# Phase1 Output Verify

> After phase1_one_shot_runner emits 13 L*.json, AI spot-checks completeness and source meaning. Triggers on phrases like "verify L docs", "check phase 1 output", "confirm extraction is correct", or automatically when /vibe-ic-phase1 / /vibe-ic-phase2 returns PASS.

- Skill: `vibeic/phase1-output-verify` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx get-skill add vibeic/phase1-output-verify`
- Raw SKILL.md: https://api.skillmd.com/api/skills/vibeic/phase1-output-verify/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: vibeic (https://skillmd.com/u/vibeic)
- Updated: 2026-11-15
- Page: https://skillmd.com/skills/vibeic/phase1-output-verify

---


# Review Phase 1 output

Read the selected rows in `<project>/reports/flow_engine.json` and open their
native `reports`. `flow/steps.json` names each current tool and
`flow/native_limits.json` retains its original limits. Review `tool_ran`,
`native_steps`, `native_metrics`, and each metric's value, operator and limit.
A missing native measurement is `NOT_MEASURED`. AI reviews the engineering
meaning of every result and fixes problems in the design, tool or plugin.

```bash
python3 programs/flow_engine.py report --project <project>
```

Compare the admitted source documents with the generated fields where their
actual consumers read them. Review the following independently.

1. Required facts are present in their owning layers. A copied token in an
   unrelated field does not close a missing requirement.
2. Port names, widths, directions, clocks and resets agree across the source,
   integration documents and verification plan.
3. Register addresses, OTP data, opcode sets and error/default behavior retain
   the source meaning, units and bounds.
4. Analog blocks, physical constraints and unavailable implementations are
   explicitly represented when the input requires them.
5. Placeholders, unsupported assumptions and open questions are disclosed.

Read representative source passages and their exact generated counterparts.
When a field is absent or wrong, repair the owning extractor or document and
rerun that generation step. Keep a concise report of the discrepancy, fix and
remaining scope. Use `phase1-completeness-deep-review` for semantic ownership
and completeness. Do not qualify content through a receipt or schema score.

## Captured rule

When one document declares different IC and IP periods, preserve the Route
column through the declaration reader and L19. Verify the public Phase 1,
native configuration and consumed SDC for each delivery route and PDK.
Each IP run uses the period declared for that IP route.
An unrecognized Route cell for the selected PDK makes the declaration
ambiguous. Name the cell and source in the diagnostic before using any period.

## Captured rule — A transaction rate is not the primary clock

Use the whole Markdown table row as context: its cells form one statement,
including commas and semicolons inside a cell. For prose, use the clause
containing the frequency. A neighboring reset or sampling instruction can
mention `clk` without assigning its clock rate to an SPI transaction. Keep
decimal frequency literals intact when splitting clauses. Do not reject a
frequency because of its numeric value.

PWM4's verification note states 1 MHz SPI frames before the main document's
25 MHz system clock. Borrowing clock words across clauses assigned 1 MHz to
`clk` and made L19, native SDC and GLS consume 1000 ns instead of the stated
40 ns. Step 1 still reported PASS. Review the clock where the actual consumers
read it, including generated input delays; preserve the original clock,
stimulus, assertions and native timing limits.

The verbatim three-document regression is in
`programs/tests/fixtures/phase1_local/primary_clock_and_serial_rate/`. The
second fixture, `primary_clock_table_row/`, retains a real analog table row
whose declared maximum follows a semicolon and comma. Preserve the existing
choice between a table's typical and maximum values during this repair.
The corrected native IP reference consumes 40 ns and passes Step 75 Candidate
gate-level simulation: three SDF Icarus cases and all four compiled functional
cases, with zero annotation or functional errors. Its final SDF row also
passes all 27 directed cases across nine corners and all four compiled cases.
The complete IP reference is 44/44 PASS, including Step 105 Signoff aggregation
and Step 106 Compliance and final adoption. The IC candidate GLS suite also
passes all four Icarus and four compiled cases; its later rows are still open.
Reference results do not qualify a completed PPA search.

## Captured rule

A hex-valued command-code column is not a register-address column. Preserve
that distinction in the documentary L4 identity parser, just as for value,
encoding, and reset columns; do not add command names to the register map to
satisfy a mistaken population. On PWM4 main v1.40.62 / image 0.4.76,
`pwm4-phase1-diagnostics-v14062-i0476-oai` measured 9/18 bindings and Phase 1
FAIL because nine SPI opcodes were misclassified. With explicit opcode-header
classification, `pwm4-phase1-table-fix-i0476-oai` measured the actual 9/9
register bindings and Phase 1 PASS (4.566 s). Limits and waivers are unchanged.
`pwm4-table-parser-outcomes-unit-i0476-oai` passed 35 tests, including command
header variants, address/offset/non-English register headers, and a genuinely
missing register that remains ABSENT_FROM_L4.

## Captured rule — An unresolved symbolic width is not a scalar declaration

When the input explicitly gives a symbolic width, a failed numeric parse does
not establish width 1. Retain the symbolic or unresolved cardinality until the
current input or a source-bound native measurement of the complete declared port
resolves it.
Check width, msb and lsb in L1 and every L9 consumer projection against that same
source. Inspect the owning table even when it has no package-position column;
a correction for package tables does not establish that this other path is sound.
Report an unsupported scalar fallback as a source-meaning finding rather than
inventing numeric geometry or treating an unknown width as a measured scalar.

