INVARIANT HUNTER

The product of this skill is a negative.

Most pattern-spotting ends in a number that looks like it means something, and the finding is written down as though it survived scrutiny. This skill takes the opposite position: a claim is not a finding until the measurement that could have refuted it has been run, and reported either way. What comes out is usually "this is not an invariant, and here is the residual that says so", which is a more useful result than a pattern nobody checked.

Two quantities are under test, both taken from the ELOHIM ghost seed so the numbers line up with the measurements that skill already makes:

• the Collatz trace. An odd step multiplies by 3 + 1/n, so the per-step multiplier cannot be constant along a non-trivial trace, and a constant multiplier is exactly what a conserved quantity needs.
• the ghost seed. A hash-derived integer is divisible by a given prime with probability about 1/p. Finding the primes below 200 that divide it is a control, not a discovery.

The instrument is also explicit about one measurement it refuses to use. The product of every multiplier in the trace telescopes to 1/n0 for any trace that reaches 1, so it cannot fail and therefore cannot refute anything. It is recorded, and it is named as the non-evidence it is. See references/traps.md.

Run the gate

From the repository root:

python3 skills/elohim-harness/scripts/harness_run.py --skill-dir skills/invariant-hunter

Exit 0 when all five gates hold, 1 when one fails, 2 when the skill or its instrument cannot be located. To run the instrument or either script on its own, cd nowhere: each resolves its own paths relative to its own file.

The instrument also honours an override, derived from the directory name the way the contract specifies:

ELOHIM_INVARIANT_HUNTER_SCRIPT=/path/to/invariant_hunter.py \
  python3 skills/elohim-harness/scripts/harness_run.py --skill-dir skills/invariant-hunter

An override that points at a missing file is an error, exit 2, never a silent fallback to the bundled copy.

The five gates

pin — the ledger pins this instrument's sha256 and byte count. Holding the instrument beside its ledger fixes only which code runs; the pin is what makes an edit to that code visible instead of silent. Drift prints both hashes and can never resolve to PASS. Expect DRIFT the first time after any change to the instrument: that is the gate working.

ledger — each recorded fact is re-measured against a fresh shard and reported with its residual. A fact whose path has vanished reports missing rather than quietly passing. These three facts are seeded from measurements the ELOHIM instrument already makes at runtime, which is what makes this the one skill in the repository whose ledger is allowed to start populated. Every other skill starts empty and discovers its own.

traps — five traps in scripts/check_traps.py, each re-deriving its property from the invocation string and the map with code of its own, then comparing that against the shard. A trap that called the instrument would agree with it by construction. Each trap is expected to be able to fail, and a refutation skill whose traps cannot fail is a pattern-spotting skill with extra steps.

hygiene — scripts/check_hygiene.py lints these sources for network imports and for identifiers a shell filter can rewrite. A gate that reaches the network is not reproducible, and a gate whose source can be silently edited is not a gate.

claim binding — elohim-harness/scripts/claim_binding.py pulls every number out of every claim sentence and requires each one to be a rendering of a value something pins, part of a formula or a scan window, or declared in claim_binding_exemptions.json with a written reason. The four gates above measure the ledger; this one reads the sentence beside the number, which is where the worst defect this project shipped lived: prose asserting the opposite of its own pins while every gate reported green, and found by a human reading the output. A number nothing pins is therefore a failure and not a warning.

Promote the negatives

Promotion is the part that is specific to this skill, and it runs backwards from the usual habit.

--discover re-derives each claim from scratch and writes it to backlog.json with a contrast: the distance the claim travels. For this skill every expected contrast is large — a claim whose contrast lands below its own tolerance would mean the claim survived, and that is a real finding that has to be promoted rather than quietly dropped.

--promote ID:PATH[:TOL] then pins one measurement as a ledger fact, and from then on the harness re-measures it on every run. What gets promoted is not a success but a recorded refutation: conserved is false, the multiplier span is 0.1999327097772694, the power-of-3 gap is 4.666283238361249e-06. Those are facts. "The Collatz map has a conserved quantity" was never one.

Read every backlog entry before promoting it. An unpromoted measurement is not a defect, and a measurement nobody can explain has no business in the ledger.

Files