays visible
The numbers this site stopped printing are on the record below, each with the value that replaced it and the reason. A record that only shows the current number asks to be trusted; a record that shows what it previously printed asks to be checked.
Figures that changed
Every number this site stopped printing stays on this record, with the value that replaced it and the reason. Four entries changed what a figure means; the rest are recounts, folded into one-line entries below.
| Ref | Figure | Was | Now |
|---|---|---|---|
| R2 | sourcehound MCP tools | 93 (profile) against 29 (repository description) | about 32 tool-name registrations across 25 tool groups |
| The two published totals measured different things, one the profile copy and one the repository description. Neither survived contact with the registry, which counts tool names and tool groups separately. The README is now authoritative for this surface. | |||
| R4 | Portfolio totals | 73 public repositories, 2.38M LOC, 31,300 tests | 77 public, 138 total |
| The aggregate LOC and test totals are withdrawn, not revised. A single summed line count across repositories of different languages and vendoring practices is not a comparable number, so the site prints a per-repository figure instead. The register carries each one separately. | |||
| R6 | Recorded cron executions | 476 | 0 |
| The 476 was never measured. It was a hardcoded fallback value in the report generator, and it fired on every run because the execution table was empty, so the missing metric was replaced by a plausible false one. The fix was to delete the default so an absent metric renders as absent. This is the most useful entry in this table: a missing metric fails visibly, and a defaulted metric lies quietly. | |||
| R16 | Operation substrate figures | 24,357 commits, cited to github.com/ishan-parihar/hermes-agent | 25 governed pools, 271 ADs / 211 RGs, cited to the private operation repository |
| The September figure counted the history of a fork of NousResearch's open-source hermes-agent and linked that fork as the operator's repository. The fork is upstream's work; the citation claimed it. The figure and the repository link are withdrawn in full, and the substrate's headline numbers are now measured from the system the operation actually runs. The case-study slug hermes-agent is retired with the figure and redirects to /work/organism. | |||
The recount log 20 figures recounted or re-scoped
- R1 Public repositories 73 → 77
- R3 operant tests and crates 9,249 test attributes, 19 crates → 9,811 test attributes, 18 crates
- R5 deckmill test attributes 185 → 196
- R7 tdg-rust test attributes 626 → 648
- R8 Public repositories 77 public, 138 total (2026-09-22), with R1 open between 73 and 77 → 67 public, 132 total (2026-10-24)
- R9 operant crates 18 → 7
- R10 operant test attributes 9,811 → 4,979 (static)
- R11 operant LOC 581,190 (539,435 Rust), raw including vendored files → 388,828 first-party Rust, plus 4,381 vendored printed separately
- R12 deckmill slide types 46 across 6 categories → 13 top-level registry entries (5 live, 8 retained as removed)
- R13 deckmill LOC 27,207 → 27,383 (src/ and crates/, build directories excluded)
- R14 deckmill test attributes 196 → 203 (static)
- R15 Recorded cron executions 0, measured against the live hermes cron runtime (92 jobs declared) → 0, and the runtime it measured was decommissioned in the October reorganization
- R17 deckmill slide types 13 top-level registry entries (5 live, 8 retained as removed), with a consolidation narrative → 46 entries: 38 live slide types, 8 retained as removed; 13 distinct layout families is a different grouping
- R18 MCP tool registrations, fleet total 140 registered tools across the fleet (2026-08-26 audit) → Withdrawn. The per-server counts stand; the fleet total does not.
- R19 Portfolio test attributes, static 34,779 summed across public repositories (2026-09-22) → Withdrawn. The per-repository figures and the executed gates stand; the portfolio sum does not.
- R20 operant test attributes 4,979 at checkout 5a116305 → 4,959 at the same checkout, with the command recorded byte for byte
- R21 operant LOC 388,828 first-party Rust, plus 4,381 vendored printed separately → 390,981 lines under crates/, plus 4,381 under vendor/, both counted by one recorded pipeline
- R22 Public repositories 67 public, 132 total (2026-10-24) → 69 public, 131 total owned non-fork (2026-10-26)
- R24 Public repositories 69 public, 131 total owned non-fork (2026-10-26) → 48 public, 116 total owned non-fork (2026-10-30)
- R23 switchpost tool surface (repository still named social-forge) 300+ tools, 42 modules, 25 platforms (2026-09-22) → 389 tools, 44 modules, 26 platforms, counted in the public tree at the pushed master head
Read the numbers where they do work
The register is the reference. The case studies are the argument: what was built, what broke, and what the measurement said about it.