A falsifiable analysis engine that tests one question:
In manufacturing brownfield retrofit, is there a set of open-source production assets that policy and procurement instruments structurally cannot "see" — and if so, does that illegibility constitute a measurable but finite economic arbitrage, or merely a phantom?
This is a research/analysis instrument, not a product and not a trading
system. It is engineered to refuse to assert a conclusion it cannot prove with
audited evidence. On the data publicly available today, it correctly returns
phantom_moat — and that is the headline result, explained below.
- Not investment advice. The codebase contains a "monetization signal" concept (a long/short framing referencing real listed companies). It is gated off and voided in every current run and is present only as a falsifiable hypothesis (H4/H7), not a recommendation. Nothing here is financial advice or a solicitation. Do not act on it.
- No empirical claim is made. All numeric inputs are
public_unaudited/self_reported. Per the engine's honesty invariant they are narrative, not evidence. The project makes no claim that any company or open-source project is over- or under-valued. - Research artifact. Outputs are illustrative of a methodology. They are explicitly not certified findings.
BlindSpotMoat models manufacturing investment as two tracks: Track P
(policy-visible: subsidized, listed, audited) and Track O (open-source
brownfield retrofit: near-zero capital, invisible to procurement frameworks). It
asks whether Track O's illegibility is a real, finite arbitrage. The engine only
accepts data classed audited as empirical evidence. The audited evidence needed
to prove the claim — production-scale deployment telemetry and full-lifecycle
cost — does not exist in the public domain, independently confirmed by six
web-enabled AIs with zero fabrication. The engine therefore halts at its first
gate and returns phantom_moat: the idea is neither accepted nor
discarded, but held as evidence-blocked provisional. Advancing it requires
non-public field research, not more analysis.
The engine's first gate (S0 = hypotheses H0 & H5) requires ≥10 Track O
assets with audited production-scale deployment evidence. That evidence is
structurally unavailable publicly (shadow-IT budgets, air-gapped OT networks,
integrator NDAs). A naive tool would have "passed" using GitHub stars and README
payback claims. This one deliberately will not. phantom_moat means:
"The methodology runs end-to-end, but the claim cannot be certified from public data. Here is exactly what is missing and where it lives."
# Python 3.10+ ; no dependencies (standard library only)
cd BlindSpotMoat
python -m engine --mode dry-run # validate data contracts, gates, thresholds
python -m engine --mode execute # run the full pipeline (human-readable)
python -m engine --mode execute --json # machine-readableExpected current output: verdict = phantom_moat, gating_halt = H0/H5,
H0 FALSIFIED (0 audited production-scale assets), H5 INCONCLUSIVE,
monetization signal voided, legal-standing layer fail_closed_ok = True.
README.md ← you are here
LICENSE ← MIT
docs/
TECHNICAL_SPECIFICATION.md ← master reference (read this to fully understand)
DESIGN.md ← architecture: PGF Gantree + nodes + dependency graph
HYPOTHESES.md ← pre-registered H0–H7 (thresholds, gating order)
REVIEW_SYNTHESIS.md ← 7-AI adversarial design review → adopted changes
DATA_COLLECTION.md ← 6-AI data collection method + findings + the gap
engine/ ← the Python implementation (stdlib only)
__main__.py contracts.py harness.py pipeline.py errors.py
nodes/ … ← one module per Gantree node
fixtures/ … ← vintage-snapshot data (6-AI cross-verified, public_unaudited)
evidence/ ← auditable primary sources behind the headline claims
design_review/ ← the 7-AI review prompt + 7 raw responses (7/7 revise)
data_collection/ ← the 6-AI collection prompt + raw responses (6/6, fabricated=0)
Korean working originals and raw PGF scratch are kept locally under
_legacy/for provenance; they are intentionally excluded from the published repository (Englishdocs/+evidence/supersede them).
Start with
docs/TECHNICAL_SPECIFICATION.md. It is self-contained: an engineer or AI seeing this project for the first time can understand it, run it, interpret it, and know how to advance it from that one document.
- Generated by the A3IE idea pipeline; selected via cross-model-certified
surprise (idea
IDEA-20-02, lens "delete the scarce resource"). - Adversarially reviewed by 7 independent AIs → 7/7 "revise". Nine changes
adopted (
docs/REVIEW_SYNTHESIS.md): moat redefined as a finite window, the self-protection layer abandoned, set-difference replaced by a graded visibility graph, hypotheses re-registered with explicit thresholds. - Implemented as this engine (design → plan → execute → verify).
- Data-collected by 6 independent web AIs (
fabricated_values=0, 6/6): the audited evidence required is structurally unavailable publicly (docs/DATA_COLLECTION.md). - Verdict:
phantom_moat(empirically robust). State: evidence-blocked provisional. Only a non-public field-research track can change this.
Steps 2 and 4 are not claims to take on trust — the raw prompts and every raw
independent response are preserved verbatim under evidence/ so the
7/7 revise and 6/6 fabricated=0 results can be audited and re-run.
The idea is not refuted and not accepted — it is blocked on evidence
that does not exist in public. The single path to a different verdict is the
field-research track (NDA integrator interviews, factory site visits,
certification-body data) detailed in docs/DATA_COLLECTION.md §4. That is a
human/commercial-data effort, out of scope for any AI or public-web collection.
MIT — see LICENSE. © 2025–2026 sadpig70.