The membrane between the organization and the market.
Later layer · deferred by design — highest stakes, last to build
GRAMS is the only organ that faces outward, which is exactly why it comes last.
Autonomous outbound communication sits on the blueprint's
deliberately-not-yet list: the membrane opens only after the factory can prove,
from its own ledger, what it did and why.
GRAMS is the one organ of the factory that faces outward. Applications,
postings, proposals, outreach, contracts, labor acquisition — every channel through
which the organization touches the external economy. The other organs run on internal
signals; GRAMS is the only one whose primary surface is not yours.
It is also the organ that requires the most explicit human oversight. Every outbound
commitment carries reputational weight, every inbound application carries judgment risk,
and the asymmetry is permanent. GRAMS is designed so that the factory handles volume
and the human handles consequence — and the line between the two is configurable, not
assumed.
The interface with the outside world is where your organization spends the most attention
and gets the worst leverage. Every application requires you. Every proposal requires
you. Every outbound reply requires you. Volume scales linearly with opportunity, and
you do not.
The industry's answer has been one of two extremes: either a SaaS form-filler that automates
surface mechanics without understanding context, or a human-staffed process that doesn't
scale. Neither is the right shape. The right shape is a membrane: most flow passes
through with policed defaults, and only the high-stakes traffic surfaces to you. The membrane
is what GRAMS is.
The Channels
Every place the organization touches the market
GRAMS doesn't invent these channels — they already exist, scattered across email, LinkedIn,
portals, contracts, ATSs, and a hundred forms. The design unifies them under a single
membrane with consistent oversight rules.
APPLICATIONS
↑ OUTBOUND
Job applications, grant submissions, RFP responses. The design: GRAMS drafts against GAMS-grounded capabilities and the GEMS-tracked portfolio, then routes for review.
POSTINGS
↑ OUTBOUND
Job postings, calls for proposals, public announcements. Drafted to match the topology GAMS says is actually staffable, not what someone wished was true.
INBOUND APPS
↓ INBOUND
Job applicants, vendor proposals, partnership requests. Triaged against the topology: fit is a function of the GAMS capacity model (a later layer), not vibes.
OUTREACH
↑ OUTBOUND
Cold and warm reach-outs to leads, candidates, partners. Personalized against what GEMS knows about your prior work, not against a generic template.
CONTRACTS
↕ BIDIRECTIONAL
Contract drafting, redlines, signature workflow. Terms grounded in viability (GAMS) and lineage (GIMS) so commitments aren't made on assumptions.
LABOR ACQUISITION
↕ BIDIRECTIONAL
End-to-end hiring: job posted → applicants triaged → interviews coordinated → offer drafted. Each step grounded in the topology, none of it requiring your continuous attention.
The Mechanism
How a request crosses the membrane
Every interaction GRAMS will handle, inbound or outbound, follows the same four-step
pattern. The pattern is what makes oversight tractable: the same review surface, regardless
of channel.
01 · CLASSIFY
Every signal crossing the membrane is classified — what channel, what counterparty, what
stakes. Classification determines oversight tier. This step is the difference
between a useful membrane and a leaky one.
02 · DRAFT
GRAMS will compose the outbound artifact by pulling from the substrate: GAMS for what's
credibly achievable, GEMS for prior portfolio evidence, GIMS for the operational
lineage. The draft is grounded in what the organization actually is, not in
what the prompt template hopes it is.
03 · TIER
Every drafted artifact lands in one of three tiers. Auto-send for the cheap, low-stakes,
high-volume traffic. Surface for review on anything carrying meaningful reputational or
contractual weight. Hand off to a human entirely when the consequence exceeds the
membrane's authority. The line between tiers is configurable.
04 · LOG
Every membrane crossing lands in GIMS as a sentence with full lineage: what the system
saw, what it drafted, which tier it took, who reviewed (if anyone), and what went out
the door. The membrane is not opaque to itself. Audit is a property of the
channel, not a feature you remember to turn on.
The Oversight Architecture
Where the line between factory and human actually lives
Most automation tools punt on this. They either run with too much latitude (and embarrass
you) or too little (and don't earn their keep). GRAMS is designed to force the question to
be answered explicitly, per channel, and re-answered as confidence grows.
THREE TIERS · ONE CHANNEL AT A TIME
AUTO
High-volume, low-stakes, high-confidence. Routine outreach acknowledgements, scheduled
check-ins, standard portal submissions where the format is settled and the content is
grounded. The membrane handles it; you don't see it unless something fails.
REVIEW
Drafted by GRAMS, queued for your eyes. Most proposals, most cover letters, anything
making non-trivial commitments. The draft is far better than blank-page; your job is
to approve or modify, not compose from scratch.
HUMAN
Reputational stakes, novel counterparties, or anything where the membrane's confidence
is low. GRAMS prepares the brief — context, prior history, recommended posture — and
hands the conversation to you entirely.
The tiering policy is per-channel and per-counterparty. Acme Corp moves from REVIEW
to AUTO after five successful exchanges. A first-time regulator stays HUMAN regardless of
volume. The default is conservative; the latitude is earned.
These tiers are not a GRAMS invention: they are the outward-facing case of the same gate the
whole factory uses — operating modes like No Outbound enforced before execution,
confirmation gates on anything external, and every crossing a sentence in the ledger.
The Defining Capability
A market interface that scales without scaling you
The scene this layer is designed to make ordinary: you open your machine. Eighteen
applications were sent overnight, each grounded in your
actual portfolio and the topology GAMS confirms as achievable. Forty-three inbound
inquiries were triaged — eleven advanced to second-contact templates auto-sent, twenty-one
await your judgment with full briefs attached, eleven were declined politely with the
reasoning logged. A draft renewal of the Acme contract sits in your review queue with the
three terms GAMS flagged as no-longer-credible already softened.
You spend an hour. Not eight. And every interaction the membrane handled was
grounded, audited, and reversible — because the membrane is part of the factory, not a
separate tool you'd have to integrate.
Where It Fits
GRAMS in the graph
GRAMS sits on the perimeter, at the top of the stack. Internally, it would consume from
every layer beneath it: the ledger from GIMS, routing and escalation from GONS-Core,
relevance and anomaly signals from GEDS, intent decomposition from GOMS, and — from the
other later layers — allocation from GAMS and portfolio evidence from GEMS. Externally, it
is the only organ the market would ever see.
GRAMS — The Membrane
Outbound and inbound channels to the market. Selective, tiered, audited. The only organ with an external surface, and the last rung of the ladder.
← GAMS
A later layer itself: allocation over a living topology. External commitments would be grounded in its viability check, so promises that aren't credible never make it out the door.
← GEMS
Portfolio evidence for proposals and applications. GEMS is a read-view over GIMS — artifacts are already ledger sentences, so every claim has a citable history.
↔ GIMS
The institutional ledger, already in the field. Every membrane crossing becomes an append-only sentence: subject, verb, object, at, by, under. Audit is structural, not optional.
↔ GONS
Inbound counterparty events would route through GONS-Core like any other question — nobody talks directly, not even the market — and the review queue would surface in GONS-Console.
↔ GEDS
The relevance engine would read counterparty patterns as tensions and anomalies: bid-success drivers, recurring objections, channels worth investing in — with handoff packets for the hard calls.
The market is asynchronous, noisy, and unreasonable. GRAMS won't fix that; it will give
the factory a stable, policed, audited interface to deal with it. The factory
handles volume; the human handles consequence.