SureThink/Product/Pillar 02
Pillar 02 · Customise the process
The person who knows the process wires the process — and can change it on a Tuesday afternoon without waiting on a release.
Triggers
email_watch
api_intake
folder_watch
schedule
Intelligence
doc_classifier
wip_extractor
ratio_analyst
Gates
rule_pack_gate
confidence_gate
Human
approval_tray
four_eyes
Output
letter_writer
record_write
Gate
i4 · rule_pack_gate
Rules in force
Hand to a person when a field is
Route to
Controls
The gate that decides whether a file goes on or stops is one panel with four settings. That is deliberate — the moment a control needs a consultant, it stops being a control your team actually owns.
The catalog
Here is the whole thing, because the interesting question is never what a demo shows — it is whether the module you need is in there.
■ triggers■ intelligence & gates■ human stations■ output
Before it touches a live file
01 · Test Bench
Point the pipeline at pinned fixtures — real files you have already decided — and watch what it would have done. No AI spend, no live data, no consequences.
run 41 fixtures
38 match prior decision
3 differ — review
cost $0.00
02 · Observe mode
A rule in observe mode reports what it would have done, on live files, without being allowed to act. Turn it loose only once its record convinces you.
CS-091 · observing 14d
would refer 6 of 212
0 false stops
ready to enforce
03 · Frozen versions
A published version cannot be edited, only superseded. A case decided in March replays under March’s logic — which is the only way a replay means anything.
v14 live since 8 Sep
v13 · 212 cases pinned
replay exact
Every version keeps the cases decided under it. Nothing is migrated onto new logic, because a decision made under old rules was made under old rules.
What it costs to run
Every model call a pipeline makes is timed, counted and priced at the moment it happens, against a rate card with effective dates. Spend is a number you watch, not a bill you receive.
| Control | What it does | Where |
|---|---|---|
| daily_budget | A ceiling per pipeline per day. When it is reached the pipeline holds rather than overspending, and the queue says why. | per pipeline |
| per_call_meter | Tokens in, tokens out, duration and outcome recorded on success and on failure — a call that errored still cost money and still shows up. | every call |
| priced_at_write | Cost is frozen when the call happens. A later change to the rate card never rewrites history. | every call |
| unpriced_model | A model with no rate card entry reports as unknown cost, never as zero. You find out by seeing a gap, not by being quietly wrong. | every call |
Test Bench runs are excluded from all of this, because they never call a model.
Start with one workflow
Thirty minutes. Bring one real file, redacted however you need it, and we'll build the pipeline that would have handled it — then run it in the Test Bench for nothing.